2697_Prelims.

indd 1 09/07/2015 09:35

2697_Prelims.indd 2 09/07/2015 09:35
2697_Prelims.indd 3 09/07/2015 09:35
 Vice President: Denise Schanck
Senior Editor: Elizabeth Owen
Assitant Editor: David Borrowdale
Production Assistant: Deepa Divakaran
Illustrator: Oxford Designers & Publishers
Layout: Techset Composition Ltd
Cover Designer: Andrew McGee
Copyeditor: Ray Loughlin
Proofreader: Susan Wood

© 2016 by Garland Science, Taylor & Francis Group, LLC

This book contains information obtained from authentic and highly regarded sources. Every effort has
been made to trace copyright holders and to obtain their permission for the use of copyright material.
Reprinted material is quoted with permission, and sources are indicated. A wide variety of references
are listed. Reasonable efforts have been made to publish reliable data and information, but the author
and the publisher cannot assume responsibility for the validity of all materials or for the consequences
of their use. All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system or transmitted in any form or by any means—graphic, electronic, or mechanical, including
photocopying, recording, taping, or information storage and retrieval systems—without permission
of the copyright holder.

ISBN 9780815344919

Library of Congress Cataloging-in-Publication Data

Donaldson, Peter, 1959-, author.

  Genetics of complex disease/Peter Donaldson, Ann Daly, Luca Ermini, Debra Bevitt.
   p. ; cm.
  Includes bibliographical references.
  ISBN 978-0-8153-4491-9 (pbk.)
  I.  Daly, Ann K., author.  II.  Ermini, Luca, 1978-, author.  III.  Bevitt, Debra, 1966-,
author.  IV. Title.
  [DNLM:  1.  Disease–genetics.  2.  Genetic Diseases, Inborn–genetics.  3.  Genetic Predisposition to
Disease. QZ 50]
 RB155
 616'.042–dc23
2015022195

Published by Garland Science, Taylor & Francis Group, LLC, an informa business,
711 Third Avenue, New York, NY, 10017, USA, and 3 Park Square, Milton Park, Abingdon,
OX14 4RN, UK.

Printed in the United States of America

15 14 13 12 11 10 9 8 7 6 5 4 3 2 1

Visit our web site at http://www.garlandscience.com

2697_Prelims.indd 4 09/07/2015 09:35

 Preface

There is a scientific revolution happening in biomedical genetics. The new genetics does
not just apply to the well-known and well-described Mendelian diseases with clear pat-
terns of inheritance, nor is it limited to major chromosomal abnormalities. What makes
the revolution so exciting is that it includes all human diseases and all aspects of human
disease. Diseases that have been largely, but not entirely, ignored in the past are the main
focus of this revolution. The potential arising from this work is astounding. It is already
having an impact and the impact will only grow over time. There are many books on
genetics, but few concentrate on complex diseases—those that do not fit the simple pat-
terns of Mendelian disease and cannot be described as chromosomal abnormalities.
Over the past 15–20 years interest in these genetically complex diseases has taken full
flight. Though earlier studies had identified some important genetic links and associations,
many of the early studies had failed to be replicated and studies in this area of genetics had
developed a poor reputation. There were some good studies and many bad studies. The
difference between good and bad studies is quite well known. However, developments in
the last 20 years have restored interest and confidence in studies of complex disease.
A number of important developments were the keys to opening up this area for high-
quality research. The two most important developments have been the Human Genome
Mapping Project and the development of supercomputers along with the necessary sys-
tems capable of handling the data that very-large-scale studies produce. These two devel-
opments go cap-in-hand, one is not possible without the other. In 2015, we have the
human genome sequence, the SNP Map and the HapMap. Of course array platforms
for genotyping and application of this knowledge as well as more sophisticated statistical
analysis have also filled an essential gap. Indeed, the genetics of today is as much about sta-
tistics as it is about biology and there are Professors of Statistical Genetics in our academic
institutions who dedicate their research to extracting important facts from the mountains
of data that current studies can generate.
This book addresses the subject of genetics of complex disease and is designed in two parts.
The first part (Chapters 1–5) provides a basic background to genetically complex diseases,
and why and how we study them. The second part (Chapters 6–12) focuses on specific
sub-branches of genetics of complex disease and specific examples to highlight the applica-
tion of genetic data in complex disease and the extent to which this data is fulfilling the
promises of the Human Genome Project.
Chapter 1 covers the necessary background to genetic variation in the human population,
i.e. our evolutionary past and how genetic variation arises. Chapter 2 goes on to define
complex diseases and compare them with Mendelian and chromosomal diseases. Chapter
3 looks at how we investigate complex diseases, including the different plans and strate-
gies available to us. Do we chose a single gene or region to study, or do we throw the net
wider and investigate the whole genome in a genome-wide association or linkage study?
Chapter 4 considers why we are interested in complex diseases, focusing on the major

2697_Prelims.indd 5 09/07/2015 09:35

 vi Preface

promises of the Human Genome Project in relation to complex disease. These suggested
that genetic testing will be used in disease diagnosis, patient treatment and management,
and in understanding disease pathology. Chapter 5 looks at how data from the studies
described below is handled in a range of different statistical tests.
Sufficient information is given in each of Chapters 1–5 to enable students to understand
the major points and, where appropriate, examples are used to illustrate the key con-
cepts (e.g. in Chapter 2, where Crohn’s disease and Hirschsprung’s disease are discussed as
two different models of genetically complex diseases). Chapter 4 uses quite a few disease
examples to illustrate how the genetic information may be used to meet the promises of
the Human Genome Project.
After Chapter 5, the book goes on to look at three specific areas: immunogenetics (Chapter
6), infectious disease (Chapter 7), and pharmacogenetics (Chapter 8).
Chapter 6 on immunogenetics deals with how common variation in genes that regulate
the immune response can increase or reduce susceptibility to common diseases. The chap-
ter concentrates on the major histocompatibility complex on chromosome 6p21.3. The
chapter includes a considerable number of recently studied examples and discusses the
different interpretations that can be applied to the data. In each case, the extent to which
these examples do or do not fulfill the promises of the genome project is considered. There
are positive examples of how genetics can be used as an aid to diagnosis (e.g. in ankylosing
spondylitis), and also how associations and linkage with certain risk alleles may be helping
us to understand disease pathogenesis (e.g. in autoimmune liver disease).
Chapter 7 on infectious disease looks at the past and the present considering how genetic
variations may influence the likelihood of infection per se and the outcome following
exposure to infectious agents. The discussion provides interesting links with mankind’s
early history. The chapter concentrates on a few selected examples to illustrate the con-
cept and demonstrate how the studies discussed are helping to fulfill the promises of the
Human Genome Project. Once again, there are clear examples of genetic investigations
impacting on our understanding of disease pathology.
Chapter 8 on pharmacogenetics discusses past and present developments in a fast-expand-
ing field that is at present providing some of the most promising results in complex disease
genetics. Studies have shown that responses to commonly used pharmacological agents
can be determined by common genetic variation. The impact of this variation ranges from
failure to respond to a drug to life-threatening toxic reactions. The potential to use genet-
ics to tailor therapy and also to develop new therapeutic agents is a real possibility in this
sub-branch of complex disease genetics, and one that major pharmaceutical companies
and academic institutions are aware of.
Chapters 9 and 10 focus on specific disease groups: cancer (Chapter 9) and diabetes
(Chapter 10). These two chapters stand alone because one group of diseases (cancer) has
a very significant impact in terms of morbidity and mortality in the developed and devel-
oping world and the other group (diabetes) is for the most part a perfect example of a
complex disease. The potential medical impact of genetic studies in these diseases is vast.
More rapid diagnosis, better patient care, personal life planning, and personal treatment
planning are all possible. As we gain a greater knowledge of the genetics of these diseases
we will start to have a better grasp on the underlying pathology of each disease, which will
open up doors for diagnosis, treatment, and management. In some cases, this will mean
simple things like changes to a person’s diet; in others, selecting the appropriate chemo-
therapeutic agent to use for a patient. To some extent some of these aims have already been
achieved, but as this book indicates, there is still much to be done.

2697_Prelims.indd 6 09/07/2015 09:35

 Preface vii

In Chapter 9 (cancer), a selected number of examples are discussed. These include breast
cancer, prostate cancer, and lung cancer. The selection is based on the most common
cancers, which are also, to some extent, those about which we know the most. Links to
useful websites are given for further information and updates. Diabetes (Chapter 10) is
discussed in its various forms, especially type 1 and type 2 diabetes and is specifically used
to illustrate the difference in the genetic portfolios in type 1 and type 2 diabetes. Here,
the question is why are two diseases that have so much in common so different in terms
of their genetic profiles?
The last two chapters deal with societal and ethical issues in the new genetic era and the
future of genetics in complex disease. This is a fast-moving area of science. The facts being
produced today will be marketed as diagnostic or prognostic indicators almost as quickly
as they are identified. Genetic testing for risk alleles will soon be normal practice, but this
will have ethical and social consequences. The potential for misuse of genetics is discussed
in Chapter 11, highlighting the importance of understanding what a genetic test in com-
plex disease is really telling you. You will need to know what a genetic profile is telling
you before getting tested. There is considerable commercial interest in genetic profiling
and this has ethical and societal impact. Other points discussed include who owns your
genome and who can access your genetic data?
Chapter 12 closes the book by looking at the techniques and technologies that have been
used and those that will be used in the future. The chapter reminds us that technologies
used in the past will also be used in the future, but it also highlights some fascinating new
possibilities. Most important will be direct sequencing either at the level of the exome (i.e.
protein-coding genes only) or the whole genome.
The structure of the book is designed to provide a basic platform on which students can
build their knowledge base. Each of the chapters (including the basic chapters) uses exam-
ples of disease to illustrate key specific points and provides a reasonable level of basic
current data on each example used. In particular, the book focuses on the promises of the
Human Genome Project that suggested genetics will be used to improve disease diagnosis,
to develop individual treatment and management plans for patients, and to inform the
debate on disease pathogenesis. At each stage and after each example, the text reflects on
the extent to which these promises have been or will be met, looking at both the present
and, if possible, the future. Links to the web are also provided for access to updates and
further information throughout the book. There is an extensive Glossary at the end of the
book.
These are very exciting times for genetics, especially in complex disease. They are also fast-
moving times. The book is written as a starting point (a first block) and for the most part it
is written in an historical style to ensure it remains in date whatever develops in the future.
This book provides a good starting point for anyone studying the genetics of so-called
complex diseases. It is written for the undergraduate student and early postgraduate stu-
dent alike. It is written for the medical and non-medically minded individual. This era is
one of the most exciting eras in modern genetics, perhaps as exciting as when the structure
of DNA was first revealed to the scientific community.
We would like to thank the staff at Garland Science, Liz Owen, David Borrowdale and
Deepa Divakaran, for their support and encouragement in producing this book.

Peter Donaldson, Ann Daly, Luca Ermini, and Debra Bevitt

2697_Prelims.indd 7 09/07/2015 09:35

 Acknowledgments

As senior author I would like to give specific thanks to: Robert Taylor (Newcastle
University) who provided advice on the mitochondrial genome, John Mansfield (Newcastle
University) who provided necessary background on inflammatory bowel disease, Roger
Williams (King’s College, London) and Oliver James (Newcastle University) both of
whom provided a supporting environment within which to learn and develop a back-
ground in liver disease and genetics as well as the necessary skills to produce this book,
Derek Doherty (Trinity College, Dublin) who worked with me on the molecular genetics
of the MHC in liver disease at King’s College Hospital, London and the many members
of different research teams who have contributed to my research between 1982 and 2015.
In addition I would like to give special thanks to the hundreds of students who, through
their positive interaction and feedback, have encouraged the writing of this book. Finally,
I would like to give very special thanks to Carolyn Donaldson who encouraged and sup-
ported production of this book from start to finish, especially during difficult times.

Peter Donaldson

The authors and publisher would like to thank external advisers and reviewers for their
suggestions and advice in preparing the text and figures.
Geoffrey Bosson (Newcastle University, UK); Margit Burmeister (University of
Michigan, USA); Angela Cox (University of Sheffield, UK); Rachelle Donn (University
of Manchester, UK); Yalda Jamshidi (St George's, University of London, UK); Martin
Kennedy (University of Otago, New Zealand); Andrew Knight (Newcastle University,
UK); Hao Mei (Tulane University, USA); John Pearson (University of Otago, New
Zealand); Logan Walker (University of Otago, New Zealand); Kai Wang (University of
Iowa, USA); Yun Zhang (Oxford Brookes University, UK).

2697_Prelims.indd 8 09/07/2015 09:35

 Contents

Preface v
Acknowledgments viii

1 Genetic Diversity 1
1.1 Genetic Terminology 2
The use of the terms genes and alleles varies, though they do have precise definitions 2
1.2 Genetic Variation 5
Genetic variation can be measured by several methods 6
Alleles on the same chromosome are physically linked and inherited as haplotypes 7
Linkage disequilibrium promotes conservation of haplotypes in populations 8
1.3 Genetics and Evolution 9
Mutation is the major cause of genetic variation 10
Genetic variation caused by mutation alters allele frequencies in populations 11
Migration and dispersal cause gene flow 12
Allele frequencies can change randomly via genetic drift 13
The thrifty gene hypothesis 16
Natural selection acting on different levels of fitness affects the gene pool 16
1.4 Calculating Genetic Diversity: Determining Population Variability 19
Genotype and allele frequencies illustrate genetic diversity 19
Allele frequency refers to the numbers of alleles present in a population 20
Heterozygosity provides a quantitative estimation of genetic variation 21
The HWP is a complex but essential concept in population genetics 21
Calculating expected genotype frequencies using the HWP 22
Different populations may have different allele frequencies 22
1.5 Population Size and Structure 26
Breeding population size is important in evolution 26
Genetic variation is not always uniform in a population 26
Wahlund’s principle 27
1.6 The Mitochondrial Genome 27
1.7 Gene Expression and Phenotype 29
Genetic variation is manifested in the phenotype 29
Phenotypes are influenced by the environment 29
1.8 Epigenetics 30

2697_Prelims.indd 9 09/07/2015 09:35

 x Contents

1.9 Genomic Imprinting 31

Conclusions 31
Further Reading 33

2 Defining Complex Disease 35

2.1 Definition of a Genetically Complex Disease 36
To fully understand complex disease it is important to deconstruct this definition 36
2.2 Chromosomal Diseases 40
Changes in chromosome number cause serious genetic diseases 40
Changes in chromosome structure can cause serious illness 41
2.3 Mendelian Diseases 43
Mendelian diseases involve a single gene and show simple patterns of inheritance 43
Penetrance is an important difference between Mendelian and complex diseases 45
Some diseases have both Mendelian and complex characteristics 47
Modifier genes may also confuse the picture 47
Mendelian traits can be studied in families 47
There are complications to Mendelian diseases 48
2.4 Variation in The Mitochondrial Genome is Associated with Disease 50
Variation in the mtDNA has been widely associated and linked with many different diseases 52
2.5 De Novo Mutations and Human Disease 53
2.6 Three Different Types of Complex Disease 54
Studying complex disease is different from studying Mendelian disease 54
Monogenic complex diseases involve a single risk allele 55
Oligogenic complex diseases involve several alleles 56
Polygenic complex diseases involve many risk alleles 56
2.7 Alzheimer’s Disease May be a Monogenic Complex Disease 57
2.8 HSCR – An Oligogenic Complex Disease 58
Sporadic HSCR illustrates the oligogenic model for complex disease 58
2.9 Crohn’s Disease is Mostly a Polygenic Complex Disease 61
Early studies of Crohn’s disease suggested a number of locations for risk alleles 61
Genetic variations in the human equivalent of the plant nod2 gene (CARD15) were the first
identified and confirmed Crohn’s disease risk alleles 62
Genetic variations in other immune regulatory genes are important risk factors in
Crohn’s disease 64
The Wellcome Trust Case Control Consortium (WTCCC1): Crohn’s disease 64
The current number of risk alleles for Crohn’s disease may be as high as 163 66
2.10 Applying Disease Models to Populations 67
Conclusions 67
Further Reading 69

3 How to Investigate Complex Disease Genetics 73

3.1 Planning Stage 1: Gathering the Basic Knowledge 73
Incidence and prevalence are measures of how common a disease is 74

2697_Prelims.indd 10 09/07/2015 09:35

 Contents xi

Incidence and prevalence can be very different or very similar depending on the prognosis
for the disease 75
Incidence and prevalence of disease may vary in different populations 76
What is the evidence for a genetic component to the disease? 76
What is known about the disease pathology? 80
Before we get down to the hard business of study planning there are one or two other
questions that it is important to ask 82
3.2 Planning Stage 2: Choosing a Strategy 84
Two basic strategies for identifying risk alleles in complex disease 84
In terms of the history of genetic studies in complex disease there are two main periods:
pre- and post-genome 84
Each of these two strategies has a substrategy 88
3.3 Good and Bad Practice 93
Accurately identifying true disease susceptibility alleles in GWAS (and other association
studies) is dependent on sample size 93
Case selection can introduce bias into a study 94
It is important to consider whether we are studying a disease, a syndrome, or a trait within
a disease subgroup 94
Selection of appropriate controls is equally important in any study 94
Errors in the laboratory and in sample handling can also introduce bias into a study 96
Statistical analysis is the key in any study of complex disease 96
SNP chip selection is an important factor to consider in study design 96
Unfortunately publication bias does occur 97
Replication in an independent sample is crucial for all association studies, especially GWAS 98
3.4 New Technologies and the Future 100
The technological advances of the past decade have had a major impact on research into
the genetics of complex disease and the rate of change is going to increase 100
New developments will come from the ENCODE project, and will also involve more
epigenetics and imputation analysis 100
The real debate about the future of complex disease research lies not in the genetics itself,
but downstream from the genetics 101
Conclusions 101
Further Reading 103

4 Why Investigate Complex Disease Genetics? 105

4.1 Why Do We Investigate Complex Disease? 106
Complex diseases do not conform to simple patterns of inheritance 106
The HGP in research into genetically complex disease 107
4.2 Disease Diagnosis 108
Early studies on the genetics of ankylosing spondylitis indicated what could be achieved
in terms of differential diagnosis in the post‑genome era 108
Genetic associations in complex disease confer small risks 110
4.3 Patient Treatment/Management and Care 110
Identifying risk alleles that predict onset of complex diseases may enable patients to
make beneficial lifestyle changes 111

2697_Prelims.indd 11 09/07/2015 09:35

 xii Contents

Predicting disease severity through genetic analysis may have clinical significance in
terms of patient management 111
Common genetic variation may predict response to treatment and be critical in patient care 113
Onset, severity, and response to treatment are all part of patient management 113
4.4 Disease Pathogenesis 114
Early studies offered potential insight into the biology of ankylosing spondylitis 115
Later GWAS have offered even further insight into the biology of ankylosing spondylitis 116
Rheumatoid arthritis has many strong genetic associations, some of which can be used to
help us unravel the pathogenesis of this disease 116
Bipolar disease is a disease for which there are many weak genetic associations, but few
strong consistent associations 122
Coronary artery disease is the most common cause of death in the developed world 127
4.5 What about the Other Diseases? 136
Conclusions 137
Further Reading 138

5 Statistical Analysis in Complex Disease: Study Planning and

Data Handling 141
5.1 Linkage Analysis 142
The LOD score is a measure of significance of linkage between a trait and a marker allele 144
5.2 The Basic Statistical Concepts of Association Analysis and their Application
in Study Design 145
In statistical terms, there are two different hypotheses to consider in the analysis of genetic
association studies: a null hypothesis and an alternative hypothesis 146
5.3 Statistical Error, Power, and P Values 146
Making the right decision and avoiding errors in the hypothesis testing 146
The likelihood of detecting a significant difference in an association study is directly related to
sample size 147
Probability (P) values are simply statements of the probability that the observed differences
between two groups could have arisen by chance 148
5.4 The Basic Statistical Considerations for Analysis of Case Control Association Studies
and their Application to Data Collection and Analysis 151
Departures from HWE can have different causes 151
Pearson’s χ2 and Fisher’s exact test are used to assess the departure from the null hypothesis 152
Fisher’s exact test calculates the exact probability (P) of observing the distribution seen in the
contingency table 154
The Cochran–Armitage test looks for a trend for a difference between cases and controls
across the ordered genotypes in the table 155
There is no simple answer to the question of which test to choose 157
Data may also be analyzed assuming a predefined genetic model 157
Logistic regression is frequently used in association studies 160
The pitfalls and problems of GWAS 162
5.5 How to Interpret a GWAS 165
There are several ways to interpret statistically significant genetic associations 165
There are several diagnostic plots that can be used for the visualization of genome-wide
association results 165

2697_Prelims.indd 12 09/07/2015 09:35

 Contents xiii

Linkage disequilibrium is a useful tool in association studies provided you know how to
handle it 167
The ability to detect a significant association through linkage disequilibrium can increase
the power of an association study 167
Most association analyses identify multiple SNPs, other genetic variants, and haplotypes 170
Conclusions 171
Further Reading 172

6 The Major Histocompatibility Complex 175

6.1 Histocompatibility 176
The idea of histocompatibility first started with blood groups 176
The MHC-encoded HLA antigens are the second major histocompatibility group 176
Naming the HLA antigens and alleles up to and including the early molecular genotyping era 177
The current naming system for HLA alleles and genes allows for a greater level of resolution
to be reported 183
The MHC encodes a cornucopia of genetic diversity within the HLA genes 184
Comparing the levels of genetic diversity at DR with those at DQ can make DQ look like
a poor relation 185
HLA class II molecules can be expressed in trans or in cis 187
The final groups of genes that need to be considered are those called pseudogenes,
gene fragments, and null alleles 188
6.2 The Extended Human MHC MAP 189
6.3 Molecular Structure of HLA Class I and Class II 191
X-ray crystallography of HLA-A2 revealed the full structure and much about the function
of HLA class I 191
The X-ray crystallography structure of HLA class II structure revealed the critical difference
between class I and class II 192
6.4 Immune Function of HLA Class I and Class II 193
Class I molecules have distinct features 193
HLA class II is different to class I 193
HLA class I and class II have important similarities 194
HLA class I and antigen engagement in the cell is different from HLA class II 194
HLA class II and antigen engagement in the cell is different 195
6.5 HLA Class I and Disease 196
Hemochromatosis is an example of a Mendelian disease which maps within the xMHC 196
Psoriasis proves the point that HLA-C is an important locus to consider in genetic studies
of the MHC 196
Type I versus type II psoriasis 197
Before we leave HLA class I we need to consider Bw4 and Bw6 197
6.6 HLA Class II and Disease 197
Severe or cataplectic narcolepsy has one of strongest HLA associations ever reported 197
There are different functional interpretations of the HLA association with narcolepsy 198
Multiple sclerosis is a disease with a strong genetic association with HLA class II 199
HLA class II and autoimmune liver disease 201

2697_Prelims.indd 13 09/07/2015 09:35

 xiv Contents

AIH is a relatively rare classical autoimmune disease of the liver 202

PSC is not a classical autoimmune disease 207
PBC is an autoimmune liver disease with a genetic component 210
6.7 Comparing the Hla Associations of the Three Liver Diseases 213
6.8 Non-HLA MHC Genes and Disease 213
The MHC class III region complement, MICA, and TNFA genes in complex disease 214
6.9 A Single Gene or a Risk Portfolio 216
A single gene may explain MHC- encoded genetic susceptibility to disease 216
Alternatively there is always room for a second bite of the cherry: a multihit hypothesis 217
6.10 How to Compare and Critically Evaluate Contrasting Studies 218
Knowing history is important when we critically review and design studies 218
Conclusions 219
Further Reading 221

7 Genetics of Infectious Disease 223

7.1 The Infection Process and Disease 223
Mechanisms of infection vary widely but common steps in the process can be identified 224
The immune response combats infectious disease 224
Individuals infected by the same pathogen may experience different outcomes 225
7.2 Heritability of Resistance and Susceptibility to Infectious Disease 225
Different populations infected by the same pathogen may experience different outcomes 225
Leprosy and tuberculosis were once believed to be inherited diseases 226
Adoption studies indicate that susceptibility to infectious disease has a heritable component 227
Rare monogenic defects in immunity can cause primary immune deficiencies 227
7.3 Identifying Alleles that Affect Risk of Susceptibility and Resistance to Infectious
Disease 228
Risk alleles can be identified using a hypothesis-driven or genome- wide approach 228
The outcome of infectious disease being tested must be clearly defined 229
7.4 Malaria 229
The life cycle of the Plasmodium protozoa is complex 229
Hemoglobinopathies confer resistance to malaria 230
Haldane’s malaria hypothesis proposed that thalassemia confers protection against malaria 232
Allison demonstrated that sickle cell trait confers resistance to P. falciparum 232
Studies on Pacific Island populations provided experimental evidence that thalassemia confers
protection from malaria 233
The mechanism of resistance to malaria conferred by hemoglobinopathies is still not fully
understood 233
Resistance to malaria conferred by HbS and thalassemia is a complex genetic trait 235
Other malaria resistance alleles have been identified via epidemiological or hypothesis-driven
studies 235
GWAS suggest that polymorphisms in immunity-related genes may affect outcome of
Plasmodium infection 235
GWAS searching for malaria resistance alleles highlight the challenges of GWAS in African
populations 235

2697_Prelims.indd 14 09/07/2015 09:35

 Contents xv

7.5 HIV-1 237

C-C chemokine receptor 5 (CCR5) acts as a co-receptor for HIV-1 in the early stages of
infection 237
Some individuals are naturally resistant to HIV infection 238
A 32-bp deletion in the CCR5 gene confers resistance to HIV-1 infection 239
Selection pressure by HIV-1 cannot account for the high frequency of CCR5-Δ32 in the
northern European population 240
CCR5-Δ32 affects the outcome of infection by West Nile virus 240
CCR5-Δ32 cannot account for all HIV-1 resistance 241
CCR5 promoter polymorphisms affect HIV-1 control 242
CCR5/CCR2 haplotypes have a complex effect on HIV-1 control 242
Polymorphisms in chemokine receptor ligand genes influence HIV-1 control 244
Polymorphisms in HLA genes affect outcome of HIV infection 245
HLA class I homozygosity is not always bad news 246
GWAS confirms the protective role of HLA-B in HIV-1 infection 247
Amino acids in the HLA-B binding groove are associated with HIV-1 control 248
GWAS revealed, for the first time, association of HLA-C with HIV-1 control 248
Some SNPs previously implicated in HIV-1 control have not yet been confirmed
by GWAS 249
Conclusions 249
Further Reading 251

8 Pharmacogenetics 253
8.1 Definition and a Brief History of Pharmacogenetics 254
8.2 Cytochrome P450 255
There is a clear relationship between genotype and phenotype for several forms of
cytochrome P450 255
The conversion of the analgesic drug codeine, which is administered as a pro-drug and
is activated to morphine by CYP2D6, is of clinical importance 258
The cytochrome P450 CYP2C9 metabolizes warfarin – a very widely used drug 259
CYP2C19 activates clopidogrel – a drug widely used to prevent strokes and heart attacks 259
8.3 Other Drug-Metabolizing Enzymes and Transporters 261
For phase II conjugation reactions, the UDP glucuronosyltransferase family makes the
largest contribution 261
Methyltransferases are also important in phase II drug metabolism 261
Polymorphisms in drug transporters also play a role in pharmacogenetics 263
8.4 Drug Targets 263
The relationship between VKOR and coumarin anticoagulants is one of the most consistently
reported genetic associations involving drug targets unrelated to cancer 263
The efficacy of β-adrenergic receptor agonists widely used in the treatment of allergies may
also be genetically determined 264
8.5 Adverse Drug Reactions 266
HLA genotype is a potent determinant of susceptibility to several different types of adverse
drug reactions 266

2697_Prelims.indd 15 09/07/2015 09:35

 xvi Contents

The anti-human immunodeficiency virus (HIV-1) drug Abacavir gives rise to hypersensitivity

in some patients 267
Drug-induced liver injury is a rare, but clinically important problem 267
There are many other susceptibility factors for serious adverse drug reactions 269
Adverse reactions to commonly used statins provide a key example of non-HLA-related
adverse drug reactions 270
Cardiotoxicity reactions to drugs do not appear to involve an immune or inflammatory
response 270
Conclusions 271
Further Reading 273

9 Cancer as a Complex Disease: Genetic Factors Affecting Cancer

Susceptibility and Cancer Treatment 275
9.1 Defining Cancer 278
9.2 Cancer as a Complex Disease 280
Early studies of cancer found evidence of genetic associations with risk 280
GWAS has revolutionized the search for cancer-promoting alleles in non-familial cancers 281
9.3 Genetic Risk Factors for Particular Cancers Detected by Gwas 281
GWAS has identified a number of biologically plausible genetic risk factors for breast cancer 281
Novel insights into lung cancer involving the target for nicotine were detected by GWAS 283
A large number of genetic risk factors for prostate cancer have been revealed by GWAS 283
9.4 General Cancer Risk Loci Detected by Gwas 285
9.5 Previously Established Cancer Risk Factors Confirmed by Gwas 286
Alcohol, smoking, and chemical exposure increase the risk of cancer 286
9.6 Individualizing Drug Treatment Based on Tumor Genotype 288
Newly developed drugs inhibit the function of mutated proteins in cancer cells 288
Specific antibodies can target tumor- specific proteins and inhibit tumor growth 289
Epigenetic changes in the tumor involving methylation may affect response to conventional
drug treatments 290
Gene expression profiling may enable personalized cancer treatment 290
Before we close the chapter on cancer it is important to recognize that there are many forms
of this disease 291
Conclusions 292
Further Reading 294

10 Genetic Studies on Susceptibility to Diabetes 295

10.1 Diabetes Mellitus 295
10.2 Genetics of T1D 297
10.3 Early Genetic Studies in T1D 297
HLA class II genotype is the strongest genetic risk factor for T1D 298
Not all of the risk for T1D above may be associated with the DQB allele or HLA class II 299
Other genetic risk factors for T1D include the genotype for the insulin gene 300
Candidate gene studies have identified a number of other non-MHC associations with T1D 300

2697_Prelims.indd 16 09/07/2015 09:35

 Contents xvii

10.4 Gwas Studies in T1D 303

The 2007 WTCCC1 study was one of the first GWAS in T1D 303
Following the introduction of GWAS in 2007, research has resulted in the identification of
at least 40 further potential T1D alleles 305
10.5 Early Genetics of T2D 306
There have been different interpretations of the associations with PPARG, KCNJ11, and
TCF7L2 307
10.6 GWAS Studies in Type T2D 307
Examples from the WTCCC1 study 308
Other risk alleles for T2D from other studies 309
10.7 The Future of Genetics in T2D 310
Future prospects in T2D research involve genome sequencing 310
Epigenetics may be important in diabetes 310
10.8 Genetics of Monogenic Diabetes 311
Conclusions 312
Further Reading 314

11 Ethical, Social, and Personal Consequences 315

11.1 Defining Ethics 316
There are philosophical arguments for and against ethical constraint in biomedical research 316
What are the practical ethical implications in the study of genetics of complex disease? 317
11.2 Ethics in Genetics: What We Can Learn from the Past? 318
The consequence of the Eugenics Movement and the ideas it spread were extremely
bad news for the developing science of genetics 318
11.3 Looking into the Future Use of Genetic Data 321
Genetic studies of complex disease will have a major impact on clinical medicine 321
The potential personal impact of data from studies in complex disease is considerable 322
11.4 Who Does the Data Belong to? Interacting with Commerce 329
Do I own my genome? 330
11.5 Who Should be Able to Access the Data? 332
Conclusions 332
Further Reading 334

12 Sequencing Technology and the Future of Complex Disease Genetics 337

12.1 DNA Sequencing: The Past, Present, and Future 338
The development of DNA sequencing using the Sanger sequencing technique opened
the way to sequencing the genome 338
The new era: next-generation DNA sequencing 340
The upcoming era: third-generation sequencing 343
12.2 The Future of Ngs in Clinical Practice and Research 347
Using NGS will enable high- resolution genotyping for SNPs in complex disease 348
Using NGS will enable better identification of CNVs 350
Sequencing the RNA transcript and the whole transcriptome is an alternative way forward 350

2697_Prelims.indd 17 09/07/2015 09:35

 xviii Contents

12.3 Whole-Genome Versus Exome Sequencing 350

12.4 The Next Generations of Genome/Exome-Wide Association Studies 351
Missing and non-genotyped SNP data can be imputed using large databases and known
patterns of linkage disequilibrium 352
GWAS identifies both synthetic (false) associations and direct (real) associations 353
The importance of linking genotype to phenotype 353
Genotyping on new arrays provides a focus and higher level of resolution for GWAS 354
Different forms of NGS technology will impact on how GWAS is used 354
12.5 Epigenetics: A Complimentary Strategy in Complex Disease Studies 357
12.6 Metagenomics and the Bacterial Genome 357
To put metagenomics into context, we need to consider the impact it  may have 359
12.7 Major Ongoing International Genome Projects 360
HapMap is a project with major significance in current research, especially GWAS 360
The 1000 Genomes Project has major potential in studies of complex disease 362
ENCODE will help to link genotype to phenotype in complex disease 363
12.8 Systems Biology 364
Considering systems biology allows us to look into the future 364
Conclusions 366
Further Reading 368

Glossary 373
Index 397
Color Inserts

2697_Prelims.indd 18 09/07/2015 09:35

 CHAPTER
1
Genetic Diversity

Human evolution is driven by a number of different factors, including migration and

settlement in different environments, genetic mutation, natural selection, and genetic
drift. The product of these different forces is genetic diversity within a population, and
understanding this genetic diversity and the reasons for it are essential when considering
the genetic basis of common human diseases.
Though the origin of modern humans is relatively recent, humans have managed to colo-
nize almost all possible environments and in doing so have been exposed to considerable
selective pressure. Consequently, there is extensive variation in the human genome and
in the phenotypic traits (e.g. skin color) expressed. In this chapter, we will review the basic
background information on mutation, natural selection, and evolution, and the way this
helps us to understand the importance of genetic variation in the human genome. We will
pinpoint the reasons why genetic variation arises in a population and introduce phenom-
ena such as epigenetics. We will also consider the mitochondrial genome.
The genetic variation described here creates a basis for genetic risk in the majority of
human diseases. Understanding this genetic diversity and how it has arisen is a necessary
precursor to understanding the genetics of complex disease. Genetic variations between
individuals determine individual susceptibility or protection from a variety of common
diseases. This is the basic subject of this book, the idea that common genetic variation
gives rise to different levels of susceptibility to common disease. The evolutionary forces
that created this genetic variation have enabled populations to thrive, throughout human
history, because some population members are likely to be less susceptible to a given ill-
ness than others and are thereby more likely to survive even the most catastrophic event.

2967_Ch01.indd 1 07/07/2015 10:27

 2 CHAPTER 1 Genetic Diversity

1.1 Genetic Terminology

As with many scientific disciplines, genetics employs a large number of specific terms
and this terminology is given in the Glossary at the back of the book. The term genome
refers to the complete set of genetic information found in a cell and includes 22 pairs of
the autosomal chromosomes plus either XX (females) or XY (males) (Figure 1.1) and a
small amount of DNA found in the mitochondria (mtDNA). Human chromosomes
are the organized packages of DNA found in the nucleus of a cell. DNA is comprised of
linear double-stranded molecules that form a helix. The strands of the helix are made up
of alternative sugars (deoxyribose) and phosphate groups. Each sugar is attached to one
of four bases A, C, G, and T, and the whole molecule is stabilized by cross-linking of the
bases A with T and C with G. The DNA structure we are most familiar with looks like
a twisted ladder, though when packaged DNA is wound around histone proteins into
compressed units. The sequence A–T/C–G provides a code for the production of RNA
and RNA production may lead to protein production. The human genome is made up of
more than 20,000 genes; each gene being a single unit of inheritance that is transmitted
from parent to offspring. The location of a gene on a chromosome is referred to as a locus
(plural: loci) and genetic variation at a locus is referred to as allelic variation, where the
different forms are known as alleles. On average, human genes encode approximately 28
kilobases (kb) of DNA with a series of small exons (protein-coding sequences) separated
by long introns (non-coding sequences). Primary transcripts can be differentially spliced
into alternative proteins, adding yet another level of complexity to the story of genetics in
complex disease. In his book The Language of Life: DNA and the Revolution in Personalised
Medicine (2010), Francis Collins refers to a single gene in the brain that is capable of mak-
ing 38,000 different proteins. This is a remarkable and unusual figure. The total impact
of intronic genetic variation on common disease is only just beginning to be investigated,
but this figure is most likely to be an exception rather than the rule.

The use of the terms genes and alleles varies, though they do have
precise definitions
The terms gene and allele are often used as though they are the same, but it is important to
note that this is incorrect and that the correct term to use when considering genetic varia-
tion is allele. A gene is, as stated above, the basic unit of inheritance. The scientific litera-
ture is peppered with examples of incorrect use of this terminology. Writers often refer to
the “cystic fibrosis gene” and the “hemochromatosis gene” as though only patients with
these diseases possess the gene, when actually all members of the population possess these
genes. In these two examples, which are both Mendelian autosomal recessive disorders,
the difference between affected patients and healthy members of the population is that
patients possess two copies of the disease-causing alleles. Unaffected population members
may have a single copy of the disease-causing allele or may not carry this allele at all.
Instead, they will have one or two copies of the non-disease-causing allele. Thus, it is the
possession of the requisite alleles that causes the disease and not the possession of the gene.
Finally, the term allele is sometimes used to include any genetic variation within a region,

Figure 1.1: Karyotypes of human chromosomes. The figure illustrates the entire autosome showing banding
patterns for each chromosome in size order. Chromosomal banding was (and is) traditionally used to identify
chromosomes and chromosomal sites for clinical diagnosis. To obtain these patterns it is necessary to first
denature the DNA with enzymes, and then dye the sample to produce light and dark bands. Karyotypes are
assigned based on the chromosome length, banding pattern, and position of the centromere. Chromosome 1 is
the longest, and chromosome 22 is the shortest among the autosomal chromosomes. (From Strachan T & Read A
[2011] Human Molecular Genetics, 4th ed. Garland Science.)

2967_Ch01.indd 2 07/07/2015 10:27

 Genetic Terminology 3

36.3
36.2
Key:
25.3
36.1 25.2
25.1 Centromere
35 24
34.3
34.2 rDNA
34.1 23
33
32.3
Non-concentric
32.2
32.1
22
heterochromatin
31.3 21
31.2 26
16 25
31.1
15 24.3 16
24.2 15.3
14 25
22.3 24.1 15.3
22.2 23 15.2 24 22
13 22 15.2
22.1 15.1 23
15.1 22.3
21 12 21.3 14 22.3 21 22.2
14
11.2 22.2
21.2 13 22.1 15.3 22.1
11.1 13.3
13.3 11.1 21.1 12 13.2 15.2
13.2 11 13.1 21.3 15.1 21.3
13.1 11.2 14.3 21.2
14.2 11 12 21.2 14
12 12 14.1 12 11 21.1
11 21.1 13
11 13 13 13.1 11.1 11.4
13.2 11.2 12 12
14.1 11.3
12 14.2 12 13.3 12 11.2 11.2 11.23
14.3 21.1 11.1 11.1 11.22
11.2 13.1
21.1
21.1
11.1
21.2 11.2 11.1 11.1 11.21
11.1
21.2 21.3 13.2 11.21
21.2 21.3 11.1 13.3 12 11.1
13 11.22 11.2
21.3 11.2 22
22 12 12
22 23 14 14 11.23
13.1 13
23 23 24 15
13.2 15
24 24.1 25 16.1 21.1
24.2 13.3 21.1
21 16.2 21.2
25 24.3 26 16.3
21 21.3 21.2
27 22 21 21.3
31
31 22 23.1 22.1
23 23.2 22.1
32.1 28 22.1 22.2
23.3 31.1
32.2 24 22.2 22.3
31.2 23
32.3 31.1 31.1 22.3
32.1 25.1
33 25.2 31.2 31.2 23.1 31.3 24
32.2 25.3 31.3 23.2
32.3 26.1 31.3 23.3 25
34 32 32
41 26.2 33.1 24
35 32 33
26.3 33.2 26
42.1 33.3 25.1 34
42.2 36 33 25.2
27 34 25.3 35 27
42.3 37.1 34
43 28 35.1 26
37.2 35.2 36
44 37.3 29 35 27 28
35.3

1 2 3 4 5 6 7 X

23.3
23.2 24
23.1
23 15.1 15.5
22 15.4
22 14 15.3
21.3
21.2 13 15.2
21 15.1
21.1 12.3 13.3
12 12.2 14 13
13 12.1 13.2
13.1 12
11.2 13 12.3 13
11.1 12 11.2 12.2 11.2 12 13
11.1 11 11.1 12 12.1 11.1 11.2 12
11.22 11.21
11.23
11 11.1
11.2 11.2 11 11.1
11.2 11.2
12 12 11.12 11.1 12.1 11.1
11.1 1 11 12.2 11.1
11 12 12.3 11.2 11.1
13 13 21.1 11.2
12 13.1 13 12
21.2 14.1 12
21.1 13.1 13.2 13
21.1 13.2 14.2 13
21.2 21.3 13.3 13.3 14
21.2 14 14.3 15
21.3 22.1 13.4
22.2 13.5 15 21.1 21 21.1
21.3 22.1 14.1 21.2 21.2
22.2 22.3 14.2 21.1
14.3 21.3 21.3
22.1 23.1 21.2 22 22.1
22.3 23.2 21 21.3
22.2 22.1 23 22.2
22.3 23.3 22 22.3
31 24.1 22.2 22 24.1
24.2 22.3 23
23 24.3 23.1 31 24.2
32 23 24
25.1 23.2 24.3
33
24.1 25.2 23.3 24.1 32 31 25
34.1 25.3
24.2 32.1 26.1
24.2 26.1 24 24.31 33
34.2 26.2 24.32 32.2 26.2
24.3 34.3 26.3 25 34 32.3 26.3
24.33

8 9 10 11 12 13 14 15

13.3
13
13.2
13.1 12 11.32
11.31
12 11.2 13.3
11.2 13
11.1 13 11.3
11.2 11.1 13.2
11.1 11.2 11.1 12 13 12 11.2
11.1 13.1
11.1 12 12 11.1
21.1 11.2 12 11.2 11.2 11.1
11.2 12.1 11
12.1 21.2 11.2 11.1 11.21
12.2 11 11.1 11.1 11.1
12.2 21.3 12 11.1 11.1 11.22
13 12.3 11.2
21.1 11.2 11.2 11.23
21 22 13.1
21.2 21 12.1
21.3 12 12.2
23 13.2 12.3
22 13.1 13.1
24 22.1 12
23 22 13.3 13.2 13.2
22.2
24 25 23 13.4 13.3 22.3 13.3

16 17 18 19 20 21 22 Y

2967_Ch01.indd 3 07/07/2015 10:27

 4 CHAPTER 1 Genetic Diversity

whether or not it is part of the exome or intronic sequence. This would not be acceptable
to all readers of this book, but those with a focused interest in this area may consider this
correct. The use of terminology changes over time.
Most individuals have two copies of a given gene – one inherited from the mother and
one from the father. As a result, they are diploid. The genotype is the set of alleles that
an individual possesses. An individual may have two identical alleles, in which case they
are homozygous, or two different copies, in which case they are heterozygous (Figure 1.2).
When we consider the expression of a genetic variant we use the term phenotype. A phe-
notype can also be referred to as a trait or characteristic and may be either physical, physi-
ological, biochemical, or behavioral. Thus, the condition of having blue eyes or dark hair
is a phenotype, but so is having sickle cell anemia. Phenotypes are most often referred to
as traits or characteristics when they do not relate to an illness or disease.
In 2001, the first draft of the map of the human genome was published. Even though this
was not the complete sequence, it marked the beginning of a new era in genetics frequently
referred to as the post-genome era. The great advantage of working in the post-genome
era is that we have access to the genome map, and the majority of human genetic varia-
tion is known and available through websites such Human Genome Resources (http://
www.ncbi.nlm.nih.gov/projects/genome/guide/human/index.shtml) and the SNP Map
(http://www.ncbi.nlm.nih.gov/SNP/).

Mother Father Mother Father

Person A has the same allele at the Person B has a different allele at the
marked gene on both chromosomes marked gene on the two chromosomes
and is therefore homozygous and is therefore heterozygous

Figure 1.2: Homozygous versus heterozygous. The figure illustrates a single pair of chromosomes in two
individuals (A and B). In contrast to the picture in Figure 1.1, the band represents a single gene. Individual A
inherits the same allele from both parents (i.e. both black) and is therefore homozygous for this genotype, while
individual B inherits different alleles (gray and black) for this gene and is therefore heterozygous.

2967_Ch01.indd 4 07/07/2015 10:27

 Genetic Variation 5

1.2 Genetic Variation

Genetic variation is by convention discussed in terms of allele frequencies. A frequency is
simply a proportion or a percentage usually expressed as a decimal fraction. For example, if
20 out of 100 of the alleles at a particular locus in a population are of the A type, we would
say that the frequency of the A allele in the population is 20% or 0.2.
The term population in human genetic studies refers to the group of individuals occu-
pying a defined area such as a country, county, city, or town. Occasionally, a popu-
lation will be defined by other characteristics, including age, ethnicity, and even in
some cases by a particular disease. The complete set of genetic information contained
within a population is called the gene pool. The gene pool includes all alleles present
in the population. Some genes do not encode variation (i.e. they are monomorphic).
A monomorphic gene only exists in a single form and therefore has a single allele at a
frequency of 100% or 1. However, the majority of our genes are polymorphic, exist-
ing in two or more (poly) forms in a population. In a population a gene may encode a
limited or small number of different alleles or it may encode a much larger number of
alleles. One example of the latter is the HLA-B gene, which encodes over 3000 poly-
morphisms and mutations. The two terms, mutation and polymorphism, are defined
and used differently by different groups. Classical geneticists, particularly those associ-
ated with the use of genetics in a clinical setting, who are involved in diagnosis and
screening for Mendelian traits, use the term mutation to refer to genetic variations
that have a causative effect [i.e. a disease-causing mutation (DCM)] and use the term
polymorphisms to describe other variations found in the population. Many evolution-
ary geneticists also prefer to use this definition in this way. However, other geneticists
prefer to use the definition provided by Cavalli-Sforza and Bodmer (1971), whereby
genetic polymorphism was defined as “the occurrence in the same population of two
or more alleles at one locus, each with an appreciable frequency.” Most authors who
apply this definition agree that polymorphic loci are those for which the frequency of
the least common allele is greater than 1%. This works well when there are only two
alleles. It can become very complicated when considering some of our most complex
genes, such as the cystic fibrosis gene CFTR with over 1910 variations, some of which
are common (e.g. Δ508), but the majority of which are rare. In this situation it is
difficult to decide which terminology applies; Δ508 is a polymorphism, while most
of the other CFTR alleles are found at frequencies of less than 1% and are therefore
mutations. The dilema is should we call all the CFTR variants, including the Δ508
mutations, polymorphisms, or should we apply a mixture of terms as implied in the
definition above? There are similar problems with the naming in the major histocom-
patibility complex (MHC) (see Chapter 6).
The problem with the use of this terminology is not simply a matter of choice. Nearly
all genetic variation arises through mutation (deletions, duplication, insertions, and
unrepaired DNA damage). Therefore, most polymorphisms are simply common muta-
tions and, as a consequence, it is not possible to insist on the strict application of this
terminology. Though the debate on the correct use of these terms continues, they are
used interchangeably in the literature on complex disease. Both terms will be applied
throughout this book: polymorphism when describing common variations associated
with specific diseases or traits, and mutations when discussing rare variations and
evolutionary principals.
A common change in a single base pair (point mutation) is called a single nucleotide
polymorphism (SNP). The site at which a SNP is encoded is marked by the “rs” number

2967_Ch01.indd 5 07/07/2015 10:27

 6 CHAPTER 1 Genetic Diversity

(ref-SNP cluster ID number) – a unique ID number based on its position on a chromo-

some. SNPs are the smallest and most common type of genetic change in humans, and
account for an estimated 90% of all variation in the genome. There are currently thought
to be more than 38 million SNPs in the genome. Consequently, SNPs are most frequently
used as markers to identify genetic variation in human disease. The high frequency of
SNPs in the genome enables high-density profiling to be undertaken, which increases the
likelihood of accurate identification of disease-promoting alleles. When SNPs were first
used for screening for disease alleles on a genome-wide basis, only common SNPs (i.e.
those where the least frequent allele was present in 5% or more of the population) were
used. The reason for this was that rare SNPs were considered to be less statistically infor-
mative. Therefore, it was considered that larger numbers would have to be included in
the study to test rare alleles in order to have adequate statistical power. The problem with
excluding rare alleles is that potentially important associations with rare SNPs may have
been missed. However, as sample collections have become larger the potential to identify
statistically significant associations with less common SNPs has grown and the lowest
applied limit for SNP frequency has been adjusted downward. For example, instead of a
lowest frequency of 5%, a 1% limit can now be applied. The potential for using even lower
frequency SNPs will increase as study cohort sizes increase.
Another form of genetic variation that is quite common in the population is copy number
variations (CNVs). These occur when there are multiple numbers or copies of a specific
gene on a chromosome. CNVs are structural variations that can occur through deletions,
duplication, insertions, and translocations. They may represent large or small areas of the
chromosome. Good examples can be seen in Chapter 8 on pharmacogenetics.

Genetic variation can be measured by several methods

Though SNPs are the preferred markers for measuring genetic variation, other mark-
ers have been used in the past, including microsatellites. These are variable number
tandem repeat (VNTR) sequences in the genome. VNTRs can be “short” (involving
two to five nucleotide repeats) or “long” (involving more substantial repeat sequences).
VNTRs are still used in studies today and are especially useful where the candidate gene
is known or a specific region is being scanned. Earlier studies used restriction enzymes
to identify different VNTRs and SNPs. To determine VNTR genotypes, one or more
restriction enzymes that cut the DNA sequence above and below the region encoding
the VNTR sequences can be used and DNA fragments of different sizes can be obtained.
After digestion with the appropriate enzyme(s), the DNA sample can be run by electro-
phoresis on either an agarose gel or a polyacrylamide gel to reveal the size(s) of the frag-
ments and thus the number of sequence repeats in each individual sample. Genotypes
can be assigned based on the pattern obtained on the gel. This method is known as
restriction fragment length polymorphism (RFLP) analysis. RFLP analysis was also
used to detect SNPs where the differences in the DNA sequence can be detected by use
of restriction enzymes that cut the DNA at a particular sequence encoded by one allele,
but not the other. Multiple enzymes were often used when genotyping SNPs in order to
obtain readable accurate results. Different enzymes are used to detect different polymor-
phisms. Later studies substituted RFLP genotyping for more reliable polymerase chain
reaction (PCR) genotyping using primers specific for the gene sequence of interest.
This method uses a polymerase enzyme purified from the hot-springs “thermophilic”
bacteria Thermus aquaticus to amplify multiple copies of the gene sequence. These ampli-
fied sequences are then run out on a gel using the same process as that used with RFLP
fragments and genotypes can be assigned from the specific banding patterns obtained for
each sample (Figure 1.3).

2967_Ch01.indd 6 07/07/2015 10:27

 Genetic Variation 7

Five most common Four of the most common IL1RN

Molecular
IL1RN alleles genotypes
weight
markers 1 2 3 4 5 1/2 1/3 1/4 1/5

600

500

400

300

200

Figure 1.3: VNTRs used to genotype the interleukin (IL)-1 receptor antagonist gene (IL1RN). Genotyping
the IL-1 receptor antagonist 86 bp VNTR sequence using PCR and agarose electrophoresis. The figure shows the
five most common alleles (1–5) and the four most common genotypes (1/2, 1/3, 1/4, and 1/5). The molecular
weight markers indicate the approximate band sizes for each allele on the agarose gel as follows: allele 1, 410 bp;
allele 2, 240 bp; allele 3, 325 bp; allele 4, 500 bp; allele 5, 600 bp. Note the figure does not show the precise
position in the gel and the ladder is illustrative only. The genotypes for each sample can be assigned using the
band sizes obtained.

Alleles on the same chromosome are physically linked and inherited

as haplotypes
As genes are inherited on chromosomes and each chromosome carries a large number of
genes, genetic variations on a specific chromosome are inherited en masse as haplotypes.
Haplotypes do not change from one generation to the next because mutation rates are low,
but will change due to recombination during crossover. The potential for change is based
on the distance between the genes. One of the very interesting observations to emerge
from analysis of haplotypes is that for any small region of a chromosome, most people in a
population will carry one of approximately six different haplotypes that can be traced back
through history to a shared ancestry in the distant past. However, because recombination
events exchange pieces of DNA between chromosomes during meiosis, person A may share
the same haplotype with person B for a region at one end of a chromosome, yet have a dif-
ferent haplotype compared with person B at a position 1 million base pairs further down
the same chromosome. Person B, however, may share the same haplotype in the second
region with person C. By studying these haplotypes, it is possible to look back at genetic
events that may have happened thousands of years ago.

2967_Ch01.indd 7 07/07/2015 10:27

 8 CHAPTER 1 Genetic Diversity

Generation 1 Generation 2

A A
Mutation

B B
Normal allele New disease-causing allele
C C

D D

Figure 1.4: The development of a new mutation or allele in the A–B–C–D haplotype. The figure
illustrates the same chromosome in two individuals in two different generations (generations 1 and 2). On
the two chromosomes there are two patterns for the haplotype A–B–C–D. One includes a black normal band
(representing the normal allele) and one includes a dark gray band (representing a “new” mutated allele). In this
illustration the mutation may have arisen through recombination in meiosis. The bands illustrated are not the
same as those seen in Figure 1.1, which are the bands based on chromosome staining for karyotyping.

African populations tend to have a greater variety of haplotypes in any given region than
other populations. This is expected for a population that is older than all others, and there-
fore has had more time to diversify and develop more haplotype variations (Figure 1.4).
In younger populations, such as those in Europe and Asia, fewer haplotypes would be
expected because these populations have descended from smaller founder populations in
which a small subset of the total available haplotypes were present and there has also been
less time for new combinations to develop.

Linkage disequilibrium promotes conservation of haplotypes

in populations
The term linkage refers to the physical association (link) between two alleles that are
on the same chromosome. Linkage disequilibrium is a population genetics phenom-
enon whereby alleles on the same chromosome are transmitted together over genera-
tions within a population and such pairs or groups of alleles are found together more
frequently than expected by chance. In other words, there is non-random segregation
of the alleles. This is due to the physical proximity of the alleles in question and the
low rate of segregation at meiosis. The phenomenon of linkage disequilibrium is com-
mon when disease-causing alleles arise in a founder and the alleles are closely linked to
other markers along a chromosome. Crossover, however, may break up this disequi-
librium. When the loci are further apart, linkage disequilibrium breaks down quickly;

2967_Ch01.indd 8 07/07/2015 10:27

 Genetics and Evolution 9

HLA-DQB1*02 HLA-DRB1*03 C4A*Q0 HLA-B8 HLA-C7 HLA-A1

Figure 1.5: Extreme linkage disequilibrium. The MHC illustrates extreme linkage disequilibrium whereby
alleles at closely linked gene loci are inherited together more often than expected by chance. One example of this
is the HLA 8.1 ancestral haplotype (shown above), which is associated with an increased risk of many different
autoimmune diseases, but may also convey some survival advantage. The individual alleles are all common in the
normal northern European population, but occur together at frequencies far greater than expected by chance.
Thus, 60% or more of HLA-B8-positives have HLA-A1, and 90% or more of HLA-B8-positives have HLA-DRB1*03
and DQB1*02. HLA-B8 is the least common of all of these alleles at around 16%, and if the assortment were
random then these pairings would be in equilibrium and the likelihood of finding HLA-B8 and HLA-DRB1*03
together would be the sum of their individual frequencies. In this case, the values are approximately 20% for
HLA-DRB1*03 and 16% for HLA-B8. This would mean that instead of seeing approximately 14% of the population
with the combination HLA-DRB1*03–HLA-B8, we would see approximately 3%.

when the loci are close together, crossover is less common and linkage disequilibrium is
more likely to persist. Linkage disequilibrium can be used to provide useful information
about the distance between genes. Where there is extreme linkage disequilibrium, hap-
lotypes may be conserved and in many cases these conserved haplotypes are common in
the population. The human MHC (Figure 1.5) illustrates these concepts well (see also
Chapter 6). Linkage disequilibrium is a major tool in understanding modern genome-
wide linkage/association studies (GWLS/GWAS).

1.3 Genetics and Evolution

Evolution in population genetics refers to changes in the gene pool resulting in the pro-
gressive adaptation of populations to their environment. Four main processes account
for most of the changes in allele frequency in populations: mutation, migration, natural
selection, and random genetic drift. Together these form the basis of cumulative change
in the genetic characteristics of populations, leading to the descent with modification that
characterizes the process of evolution.
If the population is large enough, allele frequencies remain stable and do not change
significantly as a result of random reproduction, and therefore other processes must be
responsible for changes in allele frequency. Genetic variation within populations can be
increased by migrations and mutations that introduce new alleles into the population.
Variation within populations can also be increased by some types of natural selection,
such as over-dominance, in which both alleles are favored. These evolutionary forces
that act to maintain or increase genetic variation are shown in the upper-left quadrant of
Table 1.1. The lower-left quadrant of Table 1.1 shows evolutionary forces that decrease
genetic variation within populations. These forces include genetic drift, which decreases
variation through the fixation of alleles, and some forms of natural selection, such as
directional selection, which selectively favors one allele over the other.
Evolutionary forces also affect the genetic divergence between populations and are shown
on the right quadrants of Table 1.1. Genetic divergence between populations is increased
by mutation, genetic drift, and natural selection. Different mutations can arise within each
population and therefore mutations almost always increase divergence between popula-
tions. Positive natural selection can increase or decrease divergence between populations
depending on the favored alleles. If different alleles are favored, populations will diverge;
however, if natural selection favors the same allele in different populations, genetic diver-
gence between populations will decrease. Migration reduces divergence between populations

2967_Ch01.indd 9 07/07/2015 10:27

 10 CHAPTER 1 Genetic Diversity

Table 1.1 Mutation, migration, genetic drift, and natural selection have different effects on genetic
variation within populations and on genetic divergence between populations.

Within populations Between populations

Increase genetic variation Migration Genetic drift
Mutation Natural selection
Natural selection Mutation
Decrease genetic variation Genetic drift Migration
Natural (directional) selection Natural selection

because blending the total gene pool makes populations similar in terms of their genetic
composition. Note that migration and genetic drift act in opposite directions: migration
increases genetic variation within populations and reduces divergence between populations,
whereas genetic drift reduces genetic variation within populations and increases divergence
among populations. Mutations mostly increase the genetic variability both within and
between populations, though they can occasionally restore the wild-type. Natural selec-
tion, by contrast, can either increase or reduce genetic variation within a population and
increase or reduce genetic divergence between populations.
Finally, before considering each of these processes in turn, it is important to make it clear
that populations are simultaneously affected by many evolutionary forces acting at the
same time and that evolution results from the complex interplay of these processes.

Mutation is the major cause of genetic variation

Almost all genetic variants arise through some form of mutation. New combinations of
these mutations may then arise through recombination in meiosis. Meiosis is a process
through which cells are able to divide and produce haploid gametes. It is sometimes called
a reductive division because there are two stages of cell division, but only one round of
DNA replication. Thus, four haploid gametes are created for each diploid spermatocyte
(i.e. sperm cell). In oocytes (i.e. egg cells) the situation is different. Division is asymmetric,
unlike that for spermatocytes, and the cell division results in a large secondary oocyte and
a smaller polar body that is discarded. Evolution through natural selection depends on
these processes because there has to be genetic variation in the population before evolution
can take place. There can be no selection without genetic variation in a population.
A mutation is a heritable change in the DNA sequence. This means that the structure of
DNA has been changed permanently and this alteration can be passed from mother to
daughter cells during cell division. If a mutation occurs in reproductive cells, it may also
be passed from parent to offspring. This kind of mutation is responsible for changing allele
frequencies in a population and is an essential process in evolution, as mutations provide
the variation that enables humans to change and adapt to their environment when selec-
tive pressure is applied. Some mutations may be selectively neutral, which means they do
not affect the ability of the organism to survive and reproduce. Only a very few mutations
are favorable for the organism and contribute to evolution.
Mutation rates are typically low. The mutation rate (μ) is the frequency of such change and
it is expressed as the number of mutations per locus per gamete per generation. Estimating
the mutation rate is difficult because mutations are rare. In humans, most information on
mutation rates comes from studies of rare Mendelian autosomal dominant diseases where

2967_Ch01.indd 10 07/07/2015 10:27

 Genetics and Evolution 11

it is much easier to estimate mutation rates than it is for Mendelian autosomal recessive
diseases or for non-Mendelian complex diseases. Estimates of mutation rates for a variety
of human genes lie between 10−6 and 10−5 mutations per locus per gamete per generation.
However, the estimated mutation rate is higher for some Mendelian diseases. For example,
in type 1 neurofibromatosis and Duchenne muscular dystrophy the estimated mutation
rate is as high as 10−4. This is 10–100 times greater than the general mutation rates.
The OMIM (Online Mendelian Inheritance in Man) database (http://www.ncbi.nlm.
nih.gov/omim) lists human genes and it is an excellent source for information on spe-
cific genetic diseases. For many diseases, a larger number of genetic mutations have been
identified than those listed on OMIM, but this is a good starting point to catalog genetic
variations that are linked to or associated with a specific disease and it also has a very good
bibliography for each disease.
Introducing the Hardy–Weinberg Principle
The Hardy–Weinberg Principle (HWP) or Hardy–Weinberg Equilibrium (HWE) test
is one of the central pillars of statistical analysis in population genetics (Table 1.2). The
term equilibrium in population genetics refers to something (an allele or gene) that is
in a state of balance. Equilibrium arises when alleles remain unchanged over time. The
HWE test assesses how allele frequencies have changed from generation to generation.
The HWP states that in a large breeding population, provided none of the evolutionary
forces described below are operating, allele frequencies will remain the same from genera-
tion to generation. In practice, the HWE test can be used to understand the change in
allele frequencies over time and indicate whether evolution has taken place. HWE is also
used in studies of complex disease to determine whether there is bias in the study sample
and in the qualitative assessment of studies. The HWP is a complex principle and the basic
concept and its application are discussed in more detail in Section 1.4.

Genetic variation caused by mutation alters allele frequencies

in populations
The rate at which a genetic variation increases or decreases is determined by the mutation
rate. Consider the example of a single locus with two alleles A1 and A2 with frequencies
p and q, respectively, in a population of 10 diploid individuals. In this example, the pool
of alleles for this gene within the population will consist of 20 allele copies. If there are 15
copies of A1 and five copies of A2 in the population, then the frequency of each allele is
p = 0. 75 and q = 0.25. If we suppose that a mutation changes one A1 allele into an A2
allele, after one mutation there will be 14 copies of A1 and six copies of A2, and the fre-
quency of A2 will increase from 0.25 to 0.30; a mutation has therefore changed the allele
frequency for the population. If copies of A1 continue to mutate to A2, the frequency of
A2 will continue to increase, while the frequency of A1 will decrease.

Table 1.2 The HWE (p2 + 2pq + q2 = 1) dictates that the sum of allele genotypes is always 100% and
this formula can be used to determine the expected frequency of the different genotypes in a
population.

Maternal gamete Paternal gamete

A (p) a (q)
A (p) AA (p ) 2
Aa (pq)
a (q) Aa (pq) aa (q2)

2967_Ch01.indd 11 07/07/2015 10:27

 12 CHAPTER 1 Genetic Diversity

Thus changes in the frequency of the A2 allele (Δq) depend on:

• μ: the mutation rate A1 to A2
• p: the frequency of the A1 allele in the population
When p is large, many copies of A1 are available to mutate to A2 and the amount of
change will be relatively large. However, as more mutations occur and p decreases, fewer
copies of A1 will be available to mutate to A2. The change in A2 frequency as a result of
mutation equals the mutation rate multiplied by the allele frequency:
Δq = μp
So far, we have considered only the effects of forward mutations (A1 → A2); however,
reverse mutations (A2 → A1) can also occur. Reverse mutations will occur at a rate ν,
which will probably be different from the forward mutation rate μ. When a reverse muta-
tion occurs, the frequency of the A2 allele decreases while the frequency of A1 increases.
The overall change in allele frequency (A1 and A2) is a balance between the two opposing
forces of forward and reverse mutations:
Δq = μp − νq
These allele frequencies are determined only by the forward and reverse mutation rates,
and they will increase or decrease until the HWE is established. When the equilibrium is
established, the HWP indicates that genotype frequencies will remain the same.
The mutation rates of most human genes are low and changes in allele frequencies due
to mutation in one generation are very small. Therefore, it may take a long time to reach
the HWE. For example, consider a locus where the forward and reverse mutation rates
for alleles are μ = 1 × 10−5 and ν = 0.5 × 10−5 per generation, respectively, and the allelic
frequencies are p = 0.85 and q = 0.15. The change in allele frequency per generation due
to mutation is:
Δq = μp − νq = (1 × 10−5)(0.85) − (0.5 × 10−5)(0.15) = 7.75 × 10−5 = 0.0000775
This shows that the change due to mutation in a single generation is extremely small and
because the frequency of p drops as a result of each mutation, the frequency of change will
become even smaller over time, as shown in Figure 1.6.

Migration and dispersal cause gene flow

Another process that introduces change in the allele frequencies is the gene flow. Gene
flow is the result of migration where many individuals of one population move en masse
from one geographic location to another. Though migration is the main cause of gene
flow, it can also result from population dispersal, i.e. the spreading of individuals away
from others. Migration has a similar impact to mutation as new alleles are introduced into
a local gene pool by the migrants. In this case, however, the new alleles are new only to the
population into which the migrants move and they are not the result of new mutations.
In the absence of migration, the allele frequencies in each local population can change
independently through genetic divergence. As a consequence there will be differing fre-
quencies of common alleles among local populations and some local populations will
possess certain rare alleles not found in others. This effect of the accumulation of genetic
differences among subpopulations can be reduced if subpopulations undergo migration.
Human population migration leads to mixing of the gene pool, preventing populations
from becoming too different from one another. A relatively small amount of migra-
tion among subpopulations, in the order of just a few migrant individuals in each local

2967_Ch01.indd 12 07/07/2015 10:27

 Genetics and Evolution 13

p decreases due to forward mutations

Allelic frequency (p)

At equilibrium forward mutation is

balanced by reverse mutation

p increases due to reverse mutations

Number of generations

Figure 1.6: Changes due to recurrent mutations slows as the frequency p of an allele drops. The figure
shows the influence of mutations on the frequency (p) of a single allele. The mutation rate in a single generation is
exceedingly small and because the frequency of allele p drops with each mutation, the rate of change will become
even slower over time. Reverse mutations will increase the frequency of allele p. Eventually the actions of the
opposing forces, i.e. forward and reverse mutations, will establish equilibrium in the frequencies of the alleles p and q.

population in each generation, can be sufficient to prevent the accumulation of high levels
of genetic differentiation between populations. Migration adds genetic variation to popu-
lations and increases genetic differences within the recipient population. However, genetic
diversification can also occur in spite of migration when other evolutionary forces such as
natural selection are sufficiently strong.

Allele frequencies can change randomly via genetic drift

Sewell Wright (1931) introduced the concept of random genetic drift into the study
of population genetics. Genetic drift refers to changes in allele frequencies in a popula-
tion due to random fluctuations. These are the frequencies of alleles found in gametes
that unite to form zygotes. Zygotes are single diploid cells formed by the combina-
tion of a single haploid sperm and a single haploid egg, and the alleles found in these
gametes vary from generation to generation simply by chance. The zygote referred to is
the fertilized egg cell and is the cell from which all other cells in the body are derived.
Over time genetic drift usually results in either the loss of an allele or preservation of
the allele in the population and fixation at 100%. The rate at which genetic drift occurs
depends on the population size and on the initial allele frequencies.
To illustrate the concept of genetic drift we can consider the following hypothetical simu-
lation of changes in allele frequencies for a single gene in five populations of 20 individuals
each (N = 20) over many generations (Figure 1.7). Suppose there are only two alleles, A
and B, and the allele frequencies are identical in all five populations, each with a frequency
of 0.5. In the five small populations, the allele frequencies will fluctuate from generation to
generation. Eventually, in each of the five populations one of the alleles will be eliminated

2967_Ch01.indd 13 07/07/2015 10:27

 14 CHAPTER 1 Genetic Diversity

1.0
Population 1 Population 4
Population 2 Population 5
Population 3
0.8

0.6
Frequency of A

0.4

0.2

0.0

0 20 40 60 80 100
Generations

Figure 1.7: Hypothetical model of genetic drift in five different populations. The figure illustrates the
potential influence of genetic drift in five populations. The model considers the different outcomes over 100
generations. In all cases the starting allele frequencies are 0.5 for A and 0.5 for a and each population is assigned
20 individuals (N = 20). In all cases the frequency of allele A only is considered. The simulations obtained over 100
generations indicate a variety of outcomes with peaks and troughs moving towards frequencies of 1 or 0 for A in
every case.

and the other will be fixed at 100%. As in this case the gene is now monomorphic (i.e.
there is only one allele), all individuals are homozygous for the predominant allele and
there can be no further fluctuation in that population. Genetic drift can lead to homozy-
gosity even in large populations, but this will take many more generations to occur.
Figure 1.7 also illustrates another effect of genetic drift. In the example all five popula-
tions begin with the same allele frequencies (50% or 0.5 for both alleles), but because
genetic drift is random, the frequencies in different populations do not change in the same
way and so populations gradually acquire genetic differences. Consequently genetic drift
will increase the genetic variation between different populations and there will be genetic
divergence over time. In contrast, the opposite effect may also be seen whereby there is
reduced genetic variation within populations. Through random change, an allele may
eventually reach a frequency of either 100% or 0, at which point all individuals in the
population are homozygous for one allele. When an allele has reached a frequency of 1,
we say that it has reached fixation. The other allele is lost (reaching a frequency of 0) and
can be restored only by migration from another population or by mutation. Fixation leads
to a loss of genetic variation within a population. Given enough time, all small popula-
tions will become fixed for one allele or the other. Which allele becomes fixed is random
in the absence of other forms of selection pressure, though it may be determined by the
initial frequency of the allele. If the initial frequency of two alleles is 0.5, both alleles have

2967_Ch01.indd 14 07/07/2015 10:27

 Genetics and Evolution 15

an equal probability of fixation; however, if one allele is initially more common, it is more
likely to become fixed.
Genetic drift can lead to the fixation of deleterious, neutral, or beneficial alleles, but the
effect is greatly influenced by the population size. Allele loss and fixation due to genetic
drift occur more rapidly in small populations. Therefore, in nature, both population size
and geography can influence genetic drift, and consequently the genetic composition of
a population. Some human populations have settled on small islands or in geographically
isolated areas and the allele frequencies within these small isolated populations are more
susceptible to genetic drift. A population may be reduced in size for a number of genera-
tions because of epidemic disease, famine, or other natural or even man-made disasters. As
genetic drift is a random process, small isolated populations tend to be more genetically
dissimilar to other populations. Geography and population size can influence the effect of
genetic drift by creating either a bottleneck or a founder effect.
Bottleneck effect
Changes in population size may influence genetic drift via the bottleneck effect. Natural
and man-made disasters such as famines or war may reduce the size of the founder popula-
tion. Depending on the size of the effect and the original population, this can change the
degree of genetic variability within the population. Such events may randomly eliminate
most of the members of the population with or without regard to the genetic composition
or through selection of a group, for example favoring those with specific alleles following
infectious epidemics. This can create a bottleneck effect within a population whereby the
level of genetic variation is extremely limited (Figure 1.8).

Parent Bottleneck Surviving Next

population (drastic reduction individuals generation
in population)

Figure 1.8: The bottleneck effect. The bottleneck effect can occur as a result of major environmental events
such as famine or plague whereby the founder or parent population is drastically reduced. This may affect the
degree of genetic variability within a population. Natural selection may also operate under these circumstances,
favoring those with specific alleles, especially when the disaster involves infectious disease.

2967_Ch01.indd 15 07/07/2015 10:27

 16 CHAPTER 1 Genetic Diversity

The thrifty gene hypothesis

The thrifty gene hypothesis was proposed by J. V. Neel in 1962 to explain the growing
incidence of diabetes in the Western world. Neel suggested that a thrifty genotype that
was more capable of modifying insulin release and glucose storage may have a survival
advantage. Though this worked well for our ancestors who had to survive periods of fam-
ine, possession of the thrifty genotype in a modern Western society with a plentiful food
supply may be a disadvantage as it may cause elevated insulin levels and excessive energy
stores. This is seen in clinical cases of type 2 diabetes (T2D).
This hypothesis has been supported by a number of research groups. Work on late-
Paleolithic human ancestors indicates alternating periods of abundance and famine and
recently two genes or gene sequences have been said to be associated with thrifty character-
istics. These are the insulin (INS) VNTRs and the apolipoprotein E (APOE) gene. More
variation has been recorded in the INS VNTR genes in African versus non-African popula-
tions (27 versus three variants, respectively). APOE has been linked with Alzheimer’s disease
and cardiovascular disease. APOE2, which is common in Mediterranean populations, is less
common in African and Native American populations. Interestingly, studies have shown
that women who possess APOE4 tend to have more children than those with APOE2.
Despite this emerging support for the thrifty gene hypothesis it is still controversial. One
reason for this may be that the effects seen are not genetic, but are determined by other
factors. Some authors have gone as far as to suggest this may be a thrifty phenotype as
opposed to a thrifty genotype effect. In this latter hypothesis the authors suggest that the
environment is responsible for the phenotypic variation seen and not the genes, with
nutrition in newborn and infant children being particularly important.

Founder effect
Geography and population size may also influence genetic drift via the founder effect.
The founder effect involves migration, where a small group of individuals separate from a
larger population and establish a colony in a new location. For example, a few individuals
may migrate from a large continental population and become the founders of an island
population. The founding population is likely to have less genetic variation than the origi-
nal population from which it was derived and consequently the allele frequencies in the
founding population may differ markedly from those of their original population.

Natural selection acting on different levels of fitness affects the

gene pool
The final process that brings about changes in allele frequencies is natural selection.
Selection is the differential reproduction of genotypes. Selection represents the action of
environmental factors on a particular phenotype and genotype through selective pressure.
Natural selection is the consequence of differences in the biological fitness of individual
phenotypes. Biological fitness is a measure of fertility and reproductive success of a geno-
type compared with other genotypes in a population. Genotypes with a greater level of
biological fitness contribute more to the gene pool of succeeding generations. Differential
fitness among genotypes leads to changes in the frequencies of the genotypes over time,
which in turn leads to changes in the frequencies of the alleles that make up the gene pool.
The effect of natural selection on the gene pool of a population depends on the fitness
values of the genotypes in the population. Thus, selection may operate at any time from
conception to the end of the reproductive period. There are three major forms of natural
selection: purifying or negative selection, positive or adaptive Darwinian selection, and
balancing selection (Figure 1.9)

2967_Ch01.indd 16 07/07/2015 10:27

 Genetics and Evolution 17

Column 1 Column 2 Column 3

A Negative selection

B Positive selection

C Balancing selection

Figure 1.9: Three different models of natural selection. Rows A, B and C illustrate three patterns of natural
selection (selection signatures). The columns represent three generations: the first column shows the starting
group of four individuals looking at the same chromosome in each, the second column shows the first generation
with mutations, and the third column shows the final outcome for the chromosomes in the three different
patterns of natural selection. Each circle represents a polymorphism, within a haplotype. White circles represent
mutations under neutrality, black circles indicate deleterious mutations, and gray circles indicate advantageous
mutations. Pattern A illustrates genetic polymorphisms under negative selection. Deleterious mutations arise
(black dot) and they can be removed immediately (if severely deleterious, e.g. line 3 in column 3) or kept at low
frequencies (if weakly deleterious, e.g. line 2 in column 3). Linked neutral polymorphism will also disappear (or
be kept at low frequencies, e.g. line 3 in column 3). Pattern B illustrates genetic polymorphism under positive
selection. When a new advantageous mutation arises (shaded circle in line 2, column 2), the allele increases in
frequency (in the population) along with linked neutral polymorphisms (lines 3 and 4 in column 3, which now
resemble line 2, column 2). Pattern C illustrates balancing selection. Two new alleles are shown (shaded and black
circles) and, if they confer advantage in the heterozygous state, they will increase to intermediate frequencies.
Linked neutral polymorphisms will also increase to intermediate frequencies. (From Ermini L, Wilson IJ, Goodship
TH & Sheerin NS [2012] Immunobiology 217:265–271. With permission from Elsevier.)

Purifying selection
Purifying natural selection (also called negative selection) reduces the frequency of det-
rimental alleles in a population. New mutants often have detrimental effects on biological
fitness and purifying selection reduces the number of new mutations in the gene pool. In
humans, 38–75% of all new non-synonymous mutations are thought to be affected by
moderate to strong negative selection. Deleterious mutations are generally found at low
frequencies because of the adverse effect they may have on biological fitness. Negative
selection is responsible for the removal (or maintenance at low frequencies) of mutations
associated with severe Mendelian disorders. Mendelian disease genes come under wide-
spread purifying selection, especially when the disease mutations are dominant.
Positive Darwinian selection
Some mutant alleles introduced to a population by gene flow may be advantageous. In
this case a directional genetic change may allow a population to adapt to its environment
and new, better adapted alleles may replace old, less well adapted alleles. Such selection of

2967_Ch01.indd 17 07/07/2015 10:27

 18 CHAPTER 1 Genetic Diversity

alleles that are advantageous is called adaptive Darwinian selection or positive Darwinian
selection. Under the action of positive selection advantageous alleles rapidly achieve high
frequencies within the population. This occurs at a rate much faster than that of a neu-
trally selected allele. As a consequence of this rapid increase few recombination events
will take place and any neutral variation linked to selected variants will also increase in
frequency within the population. This process often results in a transitory increase in the
strength of linkage disequilibrium between alleles on the same haplotype.
Balancing selection
A third form of natural selection is balancing selection, whereby heterozygotes show a
higher level of biological fitness than homozygotes. This leads to the maintenance of two
or multiple alleles in a population at a given locus. Polymorphisms are maintained in the
population for a longer period of time than expected. Balancing selection is often referred
to as heterozygote advantage, especially in cases where a mutant allele known to cause a
disease in homozygotes is found at a high frequency in heterozygous healthy members of
the population. Genome scans suggest that balancing selection is less extensive than posi-
tive selection. However, balancing selection does occur. The two examples below illustrate
heterozygous advantage in two autosomal recessive Mendelian diseases.
Cystic fibrosis is one of the most common autosomal recessive diseases in Northern
European populations, affecting approximately 1:2500 new born children. The caus-
ative gene in cystic fibrosis is the cystic fibrosis transmembrane conductive regulator
gene (CFTR) and there are currently 1910 mutations on the CFTR mutation database
(http://www.genet.sickkids.on.ca/StatisticsPage.html) (Figure 1.10). Carriers of the
CFTR mutations (heterozygotes) appear to have, or have had in the past, some repro-
ductive advantage over wild-type normal homozygotes. There has been debate over what
this advantage might be. The CFTR gene encodes a membrane chloride channel pro-
tein that is required by some bacteria such as those belonging to the genus Salmonella

Δ508

W1282X

G542X

G551D

N1303LYS

Rare

Figure 1.10: The five most common CFTR gene mutations. The five mutations listed account for over 70% of
the overall mutations and Δ508 is the most common of all, accounting for approximately two-thirds of all cases.
All of the other mutations, of which there are at least 1905, are found at frequencies of less than 1% and together
these account for approximately 30% of all CFTR mutations.

2967_Ch01.indd 18 07/07/2015 10:27

 Calculating Genetic Diversity: Determining Population Variability 19

Figure 1.11: Red blood cells in sickle cell disease. Sickle cells are shaped like a harvesting sickle and, unlike the
normal doughnut-shaped red blood cells, these cells can be hard with sharp edges that can damage the wall of
small blood vessels as they passage through the body. They will often clog the flow of blood and break up as they
pass through the small blood vessels.

(e.g. Salmonella typhi) to enter into epithelial cells. One explanation is that carriers of a
mutant CFTR allele may be more resistant to infection by such bacteria than those with
two copies of the wild-type gene.
Sickle cell anemia is another example. This is a genetic autosomal recessive blood disorder
that is characterized by red blood cells that occasionally assume an abnormal, rigid, sickle
shape (Figure 1.11). The β-globin allele variant, called HbS, is responsible for the sickling
of red blood cells seen in the disease. Despite the high mortality associated with homo-
zygosity the sickling allele HbS is found at high frequencies in Africa (up to 30%). One
possible explanation for the abundance of the HbS allele in Africa is that heterozygosity
confers some resistance to malaria.

1.4 Calculating Genetic Diversity: Determining

Population Variability
Genotype and allele frequencies illustrate genetic diversity
The genetic diversity of a population can be described using genotype or allele frequencies.
A large number of samples from a population are usually collected, and the genotype and
allele frequencies are calculated. The genotype and allele frequencies of the sample popu-
lation are then used to estimate the diversity of the population. To calculate a genotype
frequency, the number of individuals having the same genotype is divided by the total
number of individuals in the sample (N ). For a locus with three genotypes, AA, Aa, and
aa, the frequency ( f ) of each genotype is:

number of AA individuals
f ( AA ) =
N

2967_Ch01.indd 19 07/07/2015 10:27

 20 CHAPTER 1 Genetic Diversity

number of Aa individuals
f ( Aa ) =
N

number of aa individuals
f ( aa ) =
N
The sum of all the genotype frequencies always equals 1 (or 100%).
Genotypes are not permanent. They are disrupted in the processes of segregation and
recombination that take place when individual alleles are passed to the next generation
through the gametes. Alleles, in contrast, are not broken down and the same allele may be
passed from one generation to the next. For this reason the calculation of allele frequencies
is often the preferred choice when determining the genetic variability of a population. In
addition, there are always fewer alleles than genotypes, e.g. for the gene with two alleles A
and a above, there are two alleles, but there are three genotypes. By using alleles, popula-
tion diversity can be described in fewer terms than by using genotypes. Finally, by using
allele frequencies in case control population studies rather than genotype frequencies, no
assumptions about the impact of homozygosity or of heterozygote advantage are being
made. This is especially important in the context of complex disease where in the absence
of a clear pattern of inheritance it would not be appropriate to make any such assumption,
at least in the initial stages of analysis.

Allele frequency refers to the numbers of alleles present in

a population
The number of copies of an allele at a locus is divided by the total number of all alleles in
the sample:

number of copies of the allele

Frequency of an allele =
number of copies of all alleles at the locus

If we consider a gene with only two alleles A and a and we suppose the frequencies are p
for allele A and q for allele a; then p and q can be calculated as:

2n AA + n Aa
p = f ( A) =
2N

2naa + n Aa
q = f (a ) =
2N
In this equation nAA, nAa, and naa represent the numbers of AA, Aa, and aa individuals, and
N represents the total number of individuals in the sample it is necessary to divide by 2N
because being diploid means each individual has two alleles for each gene (one from the
maternal locus and one from the paternal locus).
The sum of the allele frequencies is always 1 (100%) ( p + q = 1); therefore where there are
only two alleles, q can be determined by simple subtraction after p has been calculated:

q = 1 − p

2967_Ch01.indd 20 07/07/2015 10:27

 Calculating Genetic Diversity: Determining Population Variability 21

These calculations apply only where there are two alleles. In cases where there are several
different alleles at a locus the calculation used is based on the same principle, but is more
complicated. Statistical software will usually be used to perform complex calculations, but
it is important to understand the underlying principles in any analysis.

Heterozygosity provides a quantitative estimation of genetic variation

Knowing the frequency of heterozygotes (i.e. those carrying both wild-type and mutant
alleles for the same gene) can be a very useful tool for the quantitative estimation of
genetic variation in a population. Where mutations are common, heterozygotes are com-
mon and homozygotes can be quite rare. Heterozygosity can provide information on the
structure and even the history of a population. High levels of heterozygosity reflect high
levels of genetic variability, while low levels of heterozygosity indicate low levels of genetic
variability. Very low levels of heterozygosity can indicate the effects of small population
sizes created by population bottlenecks. Often the observed levels of heterozygosity are
compared with what is expected under HWE (see below). If the observed heterozygosity
deviates from HWE or is lower than expected, this discrepancy may be attributed to non-
random mating. This can occur in small isolated populations when individuals select a
closely related mate more often than would be expected by chance in a larger population.
Non-random mating does not change the allele frequencies, but leads to an increase in
homozygous offspring over time because the parents are more likely to be genetically similar.
Consequently, there will be a decrease in heterozygosity in such populations. This places indi-
viduals and the population at a greater risk from Mendelian recessive diseases. The impact
of accumulating deleterious homozygous traits is called inbreeding depression. This term
refers to the loss of population vigor due to reduced genetic variability or reduced biologi-
cal fitness in a given population as a result of the breeding between related individuals. This
phenomenon is often the result of a population bottleneck. If heterozygosity is higher than
expected, an isolated breakout may have taken place through contact with individuals from
another population, which can introduce a temporary excess of heterozygotes.
Expected heterozygosity can be measured using the simple formula:

HE = 1 − ∑p
i =1
i
2

In this equation n is the number of alleles and pi is the frequency of the ith allele at a locus.
The value of this measure ranges from 0 for no heterozygosity to nearly 1 (i.e. 100%) for
a system with a large number of equally frequent alleles.

The HWP is a complex but essential concept in population genetics

The HWP, which was introduced earlier in this chapter, is one of the most important
statistical principles in population genetics, and because it is an abstract and quantita-
tive principle it is one of the hardest concepts to understand. Therefore, it needs to be
discussed in some detail. We may wonder why a recessive trait is not gradually eliminated
over the course of time or how the O blood type can be the most common blood type if
it is a recessive trait. These questions reflect the assumption that the dominant allele in
a population will always be found at the highest frequency and the recessive allele will
always be less common. The HWP addresses these questions and enables us to consider
the frequency of alleles over the time.

2967_Ch01.indd 21 07/07/2015 10:27

 22 CHAPTER 1 Genetic Diversity

The HWP depends on certain assumptions, of which the most important are:
• Mating in a population is random – there are no subpopulations that differ in allele
frequency.
• Allele frequencies are the same in males and females.
• All the genotypes are equal in viability and fertility – selection does not operate.
• Mutation does not occur.
• Migration into the population is absent – gene flow does not occur.
• Genetic drift does not occur.
• The population is sufficiently large that the frequencies of alleles do not change from
generation to generation.
The HWP states that after one generation of random mating, in a large breeding popula-
tion, where the restrictions listed above all apply, single-locus genotype frequencies can be
presented as a binomial function (where there are only two alleles) or multinomial func-
tion (where there are multiple alleles). Under the above conditions and over time, allele
frequencies will reach equilibrium and remain constant from generation to generation.

Calculating expected genotype frequencies using the HWP

The HWE can be calculated using the simple mathematical formula:
p2 + 2pq + q2 = 1
In this equation p and q represent the frequencies of alleles (Figure 1.12). It is impor-
tant to note that the sum of the allele frequencies ( p + q) is always equal to 1. To illus-
trate HWE, we can consider a single gene with two alleles A and a in a large population
with frequencies p and q, respectively (Box 1.1). First, we must assume that male and
female gametes interact randomly, and all of the major assumptions above remain true
(i.e. there is no mutation, selection, random genetic drift, or gene flow). We can then
use a simple calculation based on the Punnett Square illustrated in Table 1.2. The
Punnett square is perhaps the most common of all mathematical representations used
in the study of the genetics of complex disease. The data entered in the table can be used
to generate odds ratios (ORs) and significance values via the χ2 test. It is important to
note that the exact numbers and not the percentages must be used in the calculations to
generate accurate outcomes.

Different populations may have different allele frequencies

Note that heterozygotes are more common when allele frequencies are intermediate; how-
ever, when one allele is more common than the other, homozygosity for that allele is
increased and heterozygosity reduced. In this illustration heterozygotes have a maximum
frequency of 50%, which is achieved when p = q = 0.5. When either locus is monomor-
phic (p = 1 or q = 1), there are no heterozygotes.
HWE allows us to describe a population only considering the frequencies of n alleles at a
particular locus. For the less statistically inclined geneticists, the HWE principle is mostly
applied to ensure validation of data from population studies. Departure from the expected
distribution of genotypes generally indicates problems in sample recruitment or some
other form of population bias. Conversely, there is more confidence in the results when
there is equilibrium.

2967_Ch01.indd 22 07/07/2015 10:27

 Calculating Genetic Diversity: Determining Population Variability 23

q
1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
1

0.8

q2 p2
Genotype frequency

0.6
2pq

0.4

0.2

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
p

Figure 1.12: A plot of the HWE-based genotype frequencies (p2, 2pq, and q2) as a mathematical function
of allele frequencies (p and q). The plot illustrates the influence of allele frequencies on genotype frequencies
and shows what we can expect from the HWE test. The plot shows how the two alleles p and q determine
genotype frequencies and these change as the allele frequencies change. For example the closer q is to 1, the
lower the value of p and the higher the value for the q2 genotype (homozygous q). When p and q are both set at
0.5 (50%), then the frequency of pq heterozygotes is high.

BOX 1.1 CALCULATING EXPECTED GENOTYPE FREQUENCIES

USING THE HWP

The HWE can be calculated using the simple mathematical formula:

p2 + 2pq + q2 = 1
In this equation p and q represent the frequencies of alleles. It is important to note that
the sum of the allele frequencies ( p + q) is always equal to 1.
To illustrate HWE, we can consider a single gene with two alleles A and a in a large popu-
lation with frequencies p and q, respectively. First, we must assume that male and female
gametes interact randomly and all of the following assumptions are true:
• Mating in a population is random – there are no subpopulations that differ in allele
frequency.
• Allele frequencies are the same in males and females.
• All the genotypes are equal in viability and fertility – selection does not operate.
• Mutation does not occur.
• Migration into the population is absent – gene flow does not occur.
• Genetic drift does not occur.
• The population is sufficiently large that the frequencies of alleles do not change from
generation to generation.

2967_Ch01.indd 23 07/07/2015 10:27

 24 CHAPTER 1 Genetic Diversity

BOX 1.1 CALCULATING EXPECTED GENOTYPE FREQUENCIES

USING THE HWP (Continued)

Then we can use a simple calculation based on the Punnett Square illustrated in Table
1.2. The top side of the square is divided into proportions p and q representing the
frequencies of the male alleles A and a, respectively. The left side represents the same
proportions, but for the female alleles. If we assume there is random union of gam-
etes we can apply the product rule of probabilities. Imagine a pool with male and
female gametes, p with A alleles and q with a alleles, and where zygote formation
occurs by random union. The upper-left square represents the frequency of the homo-
zygous genotype AA. The expected frequency is simply the product of the separate
allele frequencies.
Frequency of AA = p × p = p2 (Homozygous for A)
The frequency of homozygous genotype aa is shown in the lower-right square:
Frequency of aa = q × q = q2 (Homozygous for a)
The other two squares illustrate the third possibility, i.e. Aa heterozygotes. The total pro-
portion of Aa heterozygotes can be calculated:

Frequency of upper-right square Aa = pq (Heterozygous ) (1)

Frequency of lower-left square Aa = pq (Heterozygous) (2)

Total frequency of Aa = Aa (1) + Aa (2) = 2pq (Heterozygous Aa)

The three different genotypes AA, Aa, and aa are formed in proportions p2, 2pq, and q2,
respectively. The sum of allelic frequencies is:
( p + q)2 = p2 + 2pq + q2
This illustrates the HWE. It is important to note that Hardy–Weinberg proportions are
binomial in this case (i.e. there are only two alleles). Given any set of genotype frequen-
cies (AA, Aa, and aa), the HWE predicts that after one generation of random mating,
provided the assumptions above are all met, the genotypic frequencies will be in the pro-
portions p2, 2pq, and q2. For example, given the initial genotype frequencies of AA = 0.4,
Aa = 0.4, and aa = 0.2, where p = 0.6 (frequency of allele A) and q = 0.4 (frequency of
allele a), after one generation the genotype frequencies become:
p2, 2pq, q2 = (0.6)2, 2(0.6)(0.4), (0.4)2 = 0.36, 0.48, 0.16.
The genotype frequencies will stay in these proportions generation after generation pro-
vided mating is random and the assumptions above are all met. Deviation from any of
the above conditions can led to an increase or decrease in allele frequencies from one
generation to another and this will impact on the genotype distribution. Finally, it is
important to note that this example deviates from the statement earlier in this chapter
that states it takes a long time to reach HWE. This is because mutation rates in human
genes are low; a full explanation of this is given earlier in this chapter.

2967_Ch01.indd 24 07/07/2015 10:27

 Calculating Genetic Diversity: Determining Population Variability 25

Probabilities
Probability represents the chance of a given event (or outcome) to occur. It is a measure
of the uncertainty and can be a number between 0 and 1. There are different schools
of thought regarding the concept of probability. The use of probability is illustrated in
Box 1.2. In the case of mutually exclusive events (e.g. tails or heads when the same coin is
flipped at the same time), the combined probabilities of the outcomes (i.e. heads or tails)
can be calculated by summing the individual probabilities of each event. This is known as
the sum rule. When two (or more) independent outcomes can occur simultaneously (i.e.
they are not mutually exclusive), the joint probability of the outcomes is expressed by the
product rule. The sum rule and product rule are also illustrated in Box 1.2.

BOX 1.2 PROBABILITY, SUM RULE, AND PRODUCT RULE

Probability
Probability represents the chance of a given event occurring. It is a measure of the uncer-
tainty and can be a number between 0 and 1. If probability is equal to 0 the event cannot
take place, in contrast if it is 1 the event must occur. Although there are different schools
of thought regarding the concept of probability, we prefer to define the probability as
belief in future events. For example, when we flip a coin, if we state that the probability
of observing heads (A) is 0.5 and the probability of observing tails (B) is 0.5, we believe
heads and tails have equal chance in the next flip. The probability of heads [P(A)] can be
calculated using:
P(A) = A/N
where N is the total number of outcomes (i.e. the number of times the coin is flipped).
In statistical terms, the total number of times the coin is flipped is equal to 1 and thus
the probability of heads [P(A)] is 0.5. Instead of flipping a coin, we may want to throw
a dice and estimate the probability of throwing a 5. In this case, the number of throws
is 1. If there is only one 5 on a dice and we are using a six-sided dice, the probability of
throwing a 5 is 0.167.
Sum rule
In the case of mutually exclusive events (e.g. tails or heads when the same coin is flipped
at the same time), the combined probabilities of the events can be calculated by summing
the individual probabilities of each event. This is known as the addition law of probabil-
ity, or the sum rule:
P(A or B) = P(A) + P(B) 
where A and B are two mutually exclusive events, and P represents the probability.
If we flip the same coin twice, the combined expected probability is:
P(heads or tails) = P(head) + P(tails) = 0.5 + 0.5 = 1
Product rule
When two (or more) independent events can occur simultaneously, meaning that they
are not mutually exclusive, the joint probability of the events is expressed by the product

2967_Ch01.indd 25 07/07/2015 10:27

 26 CHAPTER 1 Genetic Diversity

BOX 1.2 PROBABILITY, SUM RULE, AND PRODUCT RULE  (Continued)

rule. The product rule states that the joint probability of two independent events occur-
ring is the product of the individual probabilities:
P(A and B) = P(A) × P(B)
where A and B are two events, and P represents the probability.
If we flip two different coins, the expected probability of a head from coin 1 is 0.5 and
the probability of a head from the coin 2 is 0.5, the joint probability of two heads is:
P(two heads) = P(head coin 1) × P(head coin 2) = 0.5 × 0.5 = 0.25

1.5 Population Size and Structure

The term population has already been defined at the beginning of this chapter. Here,
we consider a population through two important parameters: population size and popu-
lation stratification. Populations are continually being modified by increase (births and
immigrations) and decrease (deaths and emigrations). Some geneticists therefore consider
a population to be best defined as the area within which individuals are likely to find a
mate. Sometimes, however, human populations may be geographically widespread and be
subdivided into local groups, called subpopulations.

Breeding population size is important in evolution

When considering population size in evolutionary terms, the relevant information con-
cerns the number of breeding individuals, which may be quite different from the total
number of individuals in the population. In some cases the breeding population may be
a small proportion of the total. In developed countries, the proportion of older people
is rapidly increasing due to improvements in healthcare. As a consequence of an expand-
ing aging population, overall population fertility inevitably declines and the proportion
of the population that can be counted as members of the breeding population decreases.
In addition, even if the size of a breeding population can be estimated with reasonable
accuracy, the breeding population number may not be indicative of the actual breeding
population size. For example, factors such as sex ratio of breeding individuals, social
status, and disease may influence their genetic contribution to the next generation. As
a result, the concept of effective population size, an ideal population of size N in which
all adults have an equal expectation of becoming parents, tends to be used. The effective
population size (usually indicated as Ne) is the number of individuals in a population
who contribute offspring to the next generation, and it can determine the amount of
genetic variation, genetic drift, and linkage disequilibrium in populations.

Genetic variation is not always uniform in a population

A population may have substructural differences in the genetic variation among its con-
stituent parts. Population substructure or stratification may be due to several different evo-
lutionary reasons. For example, a population may have localized subpopulations in which
there is genetic drift. Exchange of genotypes may not have equal probabilities throughout
a population or selection may have different effects in different parts of the population.

2967_Ch01.indd 26 07/07/2015 10:27

 The Mitochondrial Genome 27

Migrations from one population into another may also be responsible for stratification.
Thus, a population is considered stratified if:
• Genetic drift occurs in some, but not all, subpopulations.
• Migration does not happen uniformly throughout the population.
• Mating is not genetically random throughout the population.
All of the evolutionary factors that we have already discussed can contribute to the struc-
ture of a population. This structure affects the extent of genetic variation and the pattern
of distribution.

Wahlund’s principle
A population may appear to be homogeneous, but this may be a deception. This can lead
to false-positive associations, as we will see in Chapter 5. Subpopulations and population
stratification may not be obvious in studies of populations and population structure, and
as a result the study samples may include heterogeneous subsamples or clusters from the
study population. When data from these subpopulations are grouped together and differ-
ences in allele frequencies among them are inferred, a deficiency of heterozygotes and an
excess of homozygotes will be found, even if Hardy–Weinberg proportions exist within
each subsample. This effect is known as Wahlund’s principle or the Wahlund effect and
is one of the major problems in genetic association studies.

1.6 The Mitochondrial Genome

In humans and most other eukaryotes the mitochondria are large intercellular organelles
approximately 0.5–1 μm in diameter. Their main function is to generate energy in the
form of adenosine triphosphate (ATP). As a result, the mitochondria control and contrib-
ute to a range of cellular functions, including cell signaling, differentiation, and even cell
death. The mitochondria are composed of inner and outer membranes, and are packed
with multiple copies of their own DNA genome (mtDNA).
The mitochondrial genome has been extensively studied in relation to human evolution.
Generally, the mitochondrial genome is easier to analyze than the autosome by virtue of
the much shorter sequence (16.6 kb of circular DNA), greater abundance of mtDNA,
and the fact that there are only 37 genes encoded in the mtDNA genome. In addition,
because mtDNA is maternally inherited and does not recombine at meiosis, the mtDNA
genes are not reshuffled every generation through recombination (a process that tends to
obscure genetic relationships). Therefore, genetic relationships are more easily identified
in mtDNA. Usually, all the copies in an individual’s mitochondrial genome are identical
(i.e. individuals are homoplasmic), but occasional heteroplasmy is observed where there
is a mixture of two or more mitochondrial genotypes. In most individuals, though there
is no evidence of heteroplasmy, there is strong evidence that the mtDNA is undergoing
constant mutation. However, because these mutations are present at a low level they are
often not detected. When it comes to considering populations, however, there is often
considerable variation with high levels of polymorphism within and between populations.
The general low level of heteroplasmy in the population suggests that, at some point in the
germ line, the effective number of copies of the mitochondrial genome must have been
very small otherwise a greater level of diversity would be seen. More than 10,000 complete
mitochondrial genomes have now been sequenced (http://www.mitomap.org). These

2967_Ch01.indd 27 07/07/2015 10:27

 28 CHAPTER 1 Genetic Diversity

include mtDNA from a number of different populations and also from some ancient
individuals of whom the first three were sequenced in 2008, including one Neanderthal
and two Homo sapiens (the Neolithic Tyrolean Iceman and a Paleo Eskimo).
The substitution rate for the entire mitochondrial genome has been estimated as
1.665 × 10−8 (±1.479 × 10−9) substitutions per nucleotide per year or one mutation every
3624 years per nucleotide. Though the overall substitution rate is low, substitutions at
some positions occur more frequently than at others. These are referred to as hot-spot
positions and are mostly located in the control regions (positions 16,362, 16,311, 16,189,
16,129, 16,093, 195, 152, 150, and 146). The mutation rate has also been estimated for
both the control and coding regions, and this higher rate of substitution makes mtDNA
particularly valuable in studying relationships in recently diverged lineages.
When applied to modern populations, mtDNA studies clearly support the theory that
modern humans originated in Africa and spread from that continent approximately
56,000–73,000 years ago. This fits with the observation of greater genetic diversity in
African populations and can be observed through the construction of a phylogenetic tree
based on mtDNA variations from populations from all parts of the world. By applying
the molecular clock to the tree, it has been possible to demonstrate that the ancestral
mtDNA, i.e. that one from which all modern mitochondrial genomes are descended,
existed between 140,000 and 290,000 years ago. Phylogenetic reconstruction shows that
this mitochondrial genome was located in Africa and the person who possessed it must
have been African. Thus, we can conclude that the most recent common ancestor for
modern humans is African and because of the matrilineal inheritance of mtDNA, she has
been called the mitochondrial Eve (Figure 1.13). This theory also relies on the observation
that less variation in mtDNA occurs among humans than would be expected. Perhaps this
reflects the importance of mitochondrial gene function, which may drive conservation of
the mitochondrial genome or alternatively genetic variation may be reduced by genetic
drift, and thus a small population size 50,000 years ago could have the same effect as the
bottleneck from a single common ancestor 200,000 years ago.

~20,000 <8000
39,000– H, J Z
51,000 T, U, Uk, V
12,000– A, D 15,000
I, W, X R Y X
C, D, G A 15,000 A A
7,000– C, D
M1-M40 B
9,000
B
65,000– N F
70,000 M
~3000
L2 L3 M
B
L1
130,000– C, D
200,000 L0 A
B
M42 S P
Q
48,000

Figure 1.13: Mitochondrial genotypes and human migration. The figure illustrates human migration out of
Africa and subsequent migrations based on mtDNA genotypes. The “out of Africa” principle of human evolution
and the idea of the mitochondrial Eve are important in understanding human diversity. Based on this figure,
expansion began around 120,000–150,000 years ago in Africa and 56,000–73,000 years ago out of Africa, but
estimates vary and quotes depend on the hominid species being discussed. (From the MITOMAP database;
http://www.mitomap.org.)

2967_Ch01.indd 28 07/07/2015 10:27

 Gene Expression and Phenotype 29

1.7 Gene Expression and Phenotype

Genetic variation is manifested in the phenotype
The term phenotype is the result of the coordinated expression of approximately 20,000
genes that make up the human genome and the interaction with the environment. A pheno-
type can be continuous, e.g. eye or hair color, or discontinuous, e.g. phenotypes that relate
to health or physiology where there is complex interplay between alleles of different genes
and between the genotype and the environment. The balance of gene expression among
many different genes usually produces an individual with a normal (healthy) phenotype.

Phenotypes are influenced by the environment

A specific set of environmental circumstances can strongly influence the phenotype. In
order to better illustrate this concept we can consider a classical example from population
genetics. The Himalayan allele in rabbits produces dark fur on the nose, ears, and feet. The
dark pigment develops only when a rabbit is bred at a temperature of 25°C or lower; if a
Himalayan rabbit is bred at 30°C, no dark patches develop. Pigmentation is dependent on
the environment, which modifies the effect of possessing the Himalayan allele. In other
words, the rabbit may possess the allele, but possession of the allele is not sufficient for the
phenotype to be expressed. The phenotype is only manifested when the environmental
conditions are correct. The enzyme necessary for the production of dark pigment is inac-
tivated at higher temperatures. This pigment is found only in the extremities of the body.
The animal’s core body temperature is normally above 25°C, i.e. a temperature where the
enzyme is not functional.
Environmental factors also play an important role in the phenotypic expression of a number
of genetic diseases. Phenylketonuria (PKU) is caused by a gene (also called PKU) encoding
a dysfunctional phenylalanine hydroxylase enzyme (PAH) that does not convert phenylala-
nine in tyrosine, resulting in the accumulation of abnormal levels of phenylalanine that can
lead to brain damage in children. A simple environmental change in those homozygous for
the PKU allele, such as a low-phenylalanine diet, reduces the risk of brain injury.
These examples illustrate the point that genes and their products do not act in isolation;
rather, they frequently interact with other factors, including environmental factors.
In order to compare the impact of the environment versus the genome on the phenotype,
we need to construct a model that will allow us to breakdown the phenotype into genetic
and environmental components. We can accomplish this for a quantitative trait using
the equation:
Pij = Gi + Ej
In this equation Pij is the phenotype, Gi is the genetic contribution, and Ej is the envi-
ronmental component. Ej may be either positive or negative depending upon the effect
of the environment. Individuals with a particular genotype may do well in a specific envi-
ronment depending on the level of interaction with the environment. If such a specific
interaction occurs, then the basic model can be expanded to include a term for genotype–
environment interaction, resulting in:
Pij = Gi + Ej + GEij
In this equation GEij measures the interaction between genotype i and environment j.
The model could be further expanded by splitting the genetic component into additive or
dominant.

2967_Ch01.indd 29 07/07/2015 10:27

 30 CHAPTER 1 Genetic Diversity

1.8 Epigenetics
Epigenetics means above or in addition (“epi-”) to genetics and it may be typically defined
as the study of heritable changes in gene expression that are not due to changes in the
DNA sequence. Epigenetics can involve chemical modifications of DNA or proteins that
are closely associated with DNA (e.g. histones) and a prominent role for RNA is also
emerging. The structure of the chromosome in the eukaryotic cell is highly ordered and
it undergoes a process of compaction. To achieve these compact structures DNA is com-
bined with various proteins, and then coiled and super-coiled to form chromatin. The
basic unit of chromatin [the nucleosome core particle (NCP)] is composed of a 147-bp
DNA chain in a 1.7 left-handed super-helical turn around an eight sectioned octamer
composed of two copies each of four different histones (Figure 1.14). The fundamental
unit of this packaging is called the nucleosome. This involves a complex interaction with
histone proteins. This coiling to form chromatin and the interactions with histone pro-
teins are key elements in epigenetics.
DNA is also modified by biochemical processes such as methylation and two alleles
with the same sequence may have different states of methylation that confer a dif-
ferent phenotype. Though these changes do not alter the DNA sequence, they may
have major effects on the expression of the gene. Some of these changes are heritable,
though they do not affect the DNA structure. Methylation is an important factor for
post-translational modification of histones and subsequent formation of nucleosomes
for packing DNA in the nucleus. It is thought that methylation establishes epigenetic

Linker
DNA

Core DNA
H3 H4 1.7 turns
Nucleosome

5.5 nm
H1 H3 Linker
H2A DNA

H2B

11 nm

Figure 1.14: The NCP: the basic unit of chromatin. The figure shows the structure of the histone octamer
with the N-terminal tails and DNA wrapped around the histone structure. This core protein comprises four
different histones: H2A, H2B, H3, and H4. The NCP organizes 147 bp of DNA in a 1.7 left-handed helical coil.
Small DNA sections called linker DNA help to stabilize the structure by association with a linker histone H1. (From
Armstrong L [2014] Epigenetics. Garland Science.)

2967_Ch01.indd 30 07/07/2015 10:27

 Conclusions 31

inheritance as long as the maintenance methylase acts to restore the methylated state
after each cycle of replication. Thus, a methylated state can be perpetuated through an
indefinite series of somatic meiosis. Methylation is an important factor in histone for-
mation for packing DNA in the nucleus. A useful site for epigenetics can be found at
http://www.ncbi.nlm.nih.gov/epigenomics/.
Not all geneticists interested in epigenetics are focused on biochemical processes—some
are concerned with the wider meaning of the word epigenetics as “epi-,” i.e. outside genet-
ics, in more general terms. This group concerns themselves with the environment, diet,
lifestyle, and the interaction with our genes. There is significant evidence to link these
phenotypic characteristics with genetic variation, as noted in Section 1.7.

1.9 Genomic Imprinting

Genomic imprinting is also considered to be an epigenetic phenomenon, though to some
extent it is more complex than any of the above. In some families with autosomal domi-
nant disease the trait is only expressed if inherited from a parent of one particular sex.
This occurs despite the fact that the disease-causing mutation can be present in both
male and female parents. Examples of genetic imprinting are Beckwith–Wiedemann syn-
drome, Prader–Willi syndrome, Angelman syndrome, and Silver syndrome (discussed in
Section 2.3). Imprinted genes harbor regions known as imprint control elements that
act over long distances. The imprint control elements have been found to be restricted
to small areas known as imprinting centers. In the case of Prader–Willi syndrome and
Angelman syndrome, the imprinting centers have been identified at the 5′ end of the
SNURF–SNRPN gene. Both syndromes are associated with the loss of the same small
region on the long arm of chromosome 15 (see Chapter 2).
It is tempting to ask why genomic imprinting occurs. One possibility is that throughout
evolution there are different and conflicting evolutionary pressures acting on maternal and
paternal alleles for genes (such as IGF2, which affects fetal growth). From an evolutionary
standpoint, paternal alleles that maximize the survival of offspring are favored and because
low birth weight is strongly associated with infant mortality and adult health, infant size
is a major factor. Thus, it is to the advantage of the male parent to pass on alleles that pro-
mote maximum fetal growth of their offspring. In contrast, a maternal allele with more
limited fetal growth is favored from a maternal standpoint. The fetus takes nutrients from
the mother and unlimited fetal growth may limit her ability to reproduce in the future. In
addition, giving birth to very large babies is difficult and risky for the mother.

Conclusions
This chapter illustrates some of the basic science and statistical concepts that are a prereq-
uisite for understanding genetic variations in human populations. Human evolution has
created a diverse and complex gene pool, and a number of different interacting evolution-
ary forces have played a key role in the generation of this diversity. These include muta-
tion, migration, genetic drift, and the bottleneck and founder effects, to name but a few.
Considering genetic diversity is important in the study of complex diseases. The level of
genetic diversity in a population varies depending on the population size and structure.
Genetic diversity is not confined to the autosome, but is also found in the mtDNA.

2967_Ch01.indd 31 07/07/2015 10:27

 32 CHAPTER 1 Genetic Diversity

Other factors that are important in considering complex disease are gene expression and
phenotype. Epigenetics is important both in the narrow sense, where we consider such
things as methylation of genes, and in a broad sense, where we consider the impact of the
environment (e.g. nutrition) on disease.
Statistics plays an increasingly important and complex role in modern genetics. Though
there are many excellent statistical programs to use for analysis of data, all those with an
interest in this field are advised to have a basic understanding of the statistical concepts
that apply to studies of complex disease.
Diversity in the gene pool confers both advantages and disadvantages to individuals within
a population. Diversity though modification of common genes can lead to a variety of
diseases, such as those described by the classic genetic models: chromosomal, mitochon-
drial, and Mendelian traits. However, these are rare. Diversity also leads to increased risk
of more common diseases (genetically complex diseases), whereby the inheritance of an
allele confers an increased or reduced risk. This is important in evolution because a diverse
gene pool increases the likelihood that some of the population will survive even the most
severe disease.
The research that is being applied to complex disease will be increasingly applied in medical
practice, but there are also societal and ethical issues arising from our increased knowledge
of the genome and the increasing role it is beginning to play in modern medical practice.

2967_Ch01.indd 32 07/07/2015 10:27

 Further Reading 33

Further Reading
Books Ermini L, Wilson IJ, Goodship TH & Sheerin
Armstrong L (2014) Epigenetics. Garland NS (2012) Complement polymorphisms: geo-
Science. This book is a useful new guide to graphical distribution and relevance to disease.
epigenetics. Immunobiology 217:265–271.
Cavalli-Sforza LL & Bodmer WF (1971) The Ewing B & Green P (2000) Analysis of expressed
Genetics of Human Populations. W. H. Freeman. sequence tags indicates 35,000 human genes.
Nat Genet 25:232–234.
Collins F (2010) The Language of Life: DNA
and the Revolution in Personalised Medicine. Gilbert MT, Kivisild T, Grønnow B et  al.
Harper-Collins. (2008) Paleo-Eskimo mtDNA genome reveals
matrilineal discontinuity in Greenland. Science
Crawford MH (2007) Foundations of 320:1787–1789.
Anthropological Genetics. In Anthropological
Genetics: Theory, Methods and Applications Green RE, Malaspinas AS, Krause J et  al.
(Crawford MH ed.), pp 1–16. Cambridge (2008) A complete Neanderthal mitochondrial
University Press. genome sequence determined by high-through-
put sequencing. Cell 134:416–426. This is a
Darwin C (1859) The Origin of Species: By fascinating report of mtDNA sequencing in an
Means of Natural Selection or the Preservation ancient mtDNA sample.
of Favoured Races in the Struggle for Life. J
Murray. This is perhaps the most important Hales CN & Barker DJP (1992) Type 2 (non-
book in the history of genetics. insulin dependent) diabetes mellitus: the
thrifty phenotype hypothesis. Diabetologica
Hedrick PW (2011) Genetics of Populations, 35:595–601. This is an original paper that
4th ed. Jones & Bartlett. contradicts Neel’s thrifty gene hypothesis; see
Jones S (2000) The Language of the Genes, 2nd Neel (1962).
ed. Harper Collins. Hardy GH (1908) Mendelian proportions in
Lewis R (2011) Human Genetics – The Basics. a mixed population. Science 28:49–50. This
Routledge. An alternative very short textbook is the original report of the Hardy–Weinberg
on genetics that covers the basics quite well. Principal.
This is a good source for students wishing to
Mayo O (2008) A century of Hardy–Weinberg
extend their knowledge to include more on
equilibrium. Twin Res Hum Genet 11:249–256.
classical genetics in a short time.
This is a review of this difficult concept, and
Strachan T & Read AP (2011) Human is of particular use for students and research-
Molecular Genetics, 4th ed. Garland Science. ers needing to understand the principal in more
This is an excellent basic genetics textbook that detail.
goes into more depth on many of issues in this
Mueller JC (2004) Linkage disequilibrium for
chapter of this book. Chapters 2 and 3 and also
different scales and applications. Brief Bioinform
chapters 15 and 16 of HMG are particularly
5:355–364.
useful.
Nachman MW & Crowell SL (2000) Estimate
of the mutation rate per nucleotide in humans.
Articles Genetics 156:297–304.
Barreiro LB, Laval G, Quach H et  al. (2008) Neel JV (1962) Diabetes mellitus: a “thrifty”
Natural selection has driven population dif- genotype rendered detrimental by “progress”?
ferentiation in modern humans. Nat Genet Am J Hum Genet 14:353–362. This is an origi-
40:340–345. nal paper with an interesting original hypoth-
Cann RL, Stoneking M & Wilson AC (1987) esis. The hypothesis is contested by Hales and
Mitochondrial DNA and human evolution. Barker (1992).
Nature 325:31–36. Reich DE, Cargill M, Bolk S et  al. (2001)
Ermini L, Olivieri C, Rizzi E et  al. Linkage disequilibrium in the human genome.
(2008) Complete mitochondrial genome Nature 411:199–204.
sequence of the Tyrolean Iceman. Curr Biol Soares P, Ermini L, Thomson N et  al. (2009)
18:1687–1693. Correcting for purifying selection: an improved

2967_Ch01.indd 33 07/07/2015 10:27

 34 CHAPTER 1 Genetic Diversity

human mitochondrial molecular clock. Am J http://www.ncbi.nlm.nih.gov/SNP

Hum Genet 84:740–759. These two sites are essential for updates on
The International HapMap Consortium (2007) SNPs.
A second generation human haplotype map of http://www.ncbi.nlm.nih.gov/mapview
over 3.1 million SNPs. Nature 449:851–861. This is an excellent resource for those wishing
Tishkoff SA, Reed FA, Friedlaender FR et  al. to stay up-to-date on the genome.
(2009) The genetic structure and history http://www.ncbi.nlm.nih.gov/projects/SNP/
of Africans and African Americans. Science snp_summary.cgi
324:1035–1044. This site now contains more information about
Wahlund S (1928) Zusammensetzung von more than 38 million of validated reference
Population und Korrelationserscheinung vom SNPs clusters in the genome.
Standpunkt der Vererbungslehre aus betrachtet. http://www.ncbi.nlm.nih.gov/epigenomics
Hereditas 11:65–106. This is an original report This site is another excellent source for infor-
of the Wahlund principal and reminds the mation from NIH.
reader that not all major papers are published
in English. http://www.ncbi.nlm.nih.gov/omim
This is an essential site for updates and informa-
Weber JL & Broman KW (2001) Genotyping tion on genetic disease of all types.
for human whole-genome scans: past, present,
and future. Adv Genet 42:77–96 http://www.genet.sickkids.on.ca/StatisticsPage.
html
Wright S (1931) Evolution in Mendelian popu- This is an excellent place to search for updates
lations. Genetics 16:97–159. on cystic fibrosis and the CFTR gene.
http://www.mitomap.org
This is a great resource for those wishing to
Online sources investigate the mitochondrial genome further.
http://www.ncbi.nlm.nih.gov/projects/genome/ http://www.ncbi.nlm.nih.gov/omim/
guide/human/index.shtml On-line inheritance in man (OMIM) is an
This provides a very useful guide for those essential site for updates and information on
studying the human genome. genetic disease of all types.

2967_Ch01.indd 34 07/07/2015 10:27

 CHAPTER
2
Defining Complex Disease

Until recently, healthcare professionals considered medical genetics to be the province of

specialists seeking to understand rare cases of Mendelian disorders, birth defects, and
chromosomal abnormalities. Complex diseases were more or less excluded from con-
sideration as affected (informative) families were comparatively rare and genetic linkage
was difficult to prove even if families were available for testing. However, following the
publication of the Human Genome Map (HGM) and the haplotype map (HapMap)
as well as developments in genetic technology, the focus for research in medical genetics
has changed. Collins and McKusick writing in 2001 even went so far as to suggest that
“except for some cases of trauma, it is fair to say that virtually every human illness has a
hereditary component.”
However, the term hereditary (or heritability) does not necessarily imply genetic. For
example, heritable traits may be related to diet and alcohol consumption which could
impact on families and populations, causing changes in the frequency of a disease or trait
that may be falsely attributed to genetics. Heritability considered in this way is complex.
It is necessary to distinguish between heritability in the narrow sense and heritability in
the broad sense. It is also important to define the components, such as additive dominant
alleles, and epistasis. Epistasis involves gene–gene interaction. It occurs when the expres-
sion of a gene requires the involvement of one or other modifier genes. Epistasis is also
important in epigenetics—a term that refers to things outside of genetics and is discussed
in Chapter 1 and also later in this book.
However, for the purpose of this book we are concerned with the genetics of disease and
in that sense heritability can be defined as the proportion of variation in a phenotype in
a population that can be accounted for by our genes. Some authors have suggested we

2967_Ch02.indd 35 07/07/2015 10:30

 36 CHAPTER 2 Defining Complex Disease

overestimate the impact of our genes on disease. In the current millennium it is hard to
disagree with the concept that our genes play a major role in disease and disease suscepti-
bility. Even if not all of assigned heritability is genetic, it is likely that a great deal of this
heritability will be genetic and, since most diseases are not Mendelian or chromosomal, a
very significant proportion will fit into the classification of genetically complex disease. In
fact, chromosomal abnormalities account for less than 1% and Mendelian disease for no
more than 4% of human disease. Does this mean that 95% of human diseases are geneti-
cally complex diseases? Geneticists do not agree on how many diseases may have a genetic
component. However, all of the evidence suggests it may be a significant number of differ-
ent diseases, and these include some infectious, immune and metabolic diseases as well as
some cancers and some forms of toxicity.
In this chapter, we will consider the definition of genetically complex disease. We will com-
pare complex diseases to diseases that arise either as a result of major chromosomal abnor-
malities, single inherited genetic mutations (Mendelian diseases), or genetic abnormalities
in the mitochondrial genome. In addition, we will consider in detail three different models
for genetically complex disease using key examples to illustrate how each model is different.

2.1 Definition of a Genetically Complex

Disease
When constructing a definition it is often useful to consider what is not included. Complex
diseases are not simple. They do not lend themselves to simple analysis and do not con-
form to the expected patterns of inheritance that define Mendelian diseases, i.e. autosomal
recessive, autosomal dominant, and sex-linked. Complex diseases do not arise from major
chromosomal abnormalities. The terms characteristic or trait often get used interchange-
ably with the term disease and though this is not strictly correct, this has become quite
common. One of the reasons for this is defining disease is itself difficult. Whatever termi-
nology is used, complex traits or diseases are perhaps best defined in the words of Haines
and Pericak-Vance (1998) as those where “alterations in more than one gene that alone
or in concert either increase or decrease the risk of developing a trait.” We would however
modify this to say allele rather than gene, and insert disease in place of trait frequently
(but not in all cases) as we are mostly interested in diseases and the characteristic clinical
variations seen in different diseases. Thus, we will use the term disease as much as possible
in the book and not the term trait.
The problem with this definition is that it not only applies to complex disease, but can
also be applied to some Mendelian disorders such as ataxia and retinitis pigmentosa that
may be caused by more than one gene. In these cases, the genetic mutation acts in a simple
Mendelian fashion (i.e. alone) and they are not considered complex diseases. It is also the
case that some traits and diseases are known to have both pure Mendelian families and
pure sporadic complex cases.

To fully understand complex disease it is important to deconstruct

this definition
Based on the definition of Haines and Pericak-Vance, three things are immediately appar-
ent when comparing genetically complex diseases with classical Mendelian and chromo-
somal diseases. Complex diseases show complex patterns of inheritance and we are dealing
with alleles as determinants of risk rather than as absolute causes of disease. Possession of

2967_Ch02.indd 36 07/07/2015 10:30

 Definition of a Genetically Complex Disease 37

a particular risk allele may affect disease outcome or phenotype, e.g. signs, symptoms, or
response to treatment. Genetic variations and the terminology that is applied to them is
discussed at length in Chapter 1 and also in the Glossary at the end of the book, which
lists some of the appropriate terminology that applies to this area of genetic research.

Complex diseases show complex patterns of inheritance

In complex disease there may be a single risk allele or several risk alleles. In their defini-
tion, Haines and Pericak-Vance say “more than one”, we would say one or more. They
also say that these “genes”, whereas we would use the term alleles, may act “either alone or
in concert” to “increase or decrease the risk of developing a trait.” We would replace the
word trait with disease in most circumstances, but the definition remains solid and the
implications of this definition are the same. This implies that there may be one or more
than one risk allele and the patterns of inheritance seen in complex disease are likely to be
complex (Figure 2.1). This suggests that there is heterogeneity in complex disease whereby
different individuals with the same disease may inherit different alleles, i.e. a genetic port-
folio, associated with an increased or reduced risk of the disease. By inheriting different
risk alleles there is genetic heterogeneity for a given disease. At this stage it is reasonable
to say we understand little about the interactions that occur within many of the portfolios
that have been investigated, though some understanding is beginning to emerge. It is also
true to say that the interaction of the different risk alleles is likely to be complex and is
currently poorly understood for a number of reasons dealt with at the end of this book.
Studies so far have shown that there is considerable genetic sharing of risk alleles between
diseases whereby one or a group of alleles may predispose to more than one disease. Some
consider this to be pleiotrophy, but others consider this to be clinical heterogeneity.
Strictly speaking, pleiotrophy refers to mutations that have multiple effects; in these cases,
the final clinical outcome may be determined by the complement of risk alleles inherited
and by the environmental risk. Technically, pleiotrophy and clinical heterogeneity are dis-
tinct entities, but pleiotrophic alleles may be involved in clinically heterogeneous disease.
Overall, this is why complex diseases do not comply with simple Mendelian patterns of
inheritance. However, within complex disease there are recognizable patterns. These are
best defined as monogenic, oligogenic, and polygenic, depending on the number of risk
alleles. Despite this, there is no simple mathematical test to determine which group each
disease belongs to. The terms simply translate as single (mono), several (oligo), and many
(poly). Monogenic disease is relatively easy to define in Mendelian disease, but rare in
complex disease, whereas oligogenic and polygenic, though not as easily defined, are more
common. The idea that monogenetic disease can be genetically complex is very difficult
for some. However, where there is sporadic disease and a single identified common risk
allele, there is no alternative explanation at present. Future studies may prove this to be
naive, but we must allow for this possibility at present with the hope that a better explana-
tion awaits for these rare monogenic cases.

Complex diseases involve genetic variation that increases or decreases risk

Perhaps the most important detail in this definition is the use of the word “risk.” In
complex disease, inherited genetic variation gives rise to an increased or reduced risk of
a disease. This is the crucial concept of this book and much of modern medical genet-
ics. Risk alleles are those that have been found to be present at a statistically significant
increased or reduced frequency in a subpopulation with a disease or trait compared with
the healthy or normal population, either through linkage or genetic association analysis.
To date, most studies have concentrated on identifying single nucleotide polymorphisms
(SNPs) as markers for risk alleles in complex disease. The choice of SNPs is based on their

2967_Ch02.indd 37 07/07/2015 10:30

 38 CHAPTER 2 Defining Complex Disease

One allele alone leading to a disease, but the environment may play a role in
determining the phenotype so there can still be more than one phenotype.

One allele Environment

One or more alleles acting independently – genetic heterogeneity leading

to a single disease but the environment may play a role and so there can
still be more than one phenotype.

One allele

One allele Environment

One allele

One allele One allele One allele

More than one gene acting in concert leading to a

single disease but because the environment can
play a role there may be more than one phenotype.

Environment

One or more alleles leading to more

than one phenotype (pleiotrophy).

One or
more alleles

Figure 2.1: One or more alleles acting alone or in concert illustrates the key concept in the genetics of
complex disease. The first three patterns illustrate one allele alone, more than one allele acting independently,
and several alleles acting in concert. In each case the final outcome is complicated because the phenotype
is determined by both the genotype and the environment in most cases, and so there can be more than one
phenotype as illustrated by the different facial expressions. The final example illustrates several different

2967_Ch02.indd 38 07/07/2015 10:30

 Definition of a Genetically Complex Disease 39

frequency in the population. Most studies chose SNPs with a frequency of no less than
5% for the rarest alleles. However, as study groups are getting larger, the thresholds for the
lowest frequency SNPs are being set at lower levels (i.e. 1% or less). This is because greater
sample numbers are now available for testing which increases the statistical power of the
study enabling less common SNPs to be included. Assessing less common SNPs may pick
up hitherto missed genetic associations.
Other forms of genetic variation have been investigated including insertions, deletions,
and copy number variations (CNVs). Only a few complex diseases arising from insertions
and deletions have been identified so far. One major example where CNVs have been
reported is in autism.
It should be noted that statistical significance is based on a threshold model whereby the
statistical value obtained is expected to equal or exceed the threshold set prior to any inves-
tigation. This is discussed at greater length in Chapter 5.
The inheritance of a specific risk allele or group of alleles does not necessarily cause the
disease. Inheriting the risk allele or alleles simply increases the likelihood (risk) of disease.
We may say that inheritance of the risk allele is neither necessary nor sufficient for the dis-
ease to occur. Note also that the definition used by Haines and Pericak-Vance includes the
word “decrease.” This indicates the possibility that inheritance of some alleles may protect
from disease. Considering genetic protection is as important as considering susceptibil-
ity, especially when we consider infectious disease (see Chapter 7). This protective effect is
seen quite frequently with different human leukocyte antigen (HLA) alleles and haplo-
types. Many autoimmune diseases have genetic associations with groups of HLA alleles or
haplotypes, some of which increase disease susceptibility, while others reduce susceptibil-
ity. Comparison of the susceptibility versus the protective alleles has enabled investigation
of the molecular mechanisms that underpin these HLA associations (see Chapter 6). This
latter development is one of the keys in moving from simple genetic investigation to
understanding basic disease pathology.
It is also essential to dispel any idea of good and bad alleles. This idea oversimplifies the
concept of complex disease. Many of the known risk alleles that have been identified as
potent inducers of disease for some clinical syndromes have also been identified as protec-
tive for other diseases. This is illustrated by the example of the impact of the C-chemokine
receptor 5 Δ32 deletion (CCR5-Δ32), which protects from human immunodeficiency
virus (HIV) infection, but predisposes to infection with West Nile virus (see Chapter 7).
Genetic variation also affects the outcome of disease
Finally, genetic variation does not only determine susceptibility and resistance to dis-
ease per se, but also determines various outcomes following disease onset. Therefore, it
is appropriate to use the term “trait” in place of the word “disease” in this definition. By
restricting the definition to disease we narrow the field, and exclude the possibilities that
common genetic variations influence disease severity and outcome both before and fol-
lowing treatment. However, it is quite clear that morbidity, mortality, and response to
certain commonly used pharmacological agents (including adverse drug reactions) are all

Figure 2.1: (Continued) phenotypes resulting from the same risk portfolio. Phenotypes are identified by different
facial expressions: smile, open mouth, and star decoration on temple (in the last case only). Variable phenotype
and pleiotrophy are similar. Note that some authors use alleles and genes as interchangeable terms. However,
when talking about variation of a gene at a locus, the correct term is allele. Phenotype is a complex issue because
although each of these models indicates a single disease, the clinical phenotype is the product of both genetic
and environmental factors. Therefore, the phenotype can vary even with a single allele.

2967_Ch02.indd 39 07/07/2015 10:30

 40 CHAPTER 2 Defining Complex Disease

influenced by common genetic variation. The use of the word trait as opposed to disease is
important throughout the study of complex disease, but it is particularly important when
considering the role of genetic variation in infectious diseases (see Chapter 7) and also in
pharmacogenetics (see Chapter 8).
The take-home message from all of the points above is that the simple definition of a
genetically complex disease as a disease or trait where “alterations in more than one gene that
acting alone or in concert either increase or reduce the risk” needs to be deconstructed and
considered word for word. It is essential to remember we have, in the past, been dealing
with alleles that for the most part are common. However, more recently we have started
to investigate risk alleles with low frequencies. Developments in technology, statistical
analysis, and study design have all played important roles in focusing our attention on
these less common alleles. Finally, before we move on to concentrate on complex dis-
ease, let us first consider genetic diseases that are not classified as genetically complex
diseases: chromosomal, Mendelian, and mitochondrial diseases.

2.2 Chromosomal Diseases

Chromosomal and Mendelian diseases include some of the cruelest of all human diseases,
with early onset, high levels of morbidity, and high levels of mortality. The ease with
which the heritable nature of these diseases can be identified has made them perfect (if
simple) models upon which to develop genetic theory and practice, some of which is now
being applied to the study of complex diseases. Chromosomal abnormalities are fortu-
nately quite rare. They are visible (often very large) alterations of the chromosome causing
abnormalities in specific chromosomal regions. Early cytogenetic studies were restricted
to detecting changes of greater than 4 Mb of DNA, but more recent developments such
as fluorescence in situ hybridization (FISH) have enabled much smaller changes to be
identified.
The most common causative mechanisms for chromosomal abnormalities are mis­
repair of broken chromosomes or incorrect segregation during mitosis and meiosis.
Chromosomal abnormalities can be classified as either constitutional chromosomal
abnormalities (occurring in all cells) or somatic chromosomal abnormalities (occur-
ring in a selected group of cells or tissues only) and are further classified as numerical or
structural abnormalities.

Changes in chromosome number cause serious genetic diseases

Numerical abnormalities are referred to using different forms of the word “ploid,” which
means number. There are several forms of ploidy or changes in the chromosome num-
ber. Most humans are diploid with two copies of the entire genome and our gametes are
haploid. Therefore, most individuals inherit one copy of the genome from each parent,
and thus the normal genome is made up of 22 pairs of chromosomes (autosomes) and
either XX (females) or XY (males). However, some individuals inherit either additional
copies or fewer copies of the whole genome (polyploidy), or they inherit additional or
fewer copies of a specific chromosome or chromosomes (aneuploidy).
Polyploidy
Polyploidy involves the transmission from parent to offspring of multiple copies of the
genome. It can be subclassified depending on the number of copies involved. These sub-
groups include tetraploidy, triploidy (Figure 2.2), monosomy, and mosaicism.

2967_Ch02.indd 40 07/07/2015 10:30

 Chromosomal Diseases 41

Triploidy (1) (2) (3)

n n
2n
n
2n
n
n

Tetraploidy (4)

4n
2n

DNA is duplicated but

the cell fails to divide

Figure 2.2: Numerical chromosomal abnormalities. The top of this figure illustrates three different models of
triploidy: (1) dispermy, (2) diploid egg, and (3) diploid sperm. At the bottom of this figure, (4) illustrates tetraploidy
where DNA is duplicated, but there is no cell division. (Adapted from Strachan T & Read A [2011] Human
Molecular Genetics, 4th ed. Garland Science.)

Tetraploidy occurs when offspring inherit four copies of the whole genome. Tetraploidy is
rare and is most commonly caused by the failure of the cell to divide after the first DNA
duplication. Tetraploidy is not compatible with life.
Triploidy can occur through several mechanisms, the most common of which is two
sperm fertilizing a single egg (dispermy) and, as with tetraploidy, triploidy is not compat-
ible with life.
Aneuploidy
Individuals with aneuploidy have one or more chromosomes present either with an extra
copy or with a missing copy. Aneuploidy involving the autosomes is usually lethal, but
it can be compatible with life. Key examples include: trisomy 13 associated with Patau
syndrome; trisomy 18 associated with Edward’s syndrome and trisomy 21 associated with
Down syndrome. Aneuploidy associated with additional copies of the sex chromosomes
(e.g. XXX, XXY, XYY) is mostly associated with a normal lifespan and relatively few minor
clinical problems. By contrast, aneuploidy associated with missing sex chromosomes is fre-
quently lethal, though monosomy X (associated with Turner’s syndrome) is an exception.
Ninety-nine percent of all cases involving missing sex chromosomes abort spontaneously
(Table 2.1).

Changes in chromosome structure can cause serious illness

Structural abnormalities include deletions, inversions, duplications, insertions, formation
of ring chromosomes, and translocations. Inversions are subclassified as paracentric (not
involving the centromere) and pericentric (involving the centromere) (Figure 2.3). All of
these abnormalities may result from the misrepair of chromosome breaks or from failures
in recombination. Chromosomal breaks may occur naturally as a result of recombination

2967_Ch02.indd 41 07/07/2015 10:30

 42 CHAPTER 2 Defining Complex Disease

Table 2.1 Terminology applied to numerical and structural abnormalities.

Numeric abnormalities
Polyploidy Tetraploidy Four copies of whole genome
Triploidy Three copies of whole genome
Monosomy One copy of whole genome
Mosaicism Variable number of copies in different cells
Aneuploidy Trisomy Three copies of one or more chromosomes
Monosomy Only one copy of one or more chromosomes
Structural abnormalities
Deletions Unbalanced abnormalities
Inversions Paracentric or pericentric balanced abnormalities
Duplications Unbalanced abnormalities
Insertions Unbalanced abnormalities
Ring formation
Translocations Balanced abnormalities

and can occur at different phases of the cell cycle. There are natural mechanisms to prevent
cells with unrepaired chromosomes from entering mitosis. Structural abnormalities occur
when breaks are repaired incorrectly. For example, the broken ends of the chromosome
are sometimes joined incorrectly and this can lead to chromosomes without a centromere
(acentric) or with two centromeres (dicentric). Such abnormal chromosomes will not
segregate stably in mitosis and are eliminated. However, chromosomes with a single cen-
tromere carrying structural abnormalities can be propagated through mitosis.

Balanced abnormalities
When structural abnormalities are associated with no overall gain or loss of chromo-
somal material they are said to be balanced. Provided the balanced structural abnor-
malities do not result in disruption of the function of a gene through interrupted
expression, control, or activation of the gene, they are unlikely to have an adverse
effect on the phenotype. Balanced abnormalities are those involving inversions and
translocations.

Unbalanced abnormalities
When structural abnormalities are associated with gains or losses of chromosomal mate-
rial they are said to be unbalanced. In contrast to balanced abnormalities, unbalanced
chromosomal abnormalities are more likely to give rise to problems in meiosis, and they
result from deletions, insertions, and duplications. As a consequence, these abnormalities
are likely to have adverse effects on the phenotype.
As stated above, chromosomal abnormalities are rare. It is important for the reader to
note that, unlike complex disease, all carriers of these abnormalities are affected and
symptoms most often occur from birth. However, these genetic illnesses are not the
subject of this book.

2967_Ch02.indd 42 07/07/2015 10:30

 Mendelian Diseases 43

(a) Two breaks in the same arm (b) Two breaks in two different arms

a a

b b

c c

d d

e e

f f

g g

Join broken
Deletion Inversion Inversion
ends

c
a a a b
b b f
d Fusion point
c c e f
e
d d d Ring chromosome
f c
g
e b

g g

Interstitial Paracentric Pericentric

deletion inversion inversion

Figure 2.3: Outcomes after incorrect repair of two breaks on a chromosome. (a) Incorrect repair of two
breaks in the same arm of the chromosome can lead to either a parametric inversion (such as e and f inversion)
or interstitial deletion (i.e. deletion of e and f) of the fragments. In the former case, the inversion is paracentric
as it does not involve the centromere. (b) Incorrect repair occurs following breaks in two different arms of the
same chromosome. Here, the fragments coded from b–f are inverted, and in this case the inversion involves the
centromere and is referred to as a pericentric inversion. As the fragment involves the centromere, the fragment
could also rejoin to form a stable ring chromosome. In this latter model, the fragments b–f are included in the ring
structure, but fragments a and g are missing. The banding patterns represent the karyotype represented in F
1.1. (From Strachan T, Goodship J & Chinnery P [2014] Genetics and Genomics in Medicine. Garland Science.)

2.3 Mendelian Diseases

Mendelian diseases involve a single gene and show simple patterns
of inheritance
The rediscovery of the papers of Gregor Mendel (Box 2.1) in 1900 and the different
patterns of genetic inheritance that the experiments illustrated gave rise to what are

2967_Ch02.indd 43 07/07/2015 10:30

 44 CHAPTER 2 Defining Complex Disease

BOX 2.1 A SHORT HISTORY OF GREGOR MENDEL

Gregor Mendel’s research, or more specifically the rediscovery of his papers in 1900, is
rightly credited as the time at which modern genetics was born. However, people were
aware of the passage of traits from generation to generation long before the discovery
of the papers reporting Mendel’s experiments with pea plants. Indeed, the bible gives
the example of Jacob’s sheep. Gregor Mendel was born Johann Mendel in a village in
Moravia (later Czechoslovakia) in 1822; he entered an Augustinian Monastery in 1843.
He was a gifted teacher, though he failed his teaching certificate. Mendel’s experiments
with peas enabled him to identify different patterns of inheritance for specific selected
traits. The terminology applied to his work post-dates his discoveries because despite the
ground-breaking significance of his work initial interest in his papers rapidly faded, and
disappointed and burdened with administration he retired into monastic life and died in
1884. Nevertheless, he is and was the founding father of modern genetics, and we refer
to his name every time we look at a disease with a known pattern of inheritance.

called Mendelian patterns of inheritance, and today these patterns are used in classify-
ing genetic diseases and traits. Over 4609 traits are currently identified on the OMIM
(Online Mendelian Inheritance in Man) database for which the molecular basis of the risk
gene is known (http://www.ncbi.nlm.nih.gov/omim). Over 12,000 genes are also listed
on OMIM. Not all of these traits and genes are for Mendelian diseases, many are for
complex diseases, but those which are Mendelian disorders all involve a single gene (i.e.
they are monogenic) and all conform to one of five readily identifiable patterns: autosomal
dominant, autosomal recessive, X-linked dominant, X-linked recessive, and Y-linked
(illustrated by the pedigrees in Figure 2.4). Sex-linked disorders may also be referred to as
gonosomal as opposed to autosomal.
In addition, occasionally geneticists will use terms such as semi-dominant to describe
a family where heterozygotes have a phenotype. In many cases it may not be clear if the
heterozygote has a less severe phenotype than the homozygote, as would be expected in
a truly semi-dominant disease. However, for the most part, Mendelian diseases conform
to the expected patterns. Mendelian diseases can be described as those where a particular
genotype at one locus is both necessary and sufficient for the character to be expressed.
Mendelian genotypes have variable phenotypes
It is important to understand that biological systems are complex and interactive.
Therefore, though these Mendelian diseases are considered simple in as much as they
involve a single dysfunctional genetic character at a single gene locus, the biological
character affected by this disability is likely to be programmed by a large number of
interacting genes and also by environmental factors. This complexity may explain the dif-
ferent levels of phenotypic variation in individuals and families with the same Mendelian
genotype. Expression of the trait even in Mendelian disease is not always absolute. There
are variable levels of penetrance whereby possession of the mutant gene does not give
rise to the disease.
An extreme example of this incomplete penetrance is found in the autosomal recessive
disease hemochromatosis. This disease is characterized (as the name suggests) by iron over-
load in the liver and other tissues. The gene responsible for the disease has been identified
and is referred to as HFE. The causative gene is located on chromosome 6p21.3 at the

2967_Ch02.indd 44 07/07/2015 10:30

 Mendelian Diseases 45

(a)

Male Female
Affected
Carrier
Unaffected

(b) (c)

(d) (e)

Figure 2.4: Genetic pedigrees in Mendelian diseases. The five different patterns displayed represent three
generations all modeled on the same family structure. (a) Autosomal dominant: both males and females are
affected (marked in black) and the trait does not skip generations. (b) Autosomal recessive: both males and females
are affected, but here not all generations are affected (e.g. the grandparents are unaffected carriers), carriers are
marked in gray. (c) X-linked recessive: as with autosomal recessive not present in all generations; mostly males are
affected as they carry only one copy of X, but females can be affected. (d) Y-linked: males only are affected and as
they inherit a single Y chromosome only, all Y-linked Mendelian disease assumes a dominant pattern. (e) X-linked
dominant: passed from fathers to daughters in the first generation (males receive the Y chromosome from the
affected father). When these daughters have offspring, the trait is passed to both sons and daughters, but equally
sons as daughters may inherit the wild-type gene from their mothers and not develop the trait.

extreme telomeric end of the major histocompatibility complex (MHC) and was for a
short while referred to as HLA-H. A number of studies indicate that the penetrance in
those with two copies of the HFE mutation can be extremely low, with the majority of
homozygous carriers being asymptomatic possibly unaffected. The pattern of homozy-
gosity for the disease causing mutation with extremely low levels of penetrance (disease
expression), is unusual. Of course in a late onset disease such as this one, we cannot be
certain that some of the homozygotes will not develop the disease at a later date.

Penetrance is an important difference between Mendelian and

complex diseases
In Mendelian traits, we expect the trait to be expressed when one or both copies of the
disease gene (depending on a recessive or dominant pattern) are dysfunctional. The exam-
ple of hemochromatosis above is an extreme exception. For the most part, penetrance
in Mendelian disease is high and in many, but not all, cases penetrance is 100%. This

2967_Ch02.indd 45 07/07/2015 10:30

 46 CHAPTER 2 Defining Complex Disease

is in stark contrast to complex disease where penetrance is mostly low. In many cases of
complex disease the majority of those who inherit the risk allele do not express the disease
phenotype and some members of the population develop the disease in the absence of the
risk allele. In complex disease the presence of an allele confers risk, but the risk allele is
neither necessary nor sufficient for the disease to occur. Examples of this are to be found
throughout this book. One key example is ankylosing spondylitis—a degenerative disease
of the lower spine that occurs in young adults and is associated with the HLA-B*27 family
of HLA alleles. In some populations it has been found that up to 94% of patients with
this illness have HLA-B*27. The odds ratio (OR) (a crude measure of the population
risk; see Box 2.2) for ankylosing spondylitis in those with HLA-B*27, is estimated to be
as high as 171 times greater compared with those without HLA-B*27. From a personal
point of view this increased risk seems to be high until we consider that 6% of ankylosing
spondylitis patients do not have HLA-B*27 and that on average only one out of every 25
HLA-B*27-positive individuals will develop the disease (Figure 2.5). Interestingly, recent
genome-wide association studies (GWAS) have identified a range of other risk alleles for
ankylosing spondylitis outside the MHC (see Chapter 3).
The example above is an illustration of incomplete penetrance at work in complex
diseases. Penetrance is the key to understanding the difference between complex and
Mendelian disease. Whereas in Mendelian disease the presence of the allele (whether
a mutation, polymorphism, or deletion) is considered to be necessary and sufficient to
cause the disease, this is not true for complex disease. This difference is at the heart of
understanding the genetics of complex disease and the potential applications of this
evolving branch of biomedical science.

BOX 2.2 A SHORT DISCUSSION ON HOW TO CALCULATE THE ODDS

RATIO IN A CASE CONTROL ASSOCIATION STUDY

The odds ratio (OR) is a frequently used simple statistical measure of the likelihood
of risk based on the ratio of an allele in healthy controls compared with affected cases.
It is usually calculated from the figures entered in the Punnett Square (Table 1.2) as
A × D/B × C. To understand this test, consider a population in which we test a single
gene with two alleles (α and β), in 100 cases and in 100 healthy controls. First remember
that in a diploid species we have two copies of each chromosome, therefore in a group
of 100 individuals there are 200 alleles for this single gene. We perform a genetic test on
this population and the results of our testing show that the frequency of the α allele is
higher in patients than in healthy controls, e.g. there are 150 α alleles in the patients and
only 100 α alleles in the healthy controls. Because we already know the total number of
individuals in each group is 100, we can assume that the frequencies of the β alleles for
the two groups are 50 for patients and 100 for healthy controls. The OR is calculated
as above with A × D/B × C, where the A box value is the number of α alleles in patients
and the D box value is the number of β alleles in controls, while the B box value is the
number of α alleles in healthy controls and the C box value is the number of β alleles
in patients. Numerically, A × D/B × C = 150 × 100/100 × 50; this can be simplified by
deleting the common values on either side of the equation, i.e. 100, to give 150/50 and
also removing the 0 on either side to give a simple calculation of 15/5 from which the
OR value of 3 is derived.

2967_Ch02.indd 46 07/07/2015 10:30

 Mendelian Diseases 47

HLA-B*27+

AS-positive

General population

Figure 2.5: HLA-B*27 and ankylosing spondylitis. The figure illustrates incomplete penetrance for HLA-B*27
in ankylosing spondylitis (AS). Most B*27-positives are ankylosing spondylitis-negative even though most
ankylosing spondylitis-positives have B*27.

Some diseases have both Mendelian and complex characteristics

One complication not considered above can be seen when Mendelian disease becomes
mixed with complex disease. Comparing early- and late-onset Alzheimer’s disease
(EOAD and LOAD) illustrates this problem. EOAD mostly conforms to a Mendelian
pattern of inheritance and is marked by clear familial inheritance within the bounds
of the proscribed Mendelian patterns. LOAD is mostly marked by the sporadic
occurrence of the disease in (mostly) hitherto unaffected families. However, there
are individual family cases that map to a gray area in between, with some cases being
Mendelian and some cases being complex. In this situation it is important to use all
of the available information to identify the genotype and phenotype of the individual
cases in the family concerned.

Modifier genes may also confuse the picture

Another problem can arise when a Mendelian disease with a clear pattern of inheritance
has low penetrance for the disease-causing mutation. This may be due to a high level
of expression of a modifier gene. Changes in penetrance can lead to the absence of the
expressed phenotype even in a Mendelian disease. This can mean that cases look more like
complex disease cases than Mendelian disease cases and may also explain some of the cases
of monogenic complex disease.

Mendelian traits can be studied in families

Studying Mendelian traits is relatively easy compared with the study of complex diseases.
Mendelian diseases occur in families and they tend to present at an early age with a few
notable exceptions, including hemochromatosis (which presents later in life) and also
EOAD (which though called early onset, is not a genuine example of early-onset disease
as it does not occur until the affected individual is well into adult life). In Mendelian

2967_Ch02.indd 47 07/07/2015 10:30

 48 CHAPTER 2 Defining Complex Disease

disease, early onset increases the likelihood of availability of family members for genetic
testing and also influences the likelihood of family members engaging in genetic research.
Families are a rich source of information and traditional (parametric) linkage-based anal-
ysis can be applied. Of course not all families are informative. There may be too few fam-
ily members available for testing or too few affected family members. Other options in
Mendelian disease include non-parametric linkage analysis or case control population-
based association analysis. However, these other options are rarely used in Mendelian dis-
ease, except in the more common Mendelian diseases and when parametric linkage testing
is not possible. In contrast, these latter options are widely used in complex disease studies,
as discussed later in this chapter.

There are complications to Mendelian diseases

Though, as we have seen, Mendelian disease is relatively simple, there are a number of
additional complications to consider before we move on. These include locus, allelic, and
clinical heterogeneity, and also genetic anticipation and genetic imprinting.

Locus heterogeneity
Locus heterogeneity refers to the occurrence of the same trait arising from genetic varia-
tion at different loci. It is not uncommon for the same Mendelian disease to result from
mutations in different genes in different families.

Allelic heterogeneity
Allelic heterogeneity refers to the occurrence of the same trait/phenotype from differ-
ent mutations at the same locus. In these cases, a single gene is involved, but there are
multiple potential disease-causing mutations. An example of allelic heterogeneity is the
autosomal recessive disease cystic fibrosis, for which there are more than 1910 disease-­
causing mutations listed on the CFTR mutation database (http://www.genet.sickkids.
on.ca/StatisticsPage.html). Therefore, each patient may have a different CFTR genotype
and still have the same phenotype. Note, however, as with many of these Mendelian
diseases where there are multiple disease-causing mutations at a single locus, in cystic
fibrosis, there is one very common mutation, the Δ508 mutation, that accounts for the
majority of cases and a host of less common mutations that account for the remainder
(Figure 2.6).
Clinical heterogeneity
Most human diseases are very variable. Even in Mendelian disease patients with the same
genetic mutation display variation in clinical symptoms. This is referred to as clinical
hetero­geneity. It should not be assumed that this is entirely due to genetics. The environ-
ment is likely to play a significant role in the determination of clinical outcomes.
Genetic anticipation
Genetic anticipation describes the tendency of some Mendelian dominant diseases to
become more severe in successive generations. Examples of this phenomenon include
fragile X syndrome, myotonic dystrophy, and Huntington’s disease. All three of these
diseases arise from the inheritance of specific unstable trinucleotide repeats. In each case,
severity or age of onset is associated with the number of repeat sequences. The number of
repeats has been shown to increase in successive generations. Genetic anticipation in other
diseases remains controversial.

2967_Ch02.indd 48 07/07/2015 10:30

 Mendelian Diseases 49

∆508
Approx. 66%

G542X G551D

Cystic
fibrosis

N1303LYS W1282X

Another
1905
alleles all
at <1%

Figure 2.6: Mutations in the CFTR gene that cause cystic fibrosis. There are over 1910 listed mutations in the
CFTR gene. Of these, one accounts for approximately two-thirds of all mutations, while most others (with a few
exceptions) are found at frequencies of less than 1%.

Genomic imprinting
Genomic imprinting is considered to be an epigenetic phenomenon and is more com-
plex than any of the above. In some families with autosomal dominant disease the trait
is only expressed if inherited from a parent of one particular sex. This occurs despite
the fact that the disease-causing mutation can be present in both male and female par-
ents. An example of genetic imprinting is Beckwith–Wiedemann syndrome, which
is only expressed in children who inherit the mutated gene from their mother. There
are other examples of human disease where imprinting has been identified as a key
factor, including Prader–Willi syndrome, Angelman syndrome, and Silver syndrome
(Table 2.2). Research has identified approximately 100 different imprinted loci on 11
mouse chromosomes (http://www.mousebook.org/mousebook-catalogs/imprinting-
resource). Even so, imprinting is the least well understood of the many complications
that apply to Mendelian disease.
An example of genomic imprinting is given by IFG2 gene encoding insulin-like growth
factor 2. Offspring inherit one IGF2 allele from their mother and one from their father,
but the paternal allele of IGF2 is expressed, while the maternal one is completely silent. If
both alleles begin to be expressed in a cell, that cell may develop into a cancer.
Children with Prader–Willi syndrome have small hands and feet, short stature, poor sexual
development, and mental retardation, while children with Angelman syndrome exhibit
frequent laughter, uncontrolled muscle movement, a large mouth, and unusual seizures.
The imprint control centers have been found to restrict small areas known as imprinting
centers. In the case of the disorders above, the imprinting centers have been identified at
the 5′ end of the SNURF–SNRPN gene. Both syndromes are associated with the loss of
the same small region on the long arm of chromosome 15.
The most common maternal form of Angelman syndrome is a 4 kb mutation on chro-
mosome 15q11–q13 that contains the UBE3A gene encoding for E6AP ubiquitin ligase.
The mutation prevents methylation of the UBE3A gene and also silences the SNRPN

2967_Ch02.indd 49 07/07/2015 10:30

 50 CHAPTER 2 Defining Complex Disease

Table 2.2 Some examples of disease thought to involve genomic imprinting and some of the genes
involved.

Chromosomal
location of gene Candidate gene or genes if known Disease/syndrome/phenotype
11q33 PGL1 Paragangliomar 1: hearing loss
7p11.2 H19 Silver–Russell syndrome
11p15.5
5q35 NSD1/ARA263/STO Beckwith–Wiedemann syndrome
(also: Soto syndrome, Weavers
syndrome, acute myeloid leukemia)
11p15.5 H19: cyclin-dependent kinase Beckwith–Wiedemann syndrome
inhibitor IC (p57Kip2)
11p15.5 KCNQ1 overlap transcript 1 Beckwith–Wiedemann syndrome
11p15.5 H19 Wilms’ tumor 2
13q14.1–q14.2 RB1 Retinoblastoma
Microdeletion SNURF–SNRPN Prader–Willi syndrome: hypotonic,
of 15q11–q13 male hypogonadism, mental
retardation, and severe obesity
Microdeletion UBE3A Angelman syndrome: small size,
15q11–q13 severe retardation, lack speech, and
characteristic behavioral pattern
20q31 Pseudohypothyroidism type 1A
20q31 Pseudohypothyroidism type 1B

gene. If the deletion of this region is inherited from the mother, the offspring will develop
Angelman syndrome because the paternal gene is silenced and the maternal copy is almost
exclusively expressed.
Prader–Willi syndrome involves a similar mechanism, but in these cases the maternal
genes are silenced and not expressed, and the paternal genes are mutated. There may be a
number of different genes involved in cases of Prader–Willi syndrome; however, the same
pathway is affected, and the key to understanding the mechanism of the disease rests with
SNRPN and UBE3A.
It is tempting to ask why genomic imprinting occurs. This type of pressure may have sig-
nificance in evolutionary terms, and there may be conflicting pressures acting on maternal
and paternal alleles for genes, such as those that affect fetal growth (see Section 1.9).

2.4 Variation in The Mitochondrial Genome is

Associated With Disease
Not all of the human genome is inherited as autosomal DNA from the nucleus either
in our autosomes (chromosomes 1–22) or in our gonosomes (X and Y). A small

2967_Ch02.indd 50 07/07/2015 10:30

 Variation in The Mitochondrial Genome is Associated With Disease 51

proportion of our DNA is inherited from the mitochondria, the so-called mitochon-
drial genome (mtDNA). The human mitochondrial genome is extremely small com-
pared with the nuclear genome (i.e. the autosome). The mitochondrial genome is a
circular molecule of 16,569 bases in length (Figure 2.7). Mitochondrial genetics is
very different to Mendelian genetics. The mitochondrial genome is passed from gen-
eration to generation down the maternal line only and each cell may contain several
thousand copies.
The mitochondrial genome encodes a total of 37 genes: 13 encoding for proteins that func-
tion within mitochondria, two ribosomal RNAs (rRNA, 16S and 23S) and 22 transfer
RNA (tRNA) genes necessary for the synthesis of the 13 encoded polypeptides. There are
two nucleotide strands of the mtDNA: a heavy strand (H) rich in guanine and a light strand
(L) rich in cytosine. The H strand provides a template coding for the two rRNAs, 14 out of
the 22 tRNAs, and 12 out of the 13 proteins. The L strand is the template for the remain-
ing eight tRNAs and one protein. The mitochondrial genome is extremely compact with
approximately 93% of the DNA sequence representing coding sequences. Consequently, all
37 mitochondrial genes lack introns and are tightly packed within the coding region. The
remaining 7% of the mitochondrial genome encodes the displacement D loop or control
region where the replication of both the H and L strands begins. This region also encodes
the promoters for transcription of both H and L strands. The primary role of mitochondria
is to provide cells with the bulk of their adenosine triphosphate (ATP), which is used as an
energy source to drive cellular reactions. The proteins produced by the mitochondrial genes
work together to create this energy by turning complex molecules such as sugars into simple
substances such as carbon dioxide and water.

OH

F T
RNR1
V CYTB

RNR2

L1
ND5
ND1

I/M H, S2 and L2

ND2 ND4L
W R
ND3
G
COI COIII
D
COII K ATP8
and ATP6

Figure 2.7: An abridged map of the human mitochondrial genome. The genome is 16.6 kb and encodes 13
essential polypeptides of the oxidative phosphorylation (OXPHOS) system (ND1–ND6, NDL4, the cytochrome
c oxidases COI–COIII, cytochrome b, and the subunits of ATP synthase ATPase 6 and 8) and the necessary
RNA components (two rRNAs and 22 tRNAs) for their translation.

2967_Ch02.indd 51 07/07/2015 10:30

 52 CHAPTER 2 Defining Complex Disease

Variation in the mtDNA has been widely associated and linked with
many different diseases
The mitochondrial genome has been associated with many different diseases in humans
(Table 2.3). The majority of early studies concentrated on rare diseases caused by inher-
ited mutations at specific sites, such as Kearns–Sayre syndrome where there are large-scale
deletions in the mtDNA, Pearson’s syndrome, Leber hereditary optic neuropathy, mito-
chondrial myopathy, and many others, most of which occur in infants. More recently,
studies have begun to identify genetic associations with common diseases and illnesses,
including migraines, epilepsy, dementia, cardiomyopathy, renal tubular defects, adrenal
failure, and liver failure, which occur both in infants and adults (Figure 2.8). Diagnosing
and managing mtDNA disease is a major clinical challenge in the second decade of the
new millennium. One strategy that is being applied is to manipulate the mtDNA muta-
tion rates through exercise. A lack of exercise leads to a reduction in mitochondrial enzyme
activity. In contrast, endurance training improves enzyme activity. It has been suggested
that increased exercise may help to reduce the proportion of sporadic mutations, and may
be used to combat some of the clinical symptoms and illness caused by mtDNA muta-
tions. Identification of the dysfunctional mtDNA genes and sequences in mitochondrial
syndromes enables the development of new therapies based on targeting specific gene

Table 2.3 Some of the diseases that are encoded by genetic mutations and variations in the human
mitochondrial genome.

Disorder Genotype Phenotype Inheritance

Kearns–Sayre syndrome Large-scale deletion Ophthalmoplegia and Sporadic
cardiomyopathy
Chronic progressive external Large-scale deletion Ophthalmoplegia Sporadic
ophthalmoplegia (CPEO)
Pearson’s syndrome Large-scale deletion Lactic acidosis and Sporadic
pancytopenia
Mitochondrial myopathy 3243A > G TRNL1 Stroke like Maternal
3271R > C ND1 and
ND5
Myoclonic epilepsy associated 8344A > G TRNK Myoclonic epilepsy Maternal
with ragged red fibers 8356T > C TRNK
(MERRF)
Neuropathy, ataxia and 8993T > C ATP6 Neuropathy Maternal
retinitis pigmentosa (NARP)
Mitochondrial complex V 8993T > C ATP6 Brain stem Maternal
(ATP deficiency) (MILS)
Maternally inherited diabetes 3243A > G TRNL1 Diabetes and deafness Maternal
and deafness (MIDD)
Leber hereditary optic 3460G > A ND1 Ophthalmic neuropathy Maternal
neuropathy 1178G > A ND4
1448T > C ND6

2967_Ch02.indd 52 07/07/2015 10:30

 De Novo Mutations and Human Disease 53

Cardio-
myopathy

Liver failure Diabetes

The
Parkinson’s mitochondrial Myopathy
genome

Epilepsy
and Deafness
dementia
Neuropathy

Figure 2.8: Mitochondrial genome and common diseases. Some of the medical conditions that have been
claimed to be related to variation in the mitochondrial genome.

sequences using procedures like allotropic expression, whereby it is possible to express

the wild-type using transfected hybrid cell lines.

2.5 De Novo Mutations and Human Disease

One group of human conditions rarely discussed is that caused by de novo mutations.
These diseases and traits are only just being understood. De novo mutations are new
mutations, so there are no other affected family members and therefore they are not
Mendelian traits. They are also caused by simple mutations, so these are not chromo-
somal disorders involving large sections of the genome. The rate of mutation in the
genome is low (as discussed in Chapter 1) and so these mutations are rare. However,
these mutations when they are disease-causing mutations, that is having the mutation is
both necessary and sufficient for the disease to occur, cannot themselves be considered
with genetically complex diseases because carrying the mutation will cause the disease. In
this case we are not talking about mutations that simply increase disease risk. However,
de novo mutations that are not disease causing, but may be associated with disease risk,
can be considered with complex disease (as occurs in some cancers). The former group
are a category unto themselves.
One example of a trait caused by a de novo mutation is Hutchinson–Gilford progeria
syndrome (HGPS). HGPS is incredibly rare, occurring in around one per 4 million live
births in the USA. There is no family history with this trait because the mutation occurs
in the gametes, most often in the sperm. The mutation is not found in the parent’s auto-
some. The symptoms of HGPS are rapid aging; life expectancy can be very short at around
12–13 years. Patients show extremely early signs of aging, hair loss and aging of the skin,
and severe osteoporosis. We see the traits that we expect to be expressed in an average quite

2967_Ch02.indd 53 07/07/2015 10:30

 54 CHAPTER 2 Defining Complex Disease

healthy octogenarian in an HGPS patient often before the age of 10 years. This is a cruel
condition for parents and affected children alike.
Genetic testing has shown that the major mutation for this syndrome is a mutation encod-
ing an exchange of cytosine for thymine at position 1824 (C1824T) in the lamin A gene
(LMNA), which gives rise to a shortening of the lamin A protein. The gene is located on
chromosome 1q22. There is a degree of clinical heterogeneity in HGPS and this may be
in part due to genetic variation. Other, less frequent mutations have also been identified
in LMNA and these have been associated with longer life in some cases. LMNA mutations
are also associated with other genetic conditions, such as Charcot–Marie–Tooth disease
and Emery–Dreifuss muscular dystrophy.
Lamin A is an important structural protein and one of a family of intermediate filament
proteins. Intermediate filaments provide stability and strength to all our cells. Lamin A is
a scaffolding component of the nuclear envelope that surrounds the nucleus in cells, cre-
ating a mesh-like layer of intermediate filaments attached to the inner membrane of the
nuclear envelope of the cell.
In HGPS, the C1824T mutation encodes a short version of the lamin A protein called
progerin. This protein cannot be processed correctly within the cell. This can disrupt the
nuclear envelope, and over time damage the structure and function of the nucleus, caus-
ing cells to die prematurely. However, there is hope in HGPS and clinical trials have been
started, though it is early days.

2.6 Three Different Types of Complex Disease

Complex diseases as defined above do not lend themselves to simple analysis or display
predictable patterns of inheritance (i.e. Mendelian patterns). However, we can distinguish
different levels of complexity. The greatest problem is identifying genuine monogenic
complex disease. These diseases are near-Mendelian, involving only a single risk allele,
but do not behave in a simple pattern. Whether this is due to incomplete penetrance of
the Mendelian disease-causing mutation is unclear. However, until these cases have been
defined the concept of a monogenic complex disease remains open. In many of these dis-
eases there are also Mendelian families with the same disease, which adds a further layer
of complication. Even with these caveats, true monogenic complex diseases are rare; most
genetically complex diseases are associated with multiple risk alleles and are more correctly
referred to as either oligogenic or polygenic. The difference between these last two catego-
ries is very gray. Oligogenic implies that there are several risk alleles and polygenic implies
that there are many risk alleles. In practice, in oligogenic complex disease, risk alleles are
easier to identify and there are likely to be more individual Mendelian families and non-
Mendelian families for analysis.

Studying complex disease is different from studying Mendelian

disease
In contrast to Mendelian disease, which relies mostly on linkage analysis of informa-
tive families, in complex disease even if informative families are available they have
often dispersed (due to late onset) and even when DNA samples can be obtained the
only options are to use non-parametric linkage analysis or case control population-based
association analysis. The former is a linkage-based method that does not require the

2967_Ch02.indd 54 07/07/2015 10:30

 Three Different Types of Complex Disease 55

Table 2.4 Some of the common major differences between genetically complex diseases compared
with Mendelian disease.

Mendelian traits Complex traits

Mostly early onset Mostly late onset
High penetrance Low penetrance
Mostly familial Some familial, but mostly sporadic
Known pattern of inheritance Unknown pattern of inheritance
Suitable for parametric linkage analysis Not suitable for parametric analysis
Use non-parametric linkage analysis
Use mostly association analysis
Low/no environmental input (exceptions Moderate/high environmental input
include phenylketonuria, Wilson’s disease, and
hemochromatosis)

use of a genetic (Mendelian) model. Therefore, there are no preset parameters as there
are in more conventional parametric linkage analysis. Non-parametric linkage analysis
can be based on families with multiple cases of disease or on affected sibling pairs. Case
control population-based association analysis is a quite different method that simply
compares the frequency of genetic variations between groups of patients or cases and
healthy (matched) individuals from the same population. Until recently, most genetic
studies in complex disease were based on association analysis and this was thought to
be a poor mechanism for the identification of disease genes. However, for various rea-
sons, opinions have changed about the value of association studies in complex disease
(see Chapter 3). A number of other differences between studies of Mendelian versus
complex disease are listed in Table 2.4.

Monogenic complex diseases involve a single risk allele

The idea of a monogenic complex disease is difficult to explain, as discussed above.
However, the use of the term monogenic does not by itself suggest a single allele at a locus
is both necessary and sufficient for the disease to be expressed. If this were the case, then
this would be a Mendelian disease. However, to argue that all monogenic disease must be
Mendelian is incorrect because such ideas ignore incomplete penetrance. In monogenic
complex disease there is always incomplete penetrance in families, therefore the disease-
promoting allele is not sufficient for the disease to occur. In addition, because there are
cases that do not carry the disease-promoting allele, we can say that the disease-promoting
allele is not necessary for the disease to occur. A further complication is that in many dis-
ease populations where there are monogenic cases the risk allele for each case may be on
a different gene. In a disease population, different individuals may have different genetic
risk portfolios, so that in a population of patients with a specific disease there could be
some familial Mendelian cases, some complex monogenic cases, and some oligogenic or
polygenic cases. Altogether, monogenic complex disease is a gray area between Mendelian
and complex disease, and though individual cases do occur, the majority of complex dis-
ease fits into an oligogenic or polygenic pattern.
We must also consider the possibility that in some cases they are in fact Mendelian. In these
cases we may be fooled by two things. First, expression may not be seen in all individuals

2967_Ch02.indd 55 07/07/2015 10:30

 56 CHAPTER 2 Defining Complex Disease

with the risk allele, because expression is modified by another gene causing incomplete
penetrance. Second, the presence of the disease in some without the risk allele may simply
indicate a different disease-causing mutation is active in those individuals. Taken altogether
this could explain why the disease-causing mutation is neither sufficient nor necessary for
the disease to occur and may explain a large proportion of monogenic complex disease.

Oligogenic complex diseases involve several alleles

Risk alleles are easier to identify in oligogenic disease than in polygenic disease because,
by their nature, in oligogenic diseases there are less risk alleles involved (as the name
implies) and there tends to be a higher heritable component compared with polygenic
disease. As a consequence families are more frequent, onset is more likely to be early,
and the risk alleles tend to have large effects as measured by OR, relative risk, or LOD
score (these different statistical terms are explained in Chapter 5). Hirschsprung’s disease
(HSCR) is a good example of an oligogenic disease (see Section 2.8) with at least 13 iden-
tified risk alleles: two major risk alleles, three minor risk alleles, and a group that may be
modifier alleles, some of which may as their implied role suggests only act in the presence
of the major risk allele (or alleles). A further five chromosomal regions have been identi-
fied in HSCR bringing the total to 18, but no specific genes have yet been identified as
potential risk alleles.

Polygenic complex diseases involve many risk alleles

One of the difficulties of defining polygenic disease is the word “poly.” The word itself
means many, but how many is the question? There is no specific number for this term.
However, it is quite clear that some diseases are linked or associated with a large number
of different risk alleles. This does not mean that every case will inherit every risk allele or
that there may not be occasional near-Mendelian cases in the disease population. In this
regard it is important to consider patients in the context of populations and population
subgroups. The term poly applies to the patients as a group. Overall, there are many risk
alleles associated with the disease and in this case the term poly would apply. Generally,
the contrast with oligogenic disease becomes clear when we consider the size of the heri-
table component and frequency of families in the two models. The example of Crohn’s
disease (see Section 2.9) is a good example of a polygenic disease because though there
are familial cases, most cases are sporadic and the number of identified risk alleles is now
considered to be greater than 163 and rising. Compare the latter figure of 163 poten-
tial risk alleles in Crohn’s disease to the figure for HSCR, where there appears to be no
more than approximately 18 different risk-associated alleles, including five chromosomal
regions in which risk alleles have yet to be identified. If we then consider that out of this
group of 18 risk alleles for HSCR, two mutations have very major strong effects and the
remainder have weak effects, with most appearing to operate as modifiers in the presence
of the major risk allele, then differences between oligogenic and polygenic disease are quite
clearly demonstrated.
However, it would be incorrect to assume that all cases of a specific disease have the
same genetic pattern. A disease population is best defined as a group of patients with
the same or similar signs and symptoms. Though disease groups have the same clinical
picture, the disease may arise through a different pathway and the genetic contribution
to pathogenesis may be different in different cases. Nowhere is this better illustrated
than in the three selected examples described in Sections 2.7–2.9. Alzheimer’s disease
occurs in two forms: EOAD and LOAD. Most EOAD cases are Mendelian families
(but not all), while most LOAD cases are sporadic, genetically complex cases. A single

2967_Ch02.indd 56 07/07/2015 10:30

 Alzheimer’s Disease May be a Monogenic Complex Disease 57

risk allele is likely to be the total genetic contribution in some of the latter (i.e. they
are monogenic). In HSCR, not all cases are complex—a significant proportion are
due to chromosomal abnormalities or are familial with clear Mendelian inheritance
and only 70% of cases fit the pattern of oligogenic complex diseases. In Crohn’s dis-
ease, approximately 10% of cases may have a familial pattern of inheritance and some
of these are likely to be Mendelian cases, but the majority of cases are sporadic and
genetic susceptibility is determined by a large number of different risk alleles, indicat-
ing a polygenic model.

2.7 Alzheimer’s Disease May be a Monogenic

Complex Disease
LOAD is an example of a complex disease where there may be a significant number of
monogenic cases. Alzheimer’s disease is a very heterogeneous disease and it is also very
common, particularly in the developed world. This disease is characterized by progres-
sive dementia in the elderly or aging population caused by the formation of intracellular
neurofibrillary tangles (NFT) and extracellular amyloid plaques that accumulate in the
vulnerable regions of the brain, causing damage to the brain as illustrated in Figure 2.9.
It has been observed that reductions in microtubule-dependent transport occur in
Alzheimer’s disease and this may stimulate proteolytic processing of the β-amyloid precur-
sor protein (APP), resulting in the development of senile plaques that are seen in Alzheimer’s
disease. As discussed above, Alzheimer’s disease can be either Mendelian (EOAD, about 5%
of cases) or complex (LOAD, about 95% of cases). The genes involved in EOAD are APP,

(a) (b) Abberant amyloid Aging Abberant tau

processing processing
APP oxidative stress MAPT
Healthy Advanced APP tau
brain Alzheimer’s BACE disturbed calcium
β-secretase homeostasis
PSEN1,2, NCSTN decreased DNA
γ-secretase repair capacity
Aβ chromosomal hyper-
NEP instability
degradation phosphorylation
DNA damage
neuritic
plaques
soluble Aβ neurofibrillary
fibrils tangles

synaptic and neuronal

dysfunction

disrupted synaptic and

neuronal integrity

synaptic and neuronal loss

Dementia

Figure 2.9: The mechanisms of Alzheimer’s disease. (a) A healthy brain compared with a brain with advanced
Alzheimer’s disease. (b) The processes and some of the genes involved in the formation fibrils, plaques, DNA
damage, and neurofibrillary tangles. (From Armstrong L [2014] Epigenetics. Garland Science.)

2967_Ch02.indd 57 07/07/2015 10:30

 58 CHAPTER 2 Defining Complex Disease

PSEN1, and PSEN2. In LOAD, it is the apolipoprotein E4 (APOE4) allele that confers the
major risk. There is clear evidence for a causative link with APOE4 and LOAD. Prince et al.
(2004) found that the presence of the APOE4 allele was strongly associated with reduced
levels of the β-amyloid-42 protein in cerebral spinal fluid. However, there are multiple
other risk alleles that have been identified for LOAD; OMIM currently lists at least 16
possible genetic sites and many more are listed in the literature that have yet to be added to
this list. There are individual cases of LOAD where the APOE4 allele is the only identified
risk allele. APOE4 may account for as much as 50% of the genetic risk in LOAD. However,
new research may at anytime identify hitherto unknown risk alleles. In addition, we cannot
ignore the potential effects of the environment. For example, Evans et al. (2004) reported
an association between total serum cholesterol and disease progression in patients with
LOAD who did not have the APOE4 allele.

2.8 HSCR – an Oligogenic Complex Disease

HSCR was described by Harald Hirschsprung in 1886 and it is defined by the congenital
absence of ganglia in some or all of the large intestine or colon. HSCR occurs because
there is a failure of development of the central nervous system or neural crest. The con-
sequences of this are a lack of peristalsis in the gut, which leads to a toxic megacolon in
the newborn. This can be fatal when left untreated. Most cases of HSCR are sporadic and
these fit the definition of oligogenic complex disease.

Sporadic HSCR illustrates the oligogenic model for complex disease

Amiel and Lyonnet (2001) reported 12% of cases are due to chromosomal abnor-
malities, 18% are part of a syndrome (mostly Mendelian), and 70% are isolated non-­
syndromic cases, often familial but not Mendelian (Figure 2.10). HSCR is associated
with a sibling relative risk of 187. The major risk genes and alleles for HSCR have been

Chromosomal

Mendelian

Complex
(sporadic)

Figure 2.10: HSCR is either a chromosomal, Mendelian, or a complex disease. The majority of cases of HSCR
are genetically complex (70%); however, a substantial number are familial cases (inherited in a Mendelian pattern,
12%) and some occur from chromosomal abnormalities (18%).

2967_Ch02.indd 58 07/07/2015 10:30

 HSCR – an Oligogenic Complex Disease 59

Table 2.5 Genetic associations with non-syndromic HSCR: Examples

from OMIM and other sources December 2014.

Name Location Function

HSCR1 10q11–q21 RET
HSCR2 13q22.3 EDNRB
HSCR3 5p13.2 GDNF
HSCR4 20q13.32 EDN3
HSCR5 9q31 Unknown
HSCR6 3p21 Unknown
HSCR7 19q12 Unknown
HSCR8 16q23 Unknown
HSCR9 4q31 Unknown
Other minor associated genes 22q13 SOX10
4p12 PHOX2B
2q22 ZFHX1B
Other minor modifier genes 16p13.3 SOX8
8p22–p11 NRG1
Xq28 L1CAM
15q22.3–q23 BBS4
2q31 BBS5
4q27 BBS7

identified. These include a deletion in the RET transforming sequence (oncogene RET)
that encodes a receptor tyrosine kinase on chromosome 10q in the area 10q11–q21
(HSCR1) which is found in 50% of familial and 15–35% of sporadic cases, the gene
encoding endothelin receptor type B (EDNRB) on chromosome 13q (HSCR2), as well
as 11 other gene loci. Three of these 11 other genes appear to have a minor function and
six of the 11 appear to be modifier genes. In addition, there are five identified chromo-
somal locations for which the candidate genes have yet to be identified (summarized in
Table 2.5).
The RET oncogene deletion is the major genetic risk factor for the isolated non-
syndromic form of the disease, but does not account for all cases. In addition, there
is a male bias for cases with this deletion (65% males). Linkage analysis indicates that
several other genes interact with the RET deletion to increase the risk of the disease
as noted above.
The EDNRB gene was identified as a risk factor for HSCR through a series of linkage stud-
ies in a Mennonite kindred that identified deletions and mutations in chromosome 13q.
Independent work in mouse models suggested a relationship between EDNRB, which is
located within this region. As with the RET mutation, there is incomplete penetrance and
also penetrance may be sex-specific.
Both RET and EDNRB encode receptors that are active in the process of neural crest
stem cell regulation. It is therefore easy to understand how mutations in these genes can

2967_Ch02.indd 59 07/07/2015 10:30

 60 CHAPTER 2 Defining Complex Disease

give rise to HSCR. As there are two pathways in the process, it is also possible to envis-
age why a mutation in one pathway does not always lead to the clinical syndrome being
expressed.
Other susceptibility alleles for HSCR have been proposed and research suggests that
mutations in the genes encoding the ligands for RET and EDNRB may be involved as
well as a number of genes encoding for other components of the biological pathways
that are involved in neural crest stem cell development (Figure 2.11). Thus far, studies
have identified mutations in GDNF and EDN3, as well as ECE1 and NTN. An asso-
ciation with SOX10 (which may interact with ZFHX1B and SOX8) has been reported,
though this remains controversial and may relate to other similar developmental disor-
ders. Another gene implicated in HSCR is the PHOX2B gene, which interacts with the
RET gene.
Some of the associated risk alleles above and a number of other more recently identified
potential risk alleles (L1CAM, which interacts with ENDRB, and NRG1, BBS4, BBS5,
and BBS6, which all interact with RET) may act as modifiers in the presence of the major
risk alleles, which may explain why they are only seen to increase disease risk in those car-
rying the major risk allele.
Inheritance of multiple risk alleles may explain clinical heterogeneity and in a complex
model may also explain incomplete penetrance. HSCR genetics illustrates the importance
of considering the whole genome and studying complete biological systems. Recent devel-
opments have enabled both of these objectives to be possible, though much more work
remains to be done before the work on HSCR genetics is complete.

Endothelin
RET pathway
pathway

RET oncogene EDNRB gene

RET ligands EDNRB ligands

ECE1
GDNF and NTN EDN3

Ganglia cell
colonization in
the GI tract

Figure 2.11: Major pathways in neural crest stem cell development with links to HSCR. This figure
illustrates the major pathways involved in neural crest stem cell development implicated by linkage and
association with susceptibility to HSCR. The major contributing genes are RET and EDNBR. Polymorphisms
of other genes whose products are important in these RET and endothelin pathways have been identified as
potential risk factors for HSCR. GI, gastrointestinal. (Adapted from Strachan T & Read A [2004] Human Molecular
Genetics, 3rd ed. Garland Science.)

2967_Ch02.indd 60 07/07/2015 10:30

 Crohn’s Disease is Mostly a Polygenic Complex Disease 61

2.9 Crohn’s Disease is Mostly a Polygenic

Complex Disease
Crohn’s disease serves as a good model for polygenic complex disease. The disease was first
described by Dalziel, a Glaswegian surgeon, in 1913, but the definitive paper was pub-
lished by Crohn, Ginzburg, and Oppenheimer in 1932, and the name Crohn’s disease was
applied. Crohn’s disease is one of a group of inflammatory bowel diseases (IBDs), though
inflammation in Crohn’s disease is not restricted to the bowel, but can be found through-
out the gastrointestinal system, from the oral cavity to the anus. The other major form of
IBD, known as ulcerative colitis, was described in 1888 by Hale-White. Ulcerative colitis,
unlike Crohn’s disease, is restricted to the lower gut. Despite this, there is frequent clinical
overlap between Crohn’s disease and ulcerative colitis, and there is also a level of shared
pathology and genetic risk.

Early studies of Crohn’s disease suggested a number of locations

for risk alleles
Approximately 10% of cases with Crohn’s disease are familial, with the remaining 90%
being sporadic. Estimates of the sibling relative risk vary from 17- to 35-fold and there is
50% concordance for the diseases in monozygotic (identical) twins compared with 10%
concordance in dizygotic (non-identical) twins. These values are considerably lower than
those for HSCR which boasts a sibling relative risk of 187, reflecting the greater level of
heritability for HSCR compared with Crohn’s disease. Early genetic studies based on link-
age analysis of families identified several (tagged) regions of the genome with significant
linkage to Crohn’s disease, a selection of which are listed in Table 2.6. Among the identi-
fied regions, chromosome 16q (tagged as IBD1) was associated with the greatest risk of
Crohn’s disease. A brief history of early and recent genetic studies is given in Figure 2.12.
Twin studies have always been valued in genetics. They are simple studies that investigate
the degree of co-inheritance of a disease or trait in monozygotic and dizygotic twins. The
difference in the level of concordance between the two groups is considered to be a mea-
sure of the degree of heritability for the disease or trait. The drawback of using twin data is
that the twins share the same environment unless adopted into different families, and the
quality of the data obtained depends on the numbers studied. As identical twins are rare,
numbers are likely to be low in less common diseases.

Table 2.6 Major susceptibility regions for IBD identified by early linkage analysis and GWAS.

Ulcerative Putative
Region IBD-LABEL Crohn’s disease colitis candidate gene
16q12 IBD1 Strong None CARD15 (2001)
12q13.2–q14.1 IBD2 Moderate Moderate
6p21.3 IBD3 Moderate Moderate MHC/HLA
14q11–q12 IBD4 Moderate Moderate
5q31 IBD5 Strong Moderate Cytokine cluster
19p13 IBD6 Moderate Moderate 2007 WTCCC1
GWAS
1p36 IBD7 Moderate Moderate

2967_Ch02.indd 61 07/07/2015 10:30

 62 CHAPTER 2 Defining Complex Disease

2008–2009
Th17 endoplasmic
reticulum stress
and gut barrier
1988–1996 integrity
Twin Crohn’s disease
concordance 2001 meta-analysis
and ethnic NOD2/CARD15 2010
diversity Hugot/Ogura 71 alleles

1996 2006/2007 GWAS Deep

VNTR linkage IL23R sequencing
(GWLS) analysis ATG16L1 Novel therapies
list with multiple IRGM and new models
IBD regions

Figure 2.12: Time-line of genetic investigations in Crohn’s disease. The figure shows the timing of major
developments in our understanding of the genetics of Crohn’s disease from 1988 to recent.

Genetic variations in the human equivalent of the plant nod2 gene

(CARD15) were the first identified and confirmed Crohn’s disease
risk alleles
Following on from the early studies referred to above, increasing levels of fine mapping were
applied seeking to identify the specific susceptibility loci within each tagged region. The
first major Crohn’s disease risk gene was identified in 2001 as CARD15 (caspase recruit-
ment domain family member 15) on chromosome 16q12, which encodes the human
homolog of the plant root nodule protein NOD2. These studies reported two common
mutations and a deletion in the CARD15 gene. Further studies identified more, less com-
mon mutations. As expected in complex disease, possession of the CARD15 mutations
confers an increased risk of Crohn’s disease, but is neither necessary nor sufficient for the
disease to occur (Table 2.7).

Table 2.7 CARD15 mutations and variations encode major risk alleles in Crohn’s disease – estimates
from case control cohort studies.

Crohn’s disease patients

(N = 688) versus healthy OR (95% confidence
Allele controls (N = 250) interval)
SNP8 (R702W) 11 versus 4.4 2.66 (1.59–4.46)
SNP12 (G908R) 4.1 versus 0.007 6.33 (2.15–18.59)
SNP13 (insertion C at 3020 ) a
6.7 versus 2.8 2.59 (1.37–4.91)
Compound heterozygote 6.6 versus 0 24.85 (3.82–161.67)
Homozygotes 5.9 versus 0 22.19 (3.96–124.36)
a
SNP13 is not a SNP, it is an insertion. Compound heterozygotes carry a combination of SNP8, SNP12, and
SNP13. In 2002 Lesage et al. identified 27 rare variants not listed in this table.

2967_Ch02.indd 62 07/07/2015 10:30

 Crohn’s Disease is Mostly a Polygenic Complex Disease 63

NOD2 is a NACHT-LLR family member with two N-terminal caspase activation and
recruitment domains (CARDs), a nucleotide-binding domain, and a series of leucine-rich
repeats (LLRs) at the C-terminal. NOD2 expression is mostly restricted to the cytosol in
monocytes and Paneth cells. Paneth cells, named after Joseph Paneth, are specialized cells
found in the small intestine and appendix. Their precise function is unknown, though
they are thought to have antibacterial properties, and may be important in immune regu-
lation and responses to bacteria in the gut. NOD2 activates the nuclear factor NF-κB
through interaction between the LRR domain and muramyl dipeptide (MDP) on Gram-
positive bacteria. This interaction enables the CARD domains of NOD2 to interact with
the protein kinase RICK (also RIP or TRAF). RICK induces K63-linked ubiquitinylation
of the signaling molecule IKKY (NEMO), which in turn causes phosphorylation of the
NF-κB inhibitors allowing NF-κB to initiate production of a series of pro-inflammatory
and regulatory cytokines, including interleukin (IL)-1B, IL-8, IL-10, and tumor necro-
sis factor (TNF)-α (Figure 2.13). Under normal circumstances this process leads to host
recognition and destruction of bacterial pathogens in the gut. However, when CARD15
is present in the mutant form, the normal mechanism of signaling is interrupted and
there is decreased NF-κB production. However, this is not the whole story. Crohn’s dis-
ease has been associated with infection with Mycobacterium tuberculosis, Pseudomonas

Commensal
bacteria

Gastric
epithelium

NOD2

RICK TLR4 TLR2

NF-κB

Apoptosis TNF-α IL-1β IL-6 IL-8 IL-10 IL-12

Figure 2.13: NOD2 immune interactions in maintenance of homeostasis for gut commensals. NOD2
polymorphism is the major susceptibility determinant for Crohn’s disease. NOD acts through a series of
pathways to influence the immune response to gut commensals. The TLR family are major components of the
immune system involved in immune homeostasis of the gut. Missense mutations in TLR4 (D299G and T399I) are
associated with susceptibility to Crohn’s disease. TLR4 expression is increased in the intestinal epithelial cells in
active disease, but found at low levels only in healthy intestine. The figure illustrates bacterial invasion (shown
as tadpoles or small figures with tails) through the gut wall. Both TLR2 and TLR4 interact with NOD and each will
cause up- or downregulation of pro-inflammatory and regulatory cytokines. Polymorphisms that create higher or
lower activity of these immune compounds may have a significant influence on bacterial homeostasis in the gut.
This figure represents only part of what is beginning to look like a very complex picture for Crohn’s disease. NOD2
also interacts with NF-κB, and NF-κ promotes apoptosis and stimulates the production of both pro-inflammatory
and regulatory cytokines.

2967_Ch02.indd 63 07/07/2015 10:30

 64 CHAPTER 2 Defining Complex Disease

species and Listeria species. Crohn’s disease patients have exaggerated immune responses
in the gut and earlier genetic studies identified several other areas of significant linkage
for Crohn’s disease (Tables 2.6 and 2.7), suggesting other factors and pathways may also
play a role in susceptibility, in particular genes that regulate the immune response to
bacteria.

Genetic variations in other immune regulatory genes are important

risk factors in Crohn’s disease
One specific immune regulatory pathway involves the ATG16L1 (autophagy-related
16-like 1) gene. Studies have suggested that a mutation in this gene is associated with
increased risk of Crohn’s disease (the region this gene is located in has been labeled IBD10).
Activation of NOD2 through MDP interaction and interaction of the CARD domains
with RICK has also been shown to induce autophagy. Other pathways involved in the
recognition of bacterial lipopolysaccharide and peptidoglycan have been investigated in
Crohn’s disease. The toll-like receptor (TLR) genes encode a family of pattern recogni-
tion receptors that interact in a complex way to regulate immune responses, to self and
non-self and genetic variation in these receptors may be crucial in determining susceptibil-
ity to a wide range of different diseases. TLR2 and TLR4 act as extracellular receptors, and
are important in inducing immune tolerance to normal gut bacteria. TLR-2 interaction
in CARD15 mutants leads to decreased IL-1β, IL-10, and IL-12 production, whereas
TLR-4 interaction in CARD15 mutants leads to increased TNF-α release.

The Wellcome Trust Case Control Consortium (WTCCC1):

Crohn’s disease
The WTCCC1 study provided a gold standard model for case control studies in com-
plex disease. Published in 2007, this study included Crohn’s disease among the seven
diseases investigated. The study was not the first genome-wide study in Crohn’s disease,
but it was the first GWAS. This study, as the name implies, was based on a study of the
genetic variation across the whole genome using common SNPs (frequency of 5% or
more) as tagged markers for susceptibility loci in a case control population-based study.
The difference to previous studies is that GWAS are not restricted to candidate genes or
areas of the chromosomes identified by previous analysis. This hypothesis-non-restricted
(or hypothesis-non-constrained or hypotheses-free) approach aimed at identifying
hitherto overlooked susceptibility alleles in large-scale studies with a high degree of resolu-
tion. Key differences are the size of these studies, usually 2000 or more cases with an equal
(or greater) number of controls, and the number of markers used to identify associations.
This means routinely using more than 100,000 and occasionally more than 1 million
markers in the form of tagged SNPs for a study. The WTCCC1 study included analysis
of more than 500,000 such tagged SNPs in 2000 cases from seven different common
genetically complex diseases and 3000 shared controls. Analysis of Crohn’s disease identi-
fied 10 independent genetic associations at probabilities P < 5 × 10−7, of which six were
confirmations of previously reported associations:
• CARD15 on chromosome 16q12 (previously IBD1).
• A cluster of potential candidates on chromosome 5q31 (previously IBD5, the precise
identity of which is a matter of dispute).
• IL23R on chromosome 1p31 (previously IBD17).

2967_Ch02.indd 64 07/07/2015 10:30

 Crohn’s Disease is Mostly a Polygenic Complex Disease 65

• ATG16L1 on chromosome 2q37 (previously IBD10, which had the strongest reported
association for Crohn’s disease in this study P < 7.1 × 10−14).
• An unknown locus in the chromosome 10q21 region around rs10761659 (previously
IBD15, which is 14 kb telomeric of the ZNF365 gene and 55 kb centromeric of the
pseudogene antiquintin-like 4).
• An area around rs17234657 within a 1.2 Mb gene desert on chromosome 5q13.1
(previously IBD18).
In addition, four novel strong associations were identified:
• Chromosome 3p21 (previously IBD12). There are a number of candidate genes in
the area, of which MST1 (macrophage stimulating 1) is the most plausible. This gene
encodes a protein that affects macrophage function and macrophages have a major role
in phagocytosis – a key process in maintaining defense against bacterial invasion.
• The IRGM (immunity-related guanosine triphosphatase) gene on chromosome 5q33
(previously IBD19). IRGM encodes a GTP-binding protein that induces autophagy
and reduced IRGM activity could lead to persistence of intracellular bacteria, which is
consistent with the function of other genes identified as those carrying risk alleles for
Crohn’s disease.
• A cluster on chromosome 10q24.2 (previously IBD20). The likely candidate from
this cluster is NKX2-3 (NK2 transcription factor-related locus 3). Disruption of this
gene homolog in mice leads to abnormal development of the intestine and secondary
lymphoid organs. It has been suggested that mutations in this gene may disrupt gut
migration of antigen-responsive lymphocytes and influence the inflammatory response
in Crohn’s disease.
• Close to the PTPN2 gene on chromosome 18p11 (previously IBD21). PTPN2
(protein tyrosine phosphatase non-receptor type 2) is a key negative regulator of the
immune response. Members of this family have been associated with increased risk of
both rheumatoid arthritis and type 1 diabetes.
In addition to these 10 highly significant loci, there are a further eight chromosomal
regions identified as having moderate associations: 1q24, 5q23, 6p21, 6p22, 6q23, 7q36,
10p15, and 19q13. Most of these associations are unsurprising, e.g. the association with
the region of 6q23 which encodes the TNFAIP3 gene (TNF-α-induced protein 3). The
product of the TNFAIP3 gene inhibits TNF-α-induced NF-κB-dependent gene expres-
sion through RIP and TRAF-2-mediated transactivation signaling in the same way that
NOD2 works on NF-κB. This illustrates the common link between many of these genes
and autophagy and phagocytosis, both of which are key processes in the defense of the
gut from pathogenic bacteria and maintenance of homeostasis for normal gut bacteria.
Even the association with the 6p21 region is not surprising. 6p21 is the home of the
human MHC within which the HLA genes are to be found. HLA antigens are key players
in both adaptive and innate immunity, and therefore an association between an immune
inflammatory disease like Crohn’s disease and genetic variation in the MHC is highly
likely. Interestingly, the genetic association between Crohn’s disease and HLA has been a
matter of dispute, and is considerably weaker than that seen for many autoimmune and
viral diseases as well as some adverse drug reactions. Some have suggested this association
is due to the close proximity of the gene for TNF-α (TNFA), which maps within the
MHC and not the HLA genes.

2967_Ch02.indd 65 07/07/2015 10:30

 66 CHAPTER 2 Defining Complex Disease

The current number of risk alleles for Crohn’s disease may be as

high as 163
Currently, OMIM lists a total of 29 IBD genes. Interestingly, the OMIM database lists
the 10 highly significant associations reported by the WTCCC1 study (including super-
script 1, 2 and only two of the eight moderately associated regions (IBD3 and IBD6).
The remainder are not currently listed. However, this may be incorrect. A review in 2011
identified at least 14 major GWAS in Crohn’s disease and recent studies suggested that
there are at least 71 risk alleles for Crohn’s disease, with a possible total over 163 if all of
the chromosomal regions identified for IBD so far are considered.
Studying the genetics of complex disease is to some extent beginning to be similar to train
spotting as more and more genes with small effects are added to the list. More important
than the list of alleles and their locations is what do the associations tell us about the
disease pathogenesis? To understand this better it may necessary to group some of the
risk alleles according to function and pathways, seeking similarity and differences that
may explain phenotypes. This is illustrated in Figure 2.14, where the different major
risk alleles for Crohn’s disease are grouped according to function. The different GWAS in
Crohn’s disease all identify candidate genes involved in the maintenance of the intestinal
barrier and the activity of the different cells within the barrier. There are three major gene
subgroups. First, CARD15, which is expressed in the gut wall in the Paneth cells and rec-
ognizes a variety of organisms, including commensal bacteria. The second group includes
ATG16L1 and IRGM, which are involved in autophagy, i.e. the degradation of unwanted
cellular organelles and pathogens. The third group includes genes that regulate pathogenic
T cell differentiation, such as members of the IL12–IL23 pathway. Bacterial handling
and MHC class II presentation appear to be regulated by CARD15 through ATG16L1.
This indicates how we can start to consider how different genes operate to form systems
involved in disease pathogenesis.

Biological pathways in
Crohn’s disease

CARD15 XBP1
TLR4 ATG16L1 CAPN10
IRGM ORMDL3
LKKR2 ERAP2
MTMR3

IL23R CCR6 MHC TL1A

STAT3 IL27 IL-28 TNFRSF6B
JAK2 TYK2 MST1 IL22
ICOSLG IL18R1-IL18R SMAD3
AP TAGAP
CCl cluster FUT2

Figure 2.14: Genetically identified pathways associated with increased risk of Crohn’s disease. The
five arrows point to five different gene groups: innate immune signaling genes (CARD15 or NOD2 and TLR4),
autophagy (ATG16L1, etc.), the endoplasmic reticulum biology (XBP1, etc.) and genes involved in Th17 T
regulatory cell balance (IL23R, etc.), as well as a group of genes involved in the secondary immune response
(e.g. MHC, IL12B, etc.) The figure indicates the complexity of biological pathways and the likely interactive nature
of susceptibility alleles.

2967_Ch02.indd 66 07/07/2015 10:30

 Conclusions 67

The association of such a large number of susceptibility alleles, at so many loci, most with
low relative risks, suggests that Crohn’s disease is a polygenic complex disease rather than
an oligogenic complex disease. The inherited component of the disease is low in Crohn’s
disease and it is possible that many cases involve the cumulative interaction of several
mutant alleles. In contrast, in HSCR, the inherited component is greater, there are fewer
risk alleles or mutations, and most cases involve only one major risk mutation. In some
cases of HSCR, however, the effect of the major mutation can be modified by inheritance
of other risk alleles.

2.10 Applying Disease Models to Populations

A level of caution should be applied when considering these different models of geneti-
cally complex disease. In the analysis of populations we overlook the fact that cases are
individuals and that each case is different from others. It is quite possible that many indi-
vidual cases have only one risk allele. It is also possible that patients with the same clinical
disorder can have different risk alleles on the same gene or on different genes. The variable
severity of the disease phenotype seen in individual Crohn’s disease and HSCR cases may
be due to the particular portfolio of risk alleles inherited.
The size of the portfolio or genetic load on an individual is another consideration. Portfolios
could be built up in different ways. Within the portfolio there may be a single common
genetic variant that has a large effect in the majority of the cases or there may be a series of
rare variants. The former is predicted by the common disease common variant hypoth-
esis (CDCVH) and holds true for a disease like Alzheimer’s disease and the APOE4 allele.
The latter is predicted by the common disease rare variant hypothesis (CDRVH), which
predicts that if a disease is common in the population, the genetic risk may not be com-
mon in the population and may be due to the effect of multiple interacting risk alleles.
Both models apply in complex disease, even in the same disease where some cases are seen
as near-Mendelian and others as sporadic. LOAD illustrates the possibility of both mod-
els. HSCR, which is associated with a very significant genetic component, illustrates the
CDCVH model and most cases of Crohn’s disease illustrate CDRVH. However, we need
to be cautious applying models as each case may be different. Models are best applied to
populations with no assumption that all cases will fit within the model. One other con-
sideration is whether we are dealing with a small number of common variants or a large
number of rare variants. Both can apply and indeed they do appear to.
Returning to the idea of heritability, one concept that is important to consider in complex
disease is the idea of missing heritability. There is intense discussion of various hypotheses
and models for complex disease. Some suggest that we could create a model of heritability
whereby continuous inheritance of both genetic and environmental factors leads to a point
whereby some individuals express the trait. This is not unlike the risk portfolio suggested ear-
lier. Finally, rare alleles with small or moderate effects are likely to be missed in GWAS and it
is also possible that heritability estimates are overinflated.

Conclusions
One of the key concepts in this chapter and throughout the book is that complex diseases
are not simple and do not lend themselves to simple analysis. They are best defined using

2967_Ch02.indd 67 16/07/2015 10:29

 68 CHAPTER 2 Defining Complex Disease

the modified definition of Haines and Pericak-Vance (1998) as those where one or more
genetic variations (alleles) may increase or decrease the risk of a trait or disease. In this
book we are discussing disease. In complex disease we exchange trait for disease from time
to time. This exchange is not accepted by all, but using this term occasionally enables us to
be flexible when considering overlapping diseases (syndromes) and when considering dif-
ferent subgroups within a disease population. It is therefore more useful to be flexible than
proscriptive in the application of terminology. Chromosomal and Mendelian diseases are
not genetically complex diseases; however, there are some gray areas where simple defini-
tions cannot be so easily applied.
Dismantling the definition produces a clearer understanding of what this book is all about.
The most important concept throughout the book is that the inheritance of an allele at a
locus increases or reduces the risk of the disease. This means that the allele is neither nec-
essary nor sufficient to cause the disease. Inheritance of an allele is simply associated with
either an elevated or a reduced risk. In this definition, the possibility of protective as well
as susceptibility alleles is considered. Not all risk alleles are associated with the occurrence
or initial pathogenesis of a disease. Alleles may be associated with the risk of a specific dis-
ease subgroup within a population, e.g. severity defined by age of onset, disease progres-
sion (including morbidity and mortality), or response to therapy. This illustrates how not
all genetic associations can be interpreted as tags (or markers) for disease causing genes.
Complex diseases are complex and we can consider three potential models: monogenic,
involving a single gene, but not conforming to a Mendelian pattern, oligogenic involving
several, but a limited group of genes, and polygenic involving a larger group. These models
clearly overlap, but it is still useful to categorize disease like this if possible. The latter two
models illustrate not only the basic concepts of the genetics of complex disease, but also
indicate why we seek to identify risk alleles in complex diseases. Understanding the basic
principles set out here is essential in order to understand current research and the poten-
tial applications of genetic investigations such as those described in this book in modern
medical practice now and in the future.

2967_Ch02.indd 68 07/07/2015 10:30

 Further Reading 69

Further Reading
Books relationship between the genetics of Crohn’s
Collins F (2010) The Language of Life: DNA disease and the developing understanding of
and the Revolution in Personalised Medicine. the disease pathogenesis that has followed from
Harper-Collins. these genetic studies.
Haines JL & Pericak-Vance MA (1998) Collins FS & McKusick VA (2001)
Overview of mapping common and genetically Implications of the human genome project for
complex human disease traits. In: Approaches medical science. JAMA 285:540–544. This is a
to Gene Mapping in Complex Human Disease very important review that provides the reader
(Haines JL & Pericak-Vance MA eds), pp 1–16. with a guide to and an illustration of the poten-
Wiley. This book provides the best definition tial for understanding complex disease in the
of complex disease and is a good starter text on immediate post-genome period.
major issues in complex disease. Cooney R, Baker J, Brain O et  al. (2010)
Lewis R (2011) Human Genetics – The Basics. NOD2 stimulation induces autophagy in den-
Routledge. dritic cells influencing bacterial handling and
antigen presentation. Nat Med 16:90–97. This
Parham P (2009) The Immune System, 3rd ed.
paper relates genotype to phenotype – an essen-
Garland Science. This is an excellent textbook
tial next step in complex disease.
providing a good background on human immu-
nology and some examples of immunogenetics. Cuthbert AP, Fisher SA, Muddassar MM
et al. (2002) The contribution of NOD2 gene
Strachan T & Read AP (2011) Human
mutations to the risk and site of disease in
Molecular Genetics, 4th ed. Garland Science.
inflammatory bowel disease. Gastroenterology
This is an excellent basic genetics textbook that
122:867–874. This is a good study for students
goes into more depth on many of issues in this
to look at for critical analysis and to understand
chapter, especially in chapters 2 and 3 and also
basic pre-GWAS association work in Crohn’s
in chapters 15 and 16.
disease.
Evans RM, Hui S, Perkins A et  al. (2004)
Articles Cholesterol and APOE genotype interact
Amiel J & Lyonnet S (2001) Hirschsprung to influence Alzheimer disease progression.
disease, associated syndromes and genetics: a Neurology 62:1869–1871.
review. J Med Genet 38:729–739. This is an Franke A, McGovern DP, Barrett JC et  al.
excellent review of HSCR and, together with (2010) Genome-wide meta-analysis increases
Puffenberger et  al. (1994), provides a basic to 71 the number of confirmed Crohn’s disease
starting point for students investigating this susceptibility loci. Nat Genet 42:1118–1125.
disease. This paper introduces meta-analysis and sum-
Beutler E, Felitti VJ, Koziol JA et  al. (2002) marizes a series of GWAS of Crohn’s disease.
Penetrance of 845G-A (C282Y) HFE heredi- Henderson P & Satsangi J (2011) Genes in
tary haemachromatosis mutation in the USA. inflammatory bowel disease: lessons from com-
Lancet 359:211–218. This paper is a landmark plex diseases. Clin Med 11:8–10.
paper on HFE and raises a number of impor- Hugot J-P, Chamaillard M, Zouali H et  al.
tant questions. (2001) Association of NOD2 leucine-rich
Brown MA, Pile KD, Kennedy LG et al. (1996) repeat variants with susceptibility to Crohn’s
HLA class I associations of ankylosing spon- disease. Nature 411:599–603. This, together
dylitis in the white population in the United with the paper by Ogura et al. (2001), was the
Kingdom. Ann Rheum Dis 55:268–270. This first to identify the IBD1 gene as CARD15 on
paper is the gold standard paper for ankylosing 16q12 – a major step forward in our under-
spondylitis and HLA-B*27. standing of the genetics of Crohn’s disease.
Cario E (2005). Bacterial interactions with Inohara N, Ogura Y, Fontalba A et  al.
cells of the intestinal mucosa: Toll-like recep- (2003) Host recognition of muramyl dipep-
tors and NOD2. Gut 54:1182–1193. This tide mediated through NOD2. J Biol Chem
paper provides some essential insight into the 278:5509–5512.

2967_Ch02.indd 69 07/07/2015 10:30

 70 CHAPTER 2 Defining Complex Disease

Jostins L, Ripke S, Weersma RK et al. (2012). Sennvik K, Fastbom J, Blomberg M et  al.
Host–microbe interactions have shaped the (2000) Levels of alpha- and beta-secretase
genetic architecture of inflammatory bowel dis- cleaved amyloid precursor protein in the cere-
ease. Nature 491:119–124. This paper identi- brospinal fluid of Alzheimer’s disease patients.
fies 63 risk alleles for IBD. Neurosci Lett 278:169–172.
Kullberg BJ, Ferwerda G, De Jong DJ et  al. Stokin GB, Lillo C, Falzone TL et  al. (2005)
(2008) Crohn’s disease patients homozygous for Axonopathy and transport deficits early in the
the 3020insC NOD2 mutation have a defec- pathogenesis of Alzheimer’s disease. Science
tive NOD2/TLR4 cross-tolerance to intestinal 307:1282–1288.
stimuli. Immunology 123:600–605. Stoll M, Corneliussen B, Costello CM et  al.
Lala S, Ogura Y, Osborne C et  al. (2003) (2004) Genetic variation in DLG5 is associated
Crohn’s disease and the NOD2 gene: a role for with inflammatory bowel disease. Nat Genet
paneth cells. Gastroenterology 125:47–57. 36:476–480.
Lees CW, Barrett JC, Parkes M & Satsangi J Taylor RW & Turnbull DM (2005)
(2011) New IBD genetics: common pathways Mitochondrial DNA mutations in human dis-
with other diseases. Gut 60:1739–1753. This ease. Nat Rev Genet 6:389–402. This review is
paper provides a current summary of the genet- highly informative for the student or clinician
ics of Crohn’s disease and the wider connota- trying to understand the nature of mitochon-
tions of this type of research. drial disease and the mitochondrial genome.
Lesage S, Zouali H, Cezard J-P et  al. (2002) The Thousand Genomes Consortium (2012)
CARD15/NOD2 mutational analysis and gen- An integrated map of genetic variation from
otype-phenotype correlation in 612 patients 1,092 human genomes. Nature 491: 56–65.
with inflammatory bowel disease. Am J Hum The Wellcome Trust Case Control Consortium
Genet 70:845–857. This paper opened the door (2007) Genome-wide association study of
to further exploration of the CARD15 gene 14,000 cases of seven common diseases and
in Crohn’s disease and marked an important 3,000 shared controls. Nature 447: 661–678.
development in the study of Crohn’s disease This is another landmark paper highlighting
genetics. the development and application of new tech-
Ogura Y, Bonen DK, Inohara N et al. (2001) nologies (GWAS) in complex disease research,
A frameshift mutation in NOD2 associated and it provides a mass of useful information for
with susceptibility to Crohn’s disease. Nature those studying complex disease. Together with
411:603–606. This, together with the paper the other papers on Crohn’s disease, this study
by Hugot et al. (2001), was the first to identify provides some of the basic information for stu-
the IBD1 gene as CARD15 on 16q12—a major dents studying Crohn’s disease as well as a fur-
step forward in our understanding of the genet- ther six diseases not discussed in this chapter.
ics of Crohn’s disease.
Puffenberger EG, Hosada K, Washington Online sources
SS et  al. (1994) A missense mutation of the http://hapmap.ncbi.nlm.nih.gov
endothelin-B receptor gene in multigenic This is the site of the results of the International
Hirschsprung’s disease. Cell 79:1257–1266. HapMap project.
Prince JA, Zetterberg H, Andreasen N et  al. http://pngu.mgh.harvard.edu/~purcell/plink
(2004) APOE (epsilon)4 allele is associated
with reduced cerebrospinal fluid levels of A PLINK is a freely available, open-source,
(beta) 42. Neurology 62:2116–2118. whole-genome association analysis toolset
designed for quality control and analysis of
Schreiber S, Rosenstiel P, Albrecht M et  al. GWAS data.
(2005) Genetics of Crohn disease, an arche-
typal inflammatory barrier disease. Nat Rev http://www.genet.sickkids.on.ca/StatisticsPage.
Genet 6:376–388. This is an excellent review html
of the genetics of Crohn’s disease. The informa- http://www.mousebook.org/mousebook-­
tion and discussion provides insight not only catalogs/imprinting-resource
into Crohn’s disease, but is also a useful model This is an essential database for mouse imprint
to understand other similar conditions. genes and current research on imprinting.

2967_Ch02.indd 70 07/07/2015 10:30

 Further Reading 71

http://www.ncbi.nlm.nih.gov/omim 1000 genome project provided a validation

This is an essential site for updates and informa- map including 38 million SNPs in 2012 (refer-
tion on genetic disease of all types. ence above).
http://www.ncbi.nlm.nih.gov/projects/SNP/ http://www.ncbi.nlm.nih.gov/SNP
snp_summary.cgi This an essential site for updates on SNPs.
This site provides information about validated
reference SNPs clusters in the genome. The

2967_Ch02.indd 71 07/07/2015 10:30

2697_Prelims.indd 2 09/07/2015 09:35
 CHAPTER
3
How to Investigate Complex
Disease Genetics

In this chapter, we will consider the questions of how to identify risk alleles in genetically
complex disease, starting with the basic knowledge that informs study design and the dif-
ferent options that are available to us. We will look at study design from the past to the
present day and consider some of the advances. Having already seen how complex diseases
do not lend themselves to simple analysis and having been made aware in Chapter 2 that
we are dealing with mutations and polymorphisms that increase or reduce the risk of a
disease, it is possible to see how hard this task may be. In addition, it is also clear that there
may be complex interactions between different risk alleles, and between risk alleles and the
environment. Altogether this adds up to a difficult task and yet it is one that has attracted
considerable attention in medical research. We shall see how study design has advanced to
reach previously impossible heights.

3.1 Planning Stage 1: Gathering the Basic

Knowledge
Before we investigate a disease we need to know what is possible. For example, rare dis-
eases do not lend themselves to single-center studies, whereas common diseases may. It
is difficult to perform genome-wide studies on small groups of samples and small sample

2967_Ch03.indd 73 07/07/2015 10:32

 74 CHAPTER 3 How to Investigate Complex Disease Genetics

sizes do not facilitate the identification of low-risk alleles. Good planning begins with a
few simple questions:
• How common is the disease?
• What is the evidence for a genetic component to the disease?
• What is known about the disease pathology that may indicate associations or linkage
with specific biological pathways?

Incidence and prevalence are measures of how common a disease is

It is essential before starting any study to be aware of the frequency of the disease within
the population that we intend to study (Box 3.1). Disease frequency is normally measured
as either incidence or prevalence. These two terms are very different (Figure 3.1).
Incidence
Incidence refers to the number of cases of a disease within a defined time period. Usually,
incidence is measured as the number of new cases per year. However, for rare diseases, inci-
dence may be measured over a longer period in order to obtain accurate figures. Incidence
deals with new cases only and excludes all cases reported outside the time period.

BOX 3.1 INCIDENCE, PREVALENCE, AND FAMILY RISK RATIO (λ)

Incidence can be calculated from the simple equation:

number of new cases within a given time period

Incidence =
population size

Thus, in a population of 1000 with 10 new cases per year the incidence = 1/100.
Prevalence can be calculated from the simple equation:

number of cases in the population at time x

Prevalence =
population size

Thus, in a population of 1000 with 100 cases at time x, the prevalence = 1/10.

Family risk ratio (λ) can be calculated for different family members including siblings,
children or by sex (females/males). The risk ratio for siblings (λ) can be calculated from
the simple equation:

incidence or prevalence in the siblings of affected individuals

λ =
in
ncidence in the population

Thus, if the incidence in siblings is 1/50 and the incidence in the population is 1/1000,
then λ = 0.02/0.001 = 20. Unless tested in very large populations, it is appropriate to
consider more than one estimate of λ and quote a range. When calculating this value for
the first time, we need to be aware of the problems associated with small numbers, which
often lead to high estimates that may prove false on further scrutiny.

2967_Ch03.indd 74 07/07/2015 10:32

 Planning Stage 1: Gathering the Basic Knowledge 75

Jan April Aug Dec

Incidence: number of NEW cases over a set period of time

Jan April Aug Dec

Prevalence: number of cases at a specific time

Figure 3.1: Incidence and prevalence. The gray dots represent individual cases. The horizontal arrows are
time-lines. The figure illustrates the difference between incidence and prevalence. Incidence is a measure of
the number of new cases in a set period of time (often 1 year). Prevalence refers to the total number of cases at
a particular time. Incidence and prevalence can be very different or very similar depending on the severity of
disease (prognosis). When the disease is severe, survival times can be low and prevalence will match incidence;
where survival times or prognosis is good, then prevalence is often greater than incidence.

Prevalence
Prevalence refers to the number of cases at a given time point. On a specific date the num-
ber of reported cases may be 1000. This is different from incidence because it includes all
cases irrespective of when they were reported.

Incidence and prevalence can be very different or very similar

depending on the prognosis for the disease
In diseases with a poor prognosis, the survival time from diagnosis may be very short and
therefore the two measurements may be very similar. However, where the prognosis is
favorable, the survival time from diagnosis may be very long and the two measurements
will reflect this. For example, if the estimated survival time from diagnosis for a disease
is 20 years and the number of new cases per year (incidence) is 100, then at a given time
point we would expect there to be up to 2000 cases in the study population. Thus prog-
nosis is an important factor to be aware of when planning studies. There are many reasons
for this. Recruitment may be difficult where prognosis is poor and, even where incidence
is high, the poor prognosis means there is less time for sample collection and recruitment.
There are also ethical issues regarding recruitment for all studies that may be exacerbated
in these circumstances.

2967_Ch03.indd 75 07/07/2015 10:32

 76 CHAPTER 3 How to Investigate Complex Disease Genetics

Incidence and prevalence of disease may vary in different

populations
Population variation can be due to either genetic or environmental factors. For example,
malaria is rarely seen in the UK. The reasons for the virtual absence of malaria in the UK
are obvious and are clearly environmental. This example can be applied to many infectious
diseases where the pathogen is found only at a specific location or where other environ-
mental factors (e.g. clean water) play a role. Some geographic differences in incidence
and prevalence are due to genetic variation between populations. For example, the disease
autoimmune hepatitis (AIH) occurs in many different countries and is associated with
specific human leukocyte antigen (HLA) alleles (discussed in Chapter 6). Interestingly,
in the UK (and USA) there are early- and late-onset peaks for AIH within the popula-
tion, whereas in Japan the disease is mainly seen in older members of the population. The
difference between the UK/USA AIH and Japanese AIH populations is the presence of
a specific HLA haplotype (HLA 8.1) that is seen in the former populations, but rarely if
ever seen in the Japanese population. Such genetic differences are common between popu-
lations and these may be very important in the survival of the species (Figure 3.2). Other
diseases such as schizophrenia are seen at similar frequencies across the globe, and though
the frequencies are slightly higher in some developed compared with some developing
countries, this difference is not as big as expected. Also contrary to popular belief, it has
been shown by Bhugra (2005) that the illness is not more common in men than women,
as some have reported.

What is the evidence for a genetic component to the disease?

It is an essential early step to gather information on the potential for a genetic component
in the disease under study. There are several simple questions that can be asked that will
inform the study design. The most important of these is to find out whether there are

100
UK

80 France

USA
60
Frequency

40

20

X-positive X-negative

Figure 3.2: Incidence and prevalence in different populations. The figure illustrates three populations (e.g. UK,
France, and USA) with different frequencies of X disease. These values could have been measured as either incidence
(number of new cases in a set period) or prevalence (total number of cases at a time). The important point is that
incidence and prevalence for a given trait may vary between populations. The further apart these populations are
located geographically, the greater these differences are likely to be. Thus, the greatest difference in the frequency of
X disease is between the USA and the UK, while France, which is closer to the UK, has a smaller difference.

2967_Ch03.indd 76 07/07/2015 10:32

 Planning Stage 1: Gathering the Basic Knowledge 77

any informative (i.e. multicase) families with the disease. At this stage it is also important
to remember missing heritability. This reminds us that not all heritability is due to our
genes—the proportion of our heritability that is not determined by our genes is referred to
as missing heritability. The extent to which missing heritability occurs varies between dis-
eases. Inflated values for some of the data collected in the tests referred to in this book (see
Chapter 4 in particular) are the result of this missing heritability. For example, a sibling
relative risk of 50 may not all be attributed to sharing of the same genetic variation—the
shared environment can also play a significant role.
Family information
Any good geneticist will always look for informative families. To be informative, a family
must have more than one affected member (Figure 3.3). If there are enough informative
families, family studies can be performed. Linkage analysis based on multicase families
has been and still is considered the gold standard for genetic studies; however, multicase
families are rare in complex disease. In the absence of families, affected sibling pairs can be
studied. However, even if there are families or large numbers of sibling pairs and linkage
analysis is to be performed, the non-familial (sporadic) cases should not be overlooked.
Linkage analysis in families and sibling pairs is mostly parametric linkage analysis. This
means the parameters have to be known at the outset. One of these parameters is the pat-
tern of inheritance. Therefore, this suits Mendelian disease, but not complex disease where
the key parameter, i.e. pattern of inheritance, is by definition unknown. Non-parametric
linkage analysis can be used instead for linkage studies in complex disease.
It is important to remember that there are a number of caveats with respect to families in
genetic studies. Familial history is not always a sign of a genetic effect. Families share envi-
ronments, which can have a strong effect on susceptibility to a variety of different diseases.
The best example of this can be seen in an analysis of the likelihood of attending medical
school in the USA. This study found attendance was an autosomal recessive trait, when in

(a) Male Female

Affected

Unaffected

(b)

Figure 3.3: Informative families. This simple comparison between two families (a and b) illustrates the
importance of there being multiple cases in a family if linkage analysis is to be applied. Here, family (a) is
informative and family (b) is uninformative because for linkage there needs to be more than one affected family
member in each generation. In practice, the picture may be more complicated and the decision of whether or
not to use families may rest on the actual number of affected cases and the number of cases in each generation.

2967_Ch03.indd 77 07/07/2015 10:32

 78 CHAPTER 3 How to Investigate Complex Disease Genetics

fact the real impacting factors were social status and parental income. In a sense, the trait
has a significant heritable component, but that heritability is not genetic. The issue of the
term heritability has been extensively discussed in Chapter 2.
The second caveat regarding family studies is that genetic risk in families may be attribut-
able to different risk alleles compared with those involved in sporadic disease. However,
caveats aside, families do represent the genetic high ground in such studies and for many
diseases, especially near-Mendelian and oligogenic complex disease, identifying risk
alleles in families has led to significant findings. These findings have in some, but not
all cases, also led to findings in sporadic non-familial cases. Linkage analysis is no lon-
ger restricted to accessing single genes or chromosomal regions as in the past. In the
present decade, linkage studies can include the whole genome. Indeed, many genome-
wide linkage studies (GWLS) have been performed. However, this does require large
family pools. It is also important to remember that not all families are informative
(e.g. Figure 3.3b) and even where there are multiple cases in diseases with late onset or
severe prognosis, recruiting family members for studies is often difficult.
Relative risk (λ)
Apart from using families in linkage analysis, we can perform a few basic tests to provide
an indication of the size of the heritable component of a disease. One such test is based
on relative risk, which can be calculated by comparing the known incidence or prevalence
in families, with the incidence or prevalence in the population as a whole. The risk (λ) can
be calculated for specific family members, for example λs is a measure of the risk for sib-
lings of the affected (and this is the most commonly calculated risk). The risk may be dif-
ferent in different family members. Some diseases are known to be more common in the
female members of the population, such as in many, though not all, autoimmune diseases.
In these cases, calculating the risk for daughters of affected mothers may be more appro-
priate. The λ value gives a good, but crude measure of the relative risk within a population
and in practice when looking at published data it is always better to quote ranges from sev-
eral studies rather than from individual studies. Values quoted for Mendelian disease may
be as high as several thousand for dominant conditions and several hundred for recessive
diseases. Figures for complex diseases vary from zero to the low hundreds (see Table 3.1).

Table 3.1 Sibling relative risks for different diseases.

Disease Type Relative risk (λ)

Huntington’s disease Mendelian dominant 5000
Cystic fibrosis Mendelian recessive 500
Hirschsprung’s disease Oligogenic complex ~185
Crohn’s disease Polygenic complex 13–36
Ulcerative colitis Polygenic complex 7–17
Primary biliary cirrhosis Polygenic complex 10.5 (daughters of affected
mothers = 58)
Type 1 diabetes Polygenic complex ~15
Multiple sclerosis Polygenic complex ~50
Schizophrenia Polygenic complex ~10

2967_Ch03.indd 78 07/07/2015 10:32

 Planning Stage 1: Gathering the Basic Knowledge 79

It is also important to remember that this is a measure of heritability and genetic variation
may only make up part of this.
Twin studies
Twins represent a very useful study group in genetics. Twins may be either identical or
non-identical. Identical twins are the product of a single fertilization and share the same
genome, i.e. they are monozygotic twins. Non-identical twins occur when two fertil-
izations of different gametes occur at approximately the same time giving rise to two
individuals who, though born at the same time and of the same parents, are unlikely to
have a larger degree of shared genetic identity than any two other offspring of the same
parents conceived on different occasions (Figure 3.4). Non-identical twins are referred to
as dizygotic twins.
Twins can provide very useful information about the genetic component for a disease.
Simple comparison of disease concordance rates for monozygotic versus dizygotic twins
indicates the extent to which the genome plays a role in disease. Twin data has to be han-
dled with caution because concordance data for large disease populations is more reliable

Dizygotic twins: two separate fertilization events

n
2n
n

n
2n
n

Monozygotic twins: a single fertilization event 2n

followed by early division of the embryo

n 2n 2n
2n
n

2n

Figure 3.4: Monozygotic and dizygotic twins. Twinning occurs in one of two ways. Fraternal (non-identical
or dizygotic) twins occur when two eggs are independently fertilized at approximately the same time. Dizygotic
twins develop in the womb at the same time, but have separate embryonic membranes. Dizygotic twins have
the same level of genetic identity to each other as their siblings. Monozygotic twins arise from the same single
fertilization event and are produced by the division of the embryo very early in gestation. The illustration shows
this division after the two-cell stage, but it may occur anytime before, such as splitting at the morula or in the
early blastocyst stage. Monozygotic twins are essentially genetically identical.

2967_Ch03.indd 79 07/07/2015 10:32

 80 CHAPTER 3 How to Investigate Complex Disease Genetics

Table 3.2 Concordance data for monozygotic and dizygotic twins.

Concordance (%)
Disease Dizygotic twins Monozygotic twins
Type 1 diabetes 15  30–40
Crohn’s disease 7  37
Type 2 diabetes 10  90
Rheumatoid arthritis   3.6  15.4
Multiple sclerosis   5.4  25.3
Bipolar disorder   6  43
Autism   0–21 ~60
Schizophrenia 4  15–33

than for small patient populations. However, large numbers of twins are unlikely to be
available in all but the most common of complex diseases because twins are themselves
quite rare. Approximately one in every 200 pregnancies give rise to twins and most of
these are dizygotic twins. Some examples of concordance data for different diseases are
given in Table 3.2 and elsewhere throughout this book.

What is known about the disease pathology?

It is often assumed that because diseases are diagnosed according to the presence of a series
of signs and symptoms we have a clear understanding of all aspects of the pathology of
these diseases. Nothing could be further from the truth. Disease pathology reflects the
downstream events (Figure 3.5). Signs and symptoms arise because there is some form
of malfunction. However, the human body can tolerate significant trauma before organs
and structures fail. Signs and symptoms are in many cases endpoints of at least the early
phases of disease. In practice, this means that there are gaps in our knowledge of disease
pathology and much of what we do know is based on hypotheses generated from research.
Previous studies
When it comes to asking questions about disease pathology we need to consider the qual-
ity of current knowledge and the gaps in the jigsaw. In the 1990s and early years of the
new millennium, studies of complex disease were mostly based on hypothesis-driven
research investigating small numbers of candidate genes in favored regions. In the present
decade, planners have cast aside the restrictions imposed by hypotheses and have investi-
gated the whole genome in a hypothesis-free (or hypothesis-non-constrained) way. Some
argue that there is still a hypothesis, but it is just very broad, i.e. there is a disease allele or
alleles out there somewhere. However, most prefer the idea that these are hypothesis-free
studies, or at least the initial stages are, though replication and fine mapping may not be.
This change in approach reflects the introduction of new technologies in genetics of com-
plex disease. Though GWLS at low resolution in which microsatellites were used as markers
were performed in 1994 and 1996, these were rare exceptions and modern genome-wide
association studies (GWAS) and GWLS did not start appearing until after 2005. The
change in favored choice of study reflects years of frustration regarding results from earlier
traditional candidate gene-based association studies that were often contradictory. One

2967_Ch03.indd 80 07/07/2015 10:32

 Planning Stage 1: Gathering the Basic Knowledge 81

Unknown A Unknown B

Genetics
known factor C

Healthy Diseased

Figure 3.5: Disease pathology. The figure illustrates a healthy (light gray) and an unhealthy, diseased liver
(dark gray with scarring). The figure shows that the many of the factors that cause early disease are often
unknown to us, and reminds us that we tend to see patients (represented by the organs here) with specific signs
and symptoms when they present at a clinic late in disease progression. This means the opportunity to gain
knowledge of the early biological processes through which the disease arises are lost to the medical community,
preventing earlier interaction. However, our inherited genes do not change and knowledge of the risk alleles,
shown as factor C (the known factor), in complex disease is potentially the key to unraveling this conundrum.

group would report a significant association with a specific allele and another would fail to
replicate the original findings. This happened because many studies were poorly designed
using small numbers, poorly matched or inappropriate controls, and badly defined patient
groups. In general, much of this has now been dealt with. Large-size, multicenter studies
using well-defined patient and control groups have been able to identify significant risk
alleles for many diseases and many of these reports have been replicated. However, studies
still report weak associations that are not replicated by others. The reasons why this still
happens vary. Some studies are still poorly designed, but in many cases the differences
found reflect the different populations studied.
It is essential to understand the history of these studies in order to critically evaluate the
literature prior to undertaking any study of the genetics of a complex disease. Questions
about quality and reliability of previous studies are crucial if we intend to build on pre-
vious research. It is also important not to waste research resources by simply repeating
previous studies unless there is a bona fide reason to do so. Information regarding previ-
ous studies and potential risk haplotypes is essential in study planning. There may be very
little mileage in reconsidering serially confirmed associations and it may be more fruitful
to look at previously ignored chromosomal regions or genes.
Disease pathology as a means to identifying candidate pathways
Though our knowledge of early disease pathology is often incomplete there is a great deal
of information about disease pathology in the literature. Our task is to sift through the data
and identify key biological pathways that, when stressed, may contribute to the disease
expression. Biological pathways are the key to unpacking disease pathology and this process
is itself the key to identifying potential candidate genes in complex disease. For example,
pathways associated with bone formation and with inflammation may be critical in rheu-
matoid arthritis. Where there are well-established hypotheses, it may be appropriate to

2967_Ch03.indd 81 07/07/2015 10:32

 82 CHAPTER 3 How to Investigate Complex Disease Genetics

consider the constituent pathways that have led to the development of that hypothesis. For
example, early hypotheses suggested that Crohn’s disease may arise from either an abnor-
mal immune inflammatory response to gut bacteria or as a food allergy. Recent evidence
favors the former and indeed appropriate risk alleles have been identified that indicate the
immune response to gut bacteria is a key process in Crohn’s disease. Not all disease-related
genes are involved in immunity. Genes impacting on lipid metabolism are known to be
important in coronary artery disease and some of the same genes may also influence risk in
Alzheimer’s disease. In neurological diseases and some mental disorders (bipolar disorder
and schizophrenia), genetic associations have been identified with genes such as DISC1,
GRM7, and GABRB1, all of which are important in brain cell activity and neurotransmis-
sion (especially GABRB1). A more detailed consideration of these is given in Chapter 4.
When considering the pathology of a specific disease we should also consider what is
known about the pathology of other similar diseases. Many diseases are grouped into syn-
dromes or collections. Sharing of susceptibility alleles between diseases is not unusual, and
therefore looking at related diseases with similar signs and symptoms may help to identify
important pathways. A primary example of the latter is autoimmune disease where many
of the major autoimmune diseases have associations with the HLA 8.1 haplotype or with
members of the HLA DRB1*04 family of alleles. Examples include type 1 diabetes mel-
litus, the autoimmune liver diseases, AIH and primary sclerosing cholangitis, rheumatoid
arthritis, and systemic lupus erythematosus (SLE).

Before we get down to the hard business of study planning there are
one or two other questions that it is important to ask
The remaining information required for planning a study is illustrated in the organiza-
tional chart Figure 3.6 and includes:
• What is the average (mean) age of onset of the disease?
• Are there any other centers investigating the genetics of the disease?
• What is the likelihood of obtaining research funding to study the disease?
Age of onset
Age of onset is an important statistic in studying any disease. Working with children
presents a number of moral and ethical issues (discussed in Chapter 11). However, with
young families, recruiting parents and other family members for studies can be easier than
working with older families where family members are more likely to be diversely spread
and less likely to be in contact. In many cases, early-onset diseases are also more likely to
have a Mendelian or near-Mendelian pattern of inheritance. However, there are always
exceptions to every rule and several common forms of complex disease, such as autism,
attention deficit hyperactivity disorder (ADHD), childhood obesity, and asthma, are all
complex disorders that present in childhood.
Other centers
It is essential to consider the competition when planning studies. There is very little to be
gained from competing with a rival who has access to larger patient numbers and better
resources. It is more useful in such a scenario to collaborate. However, collaboration is not
always possible or appropriate. Awareness of rivals can be very useful in planning. There
are usually several ways to approach key questions and taking an alternative approach
may be very fruitful. For example, rather than looking for genetic associations with the
disease per se, it may be more appropriate to consider specific subgroups within the disease

2967_Ch03.indd 82 07/07/2015 10:32

 Planning Stage 1: Gathering the Basic Knowledge 83

Key information

Heritable
component

Sibling
Twin Family
relative
studies studies
risk (λ)

Age of onset
incidence

Current
knowledge

Competitors

Quality of
diagnosis

Clinical
variables

Response to Clinical Overlap with

treatment subgroups other diseases

Funding
options

Preliminary – Advanced
“pump priming” major funding

Figure 3.6: Planning a study: things to consider before proceeding. The organization chart illustrates the key
points that need to be considered before commencing with the detailed planning of a study. Most are common
sense, but it is important to consider each of these. This can prevent misuse of time and resources, and also lead
to new approaches to a particular disease. For example, based on the information on subgroups, we may choose
to investigate one particular group within a disease cohort rather than all cases.

population. Looking at disease phenotype is particularly important in infectious disease

where transmission is not genetically determined, but post-infection outcome or response
to therapy is.
Research funding
Research does not fund itself. Sources of funding are available, but are limited. Applications
to major bodies require a considerable effort in terms of planning and writing, and are
frequently unsuccessful. The process is competitive and is particularly difficult for those
starting out in a career in medical science, whether they are clinical or non-clinical careers.
However, the reward for perseverance in applications is worthwhile and the fruits of well-
planned studies can be significant. We need to consider where the funding will come
from when planning. Most countries have both central sources (often government funds)
and also both large and small charities capable of funding different research projects for

2967_Ch03.indd 83 07/07/2015 10:32

 84 CHAPTER 3 How to Investigate Complex Disease Genetics

particular diseases. Smaller charities are particularly helpful in funding short, pump-
priming projects as a first step in a research program.

3.2 Planning Stage 2: Choosing a Strategy

Having considered all of the above, we are ready to put together a plan to study the genet-
ics of a complex disease. The choices we make at this stage are dependent on the answers
to the questions above. In terms of overall strategy, there are two basic methods—each
with two variations.

Two basic strategies for identifying risk alleles in complex disease

These two strategies—linkage and association analysis—are discussed throughout this
book. Linkage has traditionally been seen as the gold standard for genetic studies, but is
dependent on the availability of multicase families. Even in complex diseases with sig-
nificant numbers of such families, the majority of cases are non-familial (sporadic) cases.
Therefore, if our primary strategy is to collect and study families through linkage analysis
we would hope to consider the sporadic cases at a later time. In the absence of families, our
primary strategy would be association analysis. In this instance non-familial cases would
be analyzed. Samples from affected family members may be collected, but only one fam-
ily member for each such family (usually the first diagnosed case, referred to as the index
case) would be used in the final analysis to avoid introducing bias.

In terms of the history of genetic studies in complex disease there are

two main periods: pre- and post-genome
Prior to the publication of the draft Human Genome Map (HGM) in 2001, studies were
either based on linkage or association analysis just as at present; however, the precise loca-
tion of the target genes was mostly unknown.
Linkage analysis in the pre-genome era
Linkage analysis identifies chromosomal regions that co-segregate with the disease.
Linkage analysis establishes the likelihood of a physical link between an allele at a marker
locus and a disease. The degree of linkage is expressed as the log (to base 10) of odds for
linkage between the marker and the disease, and is referred to as the LOD score. The LOD
score is discussed in detail in Chapter 5. Linkage has been traditionally used in Mendelian
monogenic disease and has been very successful. Penetrance is a factor in linkage analysis.
Linkage analysis is more likely to be successful if penetrance is high. Even in complex dis-
ease, where there are occasional familial cases, linkage can be useful provided penetrance
is high. Linkage analysis does not identify a specific gene or group of genes or establish
causal links; it simply establishes the statistical likelihood of a link with a marker or allele
in a region of a chromosome. As the location of most human genes was unknown prior to
2001, researchers would have to best-guess gene locations based on whatever information
they had available at the time. There are positional and positional-independent routes to
identify disease genes. Positional studies use positional cloning to identify a disease gene
based on its approximate chromosomal location. There are several steps in this process.
First, a candidate region needs to be identified, clones from the region are produced, and
all of the genes in the region identified and prioritized for mutation screening (Figure 3.7).
This was a considerable undertaking prior to the publication of HGM, and positional

2967_Ch03.indd 84 07/07/2015 10:32

 Planning Stage 2: Choosing a Strategy 85

Prevalence/incidence

Heritability data: twins:

families: sibling relative risk (λ)

Segregation analysis:
estimate of number of
risk alleles

Linkage analysis Positional cloning

Association analysis

Identify biological
factors in disease

Figure 3.7: Logical sequence for disease gene identification prior to the HGM. This figure illustrates the
normal sequence of events proposed for studies designed to identify disease genes prior to the publication of
the HGM. However this sequence is more suitable for Mendelian disease than for complex diseases. Segregation
analysis has rarely been performed in complex diseases. Indeed, many complex diseases have never even been
subject to linkage analysis. The sequence that suggests association analysis should only be performed after
linkage analysis is not practical in many complex diseases. This “logical” sequence suits the rare, highly penetrant
near-Mendelian complex diseases, but does not work with diseases which are mostly sporadic or where
transmission is horizontal, e.g. many infectious diseases. Only when there are sufficient family cases can linkage
be performed. Nevertheless, this approach has been used in a number of diseases at least once (e.g. type 1
diabetes). This sequence is especially useful where there are a reasonable number of cases in the collection
(i.e. the disease or trait is common) and when possible it remains a logical sequence.

cloning was mostly applied to Mendelian disease and rarely (if at all) used in complex
disease. Today, in the post-genome period, the precise location of most human genes is
known and the first steps in this method (cloning) have therefore become redundant. In
addition, since the availability of the SNP Map (http://www.ncbi.nlm.nih.gov/SNP/) and
the HapMap (http://hapmap.ncbi.nlm.nih.gov/) the majority of the mutations and poly-
morphisms are also known, so the whole process of testing candidate genes for relationships
with disease susceptibility can be short circuited.
The problem with linkage analysis is that it requires informative families. Unfortunately,
many complex disease cases are sporadic cases, without a family history, and linkage would
not or could not be applied, and the number of familial cases available for linkage analysis
would be insufficient to ensure adequate statistical power in any analysis undertaken.
Even where there are significant numbers of family members, such as in cardiovascular
disease, diabetes, and depression, cases do not cluster in a Mendelian fashion. This is partly
because most common diseases are complex and phenotype is determined by the interac-
tion of several factors, both genetic and environmental. Individual genetic variations often
have relatively small effects on the disease risk and are more likely to be detected in large-
scale case control association studies.
Association analysis in the pre-genome era
Association analysis is perhaps the most important method used to identify risk alleles in
complex disease. Case control association analysis tests for correlation between the inheritance

2967_Ch03.indd 85 07/07/2015 10:32

 86 CHAPTER 3 How to Investigate Complex Disease Genetics

of one or more genetic markers or alleles and a disease. However, unlike linkage analysis,
association analysis does not establish a physical link. This approach compares the two groups
from the same population. These groups are usually (though not always) cases (patients) and
healthy controls. The history of association analysis is one of mixed fortunes and to a conven-
tional geneticist who deals with Mendelian diseases association analysis has had a very bad
reputation. However, this bad reputation is not entirely deserved, as discussed below.
Some authors refer to the common disease common variant hypothesis (CDCVH) in con-
sidering association studies and even go so far as to suggest it is the foundation for associa-
tion studies, especially GWAS. It may be useful to consider this idea before embarking on
an association study. The hypothesis predicts that common disease-causing alleles will be
found in all human populations with specific diseases.
Studies starting in the 1960s and 1970s began to identify both genetic linkage and genetic
associations in complex disease. One of the major areas for association analysis was the
human major histocompatibility complex (MHC) on chromosome 6p21.3, with the first
association reported in 1967. The clinical need to HLA-match patients requiring either
bone marrow or kidney transplants revealed an abundance of certain HLA phenotypes
in patient groups with specific presenting diagnoses. Many of these genetic associations,
though not all, have stood the test of time. We should note here that it is important to
acknowledge that there have been significant developments in the methods used for HLA
typing with a major shift from what was low (slow)-resolution phenotyping to very high
(rapid)-resolution genotyping (see Chapter 6). It is also fair to say that the majority of
these genetic associations were poorly understood at the time and to some extent geneti-
cists are still cautious when considering the human MHC.
Expectations in the earlier association studies were quite different from those of today. A
study reporting an association with an increased risk [measured as odds ratio (OR) or rela-
tive risk] of less than 3 would be unusual. Almost all studies expected large risks and studies
were often small. Many of the statistical principles considered in Chapter 5 were not con-
sidered in those early studies. Consequently, association analysis developed the bad reputa-
tion referred to above, but some good things have arisen from this bad reputation. The high
number of inconsistencies between studies created a platform from which statisticians were
able to critically review the past, and launch new, better guidelines for research in this area.
There are a few exceptions to the statement above, for example large OR values above 2 or
3 were reported for PTPN22 in type 2 diabetes and rheumatoid arthritis, and CARD15
in Crohn’s disease. Many of the early associations in pharmacogenetics also reported high
OR values (see Chapter 8). Other than these, early studies outside the MHC rarely found
strong associations with OR values of 3 or more. As a consequence, these studies were
more susceptible to finding false positives that proved difficult to replicate. Of course with
small studies many weak (but important) associations were also missed and for every false
positive it is almost certain there were several false negatives.
It is important when considering data to remember that OR values are rough calculations
at best and should always be considered within the range identified by the confidence
intervals (CIs). In addition, some of the early associations remain controversial even
today despite multiple studies and meta-analysis.

Transmission disequilibrium testing and family-based association testing

The transmission disequilibrium test (TDT) or family-based association test (FBAT)
are designed to detect the presence of a genetic association (albeit through linkage) between
a genetic marker and a disease or disease subgroup. The TDT was originally developed for

2967_Ch03.indd 86 07/07/2015 10:32

 Planning Stage 2: Choosing a Strategy 87

use in family-based studies. The test is based on the probability of equal transmission of
an allele or genotype from parents to affected and unaffected offspring. If the marker allele
in the group being tested is equally transmitted, then there is no disequilibrium (that is
there is equilibrium); if they are not equally transmitted, then there is transmission dis-
equilibrium. If sufficient families are tested, then the association can be statistically proven.
Consider a single gene with alleles A and B. Using heterozygous parents in a TDT test, the
genotype distribution among tested offspring should be 1:2:1, i.e. we would expect 25% to
be homozygous genotype AA, 50% to be heterozygous genotype AB, and 25% to be homo-
zygous genotype BB, if the transmission is in equilibrium. Otherwise there is transmission
disequilibrium suggesting an association with one or other allele (A or B) with the disease.
Usually the sample consists of a set of families each with four individuals, i.e. the affected
and unaffected offspring and the two parents, though in some studies trios can be used
(affected cases and parents only) and occasionally families with more than one affected
case can also be used. A variation on the original TDT is the so-called FBAT. Both TDT
and FBAT are family-based association tests and can be used in most complex diseases.
However, for many diseases they are rarely used because obtaining family material is dif-
ficult for the reasons discussed above (Figure 3.8).

A, B A, B A, B A, B

B, B A, B

A, B A, B A, B A, B

A, B B, B

Figure 3.8: TDT. The TDT test is a family-based association test. Using heterozygous parents, affected and
unaffected siblings are tested to see if the alleles are transmitted equally as expected. Unaffected siblings are not
always required for TDT. This illustration shows four families with two potential alleles A and B. All of the parents
are heterozygous AB. If the alleles are equally transmitted, then the frequency of the genotypes in the affected
offspring should be 1:2:1 in the order AA, AB, and BB, respectively, or alternatively each allele should be found at
a frequency of 50%. In this case, two of the offspring have the BB genotype and two are heterozygous AB positive
(the frequency of the B allele is 75% compared with 25% for the A allele). This suggests disequilibrium with a
preponderance of the B allele. In a larger cohort of cases it would be possible to analyze the data and be more
precise about the statistical significance of this observation.

2967_Ch03.indd 87 07/07/2015 10:32

 88 CHAPTER 3 How to Investigate Complex Disease Genetics

Choosing study strategies in the post-genome era

Linkage is and can be used when families are available. However, even when the primary
strategy is to collect and study families, a good plan would include collection of sporadic
cases for consideration at a later time. It must be stressed that linkage analysis is rare. In
the absence of families our primary strategy would be association analysis. In this instance,
non-familial cases would be collected with one affected from each family if families are
present in the case group.
One justification for studying families when the majority of cases are sporadic is that fami-
lies may have a stronger genetic predisposition than sporadic cases and hence identifying
risk alleles may be easier in families. However, there is no guarantee that the same risk
alleles will be found in familial cases and sporadic cases, but it is possible that the risk alleles
for familial disease will identify pathways in which there is polymorphism in other genes
that explains genetic susceptibility and resistance to sporadic disease. An example of such a
difference is early- and late-onset Alzheimer’s disease (EOAD and LOAD; see Chapter 4),
where EOAD is associated with genetic variation in the APP, PSEN1, and PSEN2 genes,
and LOAD is associated with genetic variation in the APOE gene (among others).
In the past, linkage was favored because early studies of complex disease based on associa-
tion analysis were given a bad press largely due to small numbers tested and the irrepro-
ducible results. This was addressed extensively in the late 1990s and at the start of the new
millennium. This subject is addressed in published reviews by Cardon & Bell (2001) and
by Colhoun et al. (2003). Association analysis is currently the strategy of choice in geneti-
cally complex disease. In particular, association studies are more appropriate for infectious
diseases where transmission is horizontal and not vertical.

Each of these two strategies has a substrategy

Having decided which of the two main strategies to apply (i.e. linkage or association or
both), it is important to gather together the information to determine our plan. Will we
use a genome-wide approach or will we be selective? If we decide to be selective, what will
we investigate? The options are:
• The whole genome.
• Genes within one or more chromosomal regions.
• Genes within one or more extended haplotypes.
• Genes identified within one or more specific biological pathways.
• A single candidate gene.
The choice of which option to take is dependent on our knowledge of previous studies,
current studies and the disease pathology.
Genome-wide studies
Genome-wide studies can be either linkage (GWLS) or association (GWAS) depending on
the availability of families. Both are mostly post-genome phenomena. The Human Genome
Mapping Project (HGMP) followed by the SNP Map and HapMap has opened up the
genome to extensive high-resolution study. It is now possible to study millions of genetic
markers, mostly single nucleotide polymorphisms (SNPs) or tagged SNPs as they are called,
but also copy number variants (CNVs), across the whole human genome (Figure 3.9). It
is also possible to use databases with haplotype data to identify common linkage patterns
between tagged SNPs and impute missing data from these databases, thereby extending the

2967_Ch03.indd 88 07/07/2015 10:32

 Planning Stage 2: Choosing a Strategy 89

Disease Control

DNA collection

All SNPs
Patients Controls

Frequency for a
single SNP allele
in patients (black)
and controls
(white)

Figure 3.9: GWAS. The illustration shows two populations of diseased patients and healthy controls. The DNA
samples for each individual (shown as small dots) in each group are collected. The frequency of a series of
SNPs (around 500,000 to 1 million plus in some studies) across the whole genome is tested in each group on a
commercial chip or platform. The frequency data for one SNP allele in each group is shown in the gray sphere at
the bottom of the figure. Data for each SNP are subject to statistical testing to determine whether there are any
statistically significant differences between them. The predominance of black dots in this illustration suggests an
association with this allele.

analysis within the target haplotypes to include more data (see Chapters 5 and 12). Impute
can be used provided there are sufficient fully genotyped samples with complete haplotypes
in a database. One can impute the missing genotypes for a set of study samples based on
the expected pattern of linkage. The method exploits linkage disequilibrium. Essentially, it
involves replacing missing data with substituted values. This means that any bias generated
by discarding data from samples that are not fully genotyped is avoided.
The level of resolution from a study can be significantly increased using impute software
to identify SNPs through known linkage disequilibrium, so that two or even four times
as many SNP genotypes can be assigned than were actually tested (Figure 3.10). Impute
requires complex computer software such as PLINK and access to good-quality databases
such as that generated by the 1000 Genomes Project. The greater the number of whole-
genome sequences put into public access systems, the greater the quality of the imputed
data will be. Of course the success of this method depends on the degree of linkage dis-
equilibrium between tagged SNPs on the target haplotypes. The beauty of genome-wide
studies, at least in the initial stages, is that they do not require a hypothesis. This allows us

2967_Ch03.indd 89 07/07/2015 10:32

 90 CHAPTER 3 How to Investigate Complex Disease Genetics

1 2 3 4 5 6

Figure 3.10: Imputation analysis. The figure illustrates six different haplotypes in six individuals numbered 1–6
for a combination of six biallelic genes illustrated as double rings. Each gene is polymorphic; five of the genes
have been genotyped. There are two possibilities for all six genes (i.e. two alleles). The alleles are illustrated
by gray shading or no shading in those that have been genotyped. The gene illustrated in black has not been
genotyped. Imputation analysis can enable the genotype of the black gene on this haplotype to be estimated
based on the known pattern of linkage disequilibrium for the surrounding genes for which this data has been
obtained. Therefore, if we know that individuals with the haplotype which runs (top to bottom) gray, white, gray,
white always carry the gray allele at the fifth position, then we can assign this allele to individual 1 at position 5. If
we were to know that all carriers of white at position 4 and gray at position 6 also carry gray at position 5, we can
also impute the gray allele for these individuals even though they have not been genotyped for this position. This
imputed data can be included in the analysis. As linkage disequilibrium is so strong in certain areas of the human
genome, imputation analysis from initial GWAS data can massively extend the studies. Thus, an initial GWAS
study based on 300,000–400,000 SNPs can generate data for up to 1 million or more markers.

look more freely at the genome and identify hitherto unthinkable genetic relationships.
This can lead to favored hypotheses being rejected or being accepted, or in extreme cir-
cumstances it can turn some favored hypotheses on their heads. In all cases it is likely to
identify new risk alleles.
Selecting this approach in an association study would depend very heavily on previous
studies and sample size (see below). In the absence of any data at all, provided sufficient
candidates could be recruited, this strategy would be a good first choice. Even when num-
bers are small, GWAS can be performed, but it is essential to understand the limits of
using these studies on small numbers. The reason GWAS are based on large numbers is
that they are required to have sufficient statistical power (see Chapter 5) to detect weak
genetic effects with a high degree of confidence. If we look at small numbers we may only
be able to detect strong associations with a reasonable degree of confidence.
If we apply a non-hypothesis-constrained approach we cannot claim to be testing a spe-
cific hypothesis, we are simply looking for associations (or linkage) with tagged SNPs or
other genetic variants that mark areas of statistical significance and therefore genetic inter-
est. The great thing about this approach is that it is hypothesis generating. Thus, from the

2967_Ch03.indd 90 07/07/2015 10:32

 Planning Stage 2: Choosing a Strategy 91

ashes of hypothesis-free GWAS a mass of data arise that can identify regions around genes
that contribute to biological systems that may be of specific importance in understanding
the pathology of the disease under study. This is very much a first phase and a first study
would require replication to validate any findings.

Studying a selected chromosomal region

This is a second-phase process and would only be undertaken where prior studies had
identified a region or regions of interest. GWAS is likely to generate interest in several
chromosomal regions. The tagged SNPs may indicate association with a specific gene,
but are more likely to identify associations across an area for further study. In this case,
high-resolution analysis using selected tagged SNPs will identify peaks of association in or
around genes with the region. However, just like initial GWAS, the findings of this type
of study should be replicated in a second cohort and the results may not identify a spe-
cific gene, but may identify a group of genes or a haplotype. Currently, the application of
GWAS to a single chromosome is not an option, as most commercial SNP chips are not
set up to do this and the cost is too great to do a single study.

Studying extended haplotypes

Haplotypes in genetic terms are groups of alleles at different loci that are inherited as a
unit. Extended haplotypes are particularly lengthy sections of DNA where multiple alleles
are inherited in specific groups. One particular feature of haplotypes is that the alleles that
compose them are found to be inherited together more often than expected by chance,
i.e. they are in linkage disequilibrium. As a result of this, extended haplotypes can be both
good and bad news. Linkage disequilibrium means that alleles of two or more genes are
found together more often than they should be if there was independent (equal) segrega-
tion. These genetic packages may contain several important genes all related to the same
pathway. For example, the human MHC contains many of the key genes involved in
T cell immunity (among other functions). When polymorphism is widespread within a
haplotype, as it is within the MHC, it can be very difficult to pinpoint the precise suscep-
tibility allele, but it is not impossible if the data set is large enough. Combining haplotype
data with GWAS data using imputation to assign HLA haplotypes has recently generated
some very interesting results in the autoimmune liver disease primary biliary cirrhosis.

Investigating pathways
Proteins almost always work in groups or pathways. Systems biology is a new approach to
biological processes that concentrates on the study of interaction of pathways. Those who
study the genetics of complex disease can learn a lot from considering the ways in which
pathways interact to form systems. If we take this approach to studying a complex dis-
ease, we would need to have a reasonable amount of prior data or a plausible hypothesis.
This is likely to be a phase 2 or phase 3 investigation based on prior knowledge. Evidence
from recent studies in Crohn’s disease indicates the value of considering processes com-
pared with considering individual genes in complex disease. The identification of nod2
(CARD15) and the autophagy gene (ATG16L1) both indicate links to pathways associated
with immune tolerance to Gram-positive commensal bacteria in the gut. Most studies of
Crohn’s disease now look for links to these and related pathways.

Investigating a single gene: candidate gene studies

In the past, most association studies considered variation at only one or two candidate
gene loci. A few considered haplotypes with specific regions based on selected candidates.
This approach is hypothesis driven and, as such, is limited to searching for associations at

2967_Ch03.indd 91 07/07/2015 10:32

 92 CHAPTER 3 How to Investigate Complex Disease Genetics

specified loci. Gene loci were selected where there was at least some potential functional
relationship between the product of the gene under investigation and the disease. The
problem with this candidate gene approach is that our knowledge of the biological rel-
evance of most human genes is limited by our lack of understanding of human biology
and disease pathology. If we take a narrow view and study only one or two genes, we will
almost certainly miss a great deal. If we cast our nets wider, perhaps studying genes in
whole pathways and systems, then there is more chance of finding significant associations.
Investigation of a single gene is less likely in the present era than it would have been 20
years ago. However, where prior studies indicate a strong relationship with a particular
gene or region close to a gene, then it may be relevant to perform high-resolution genotyp-
ing of that gene to identify polymorphisms (alleles) associated with susceptibility to the
disease under study. It may occasionally be prudent to investigate a region or haplotype
one gene at a time. This can be useful if resources are limited and it can also be helpful
to produce data for initial funding applications. Often researchers will use small funds
to pump-prime preliminary studies to gain data for larger research funding applications.
Candidate gene studies have rarely been very successful in identifying associations with dis-
eases. Most associations that were initially positive have failed to been confirmed. The reason
for this is in the selection of the candidate, which depends on the validity of the biological
hypothesis being tested, which in turn depends upon our knowledge of the disease patho-
genesis. Our knowledge of the early pathogenesis for most diseases is relatively poor; patients
present late in the disease process and identifying biological features of the early disease is
impossible. However, the good news for geneticists is that while pathology changes, the
genome does not. The future of medicine lies in early identification and prevention of dis-
ease. To achieve this goal it is necessary to identify risk alleles and use genetic studies to
inform the debate on disease pathogenesis. In 2015, this almost always means throwing the
net as wide as possible, i.e. genome-wide, at least in phase 1 of the study and coming back to
regions, pathways, haplotypes, and candidate genes later in the study as our knowledge base
progresses and we focus on the real disease-causing polymorphisms (Table 3.3).

Table 3.3 Current organization for genetic studies in complex disease including options for genome-
wide linkage and genome-wide association analysis.

Linkage Association
Collect families Collect cases
Genome-wide studies (GWLS) Genome-wide studies (GWAS)
Replication Replication
Chromosomal regions Chromosomal regions
High resolution High resolution
Replication Replication
Extended haplotype(s) Extended haplotype(s)
High resolution High resolution
Target gene Target gene

Biological Function Biological Function

Pathways and Systems Pathways and Systems

2967_Ch03.indd 92 07/07/2015 10:32

 Good and Bad Practice 93

3.3 Good and Bad Practice

It is important in planning to consider what constitutes good and bad practice. The pub-
lished literature on genetics of complex disease is peppered with learned text on the prob-
lems we can encounter and it is also peppered with contradictory studies. Much of the
critique is aimed at association studies. However, things have changed and association
studies have become the method of choice. Why has there been such a change of heart?
The answer lies in the field of statistics, or statistical genetics, but what are the problems?
• Sample size
• Case selection
• Controls selection
• Sampling errors
• Statistical analysis
• SNP chip selection
• Publication bias

Accurately identifying true disease susceptibility alleles in GWAS

(and other association studies) is dependent on sample size
This is a very basic idea and is common sense. Studying larger groups produces data for
which there is a greater degree of statistical confidence. There is no doubt about the truth
of this statement. The problem arises when we look at the practical issues of recruiting
large numbers. Single centers may not have access to large numbers of patients. Indeed,
if the incidence of the disease is low, then it may take many years to recruit reasonable
numbers for a study. This does not excuse the use of small numbers, but it does to some
extent explain why small numbers have been used.
Sample size is the biggest concern in all studies because the greater the sample size, the
greater the statistical power of the study. For a standard GWAS, where between 500,000
and 1 million SNPs are tested, a sample of approximately 2000 cases and 2000 controls
are considered reasonable, though more is considered to be even better. If fewer samples
are used then confirmation of any associations in a second set is advised (replication,
see below). The 2007 Wellcome Trust Case Control Consortium (WTCCC1) study pro-
vides a good example of how to design a GWAS. In planning any study it is important
to consider the limits of the study. The numbers of cases and controls that are required
in an association study to detect risk alleles, depends on the frequency of allele and the
size of effect that the allele is expected or likely to have. This is considered in more detail
in Chapter 5. In order to determine the appropriate sample size, we can perform power
calculations.
Power calculations
In order to determine the appropriate size for a study, power calculations are recommended
during the planning stages. This can be done quite simply using freely available computer
software. The calculation in its most simple format determines the level of confidence that
can be assigned to a study based on a predicted relative risk (or OR) and given sample
size. Two variables are critical in this calculation: the size of the patient population being
studied and the expected impact of the allele (or alleles) in the test. Where the expected

2967_Ch03.indd 93 07/07/2015 10:32

 94 CHAPTER 3 How to Investigate Complex Disease Genetics

OR (a measure of the genetic impact) is high, then a smaller sample size can be used with
a reasonable degree of statistical confidence. When the expected OR is low, the sample size
needs to be greater to achieve the same level of statistical confidence. Power calculations
are particularly important in relation to negative studies that fail to find any association. It
has been suggested that the reason many studies have failed to replicate original results is
due to the lack of statistical power in the replication cohort. This is true in many studies,
but not in all. Once someone else has published their work it can be very difficult to pub-
lish results if they contradict the original findings. Indeed, there is likely to be a publica-
tion bias in favor of the initial studies. False negatives do occur, but so do false positives.
Looking at the early history of association studies we can see that the most frequently rep-
licated findings are those where the level of association is high, with an OR greater than 3.

Case selection can introduce bias into a study

Selection of both cases and controls can introduce bias into a study. To avoid recruitment
bias, case selection needs to be considered at the beginning of the study during the planning
process. Criteria for inclusion and exclusion should be as precise as possible. In syndromic
diseases, such as inflammatory bowel disease (IBD), the subgroups should be identified
clearly. The diagnostic criteria must be set out in order to create a homogeneous study popu-
lation. Cases where the diagnosis is in doubt should be excluded to reduce the likelihood
of bias. Post hoc changes should not be made. Where there are multiple cases in a family, it
is standard practice to consider only the index case. Healthy relatives from patient’s fami-
lies should also be excluded. The methods for detecting sample relatedness are discussed in
Chapter 5. Where studies are being designed for replication of previous research, the plan
should if possible meet the same criteria and employ the same clinical parameters.

It is important to consider whether we are studying a disease,

a syndrome, or a trait within a disease subgroup
We could consider allergy to be a syndrome if we are not specific about the allergen. IBD is
a syndrome even though it carries the label disease. Some disorders are syndromes. A syn-
drome can be best described as a collection of diseases with similar signs and symptoms,
but in this case different and distinct pathogenesis. The Oxford English Dictionary defines a
syndrome as a disease with concurrent symptoms. A syndrome usually involves more than
one disease, e.g. IBD, which can describe Crohn’s disease, ulcerative colitis, or a less severe
and less well defined disorder simply referred to as “inflammatory bowel disease.” There is
some overlap between different forms of IBD (Crohn’s disease and ulcerative colitis), but
the overlap is not complete. This introduces a greater variety into the sample being tested.
At the other end of the scale we may want to test patients in subgroups looking at disease
severity, progression, and response to treatment. These are all qualitative traits, whereas
clinical expression of a disease is a quantitative trait. Some consideration of these differ-
ent parameters will be given in later chapters of the book. At this stage it is important to
state clearly that we are not confined to quantitative traits, we can also consider qualitative
traits.

Selection of appropriate controls is equally important in any study

Recruitment bias discussed above can also be a problem with controls. Selection of appro-
priate controls for comparison is a key requirement of association studies. However, few
studies look for completely matched controls. Complete matching is not essential in most
cases, but it is desirable and should be seen as a gold standard in good practice.

2967_Ch03.indd 94 07/07/2015 10:32

 Good and Bad Practice 95

Healthy controls
Most studies refer to their control group as healthy controls. This is at best a vague descrip-
tion and is mostly incorrect. What are healthy controls? In general terms, the authors of
such studies mean individuals who have not been diagnosed as positive for the disease under
study. It is arguable that comparator populations should contain the normal number of cases
of other diseases that are not included in the study. Exclusion of all diseases could create a
genetic bias in the controls. Therefore a good comparator population will not all be healthy.

Matched controls
Matched implies a degree of similarity between the patient group and the control group.
Most studies use racial and ethnically matched controls. This makes sense because there
are racial and ethnic differences in the distribution of different gene polymorphisms.
In addition, there are racial and ethnic differences in environmental factors, examples
include smoking, diet, and alcohol consumption, that may impact on disease susceptibil-
ity or severity. This is the definition of matched for most studies and no further matching
applies. However, further matching may be appropriate where there are subpopulations
(population stratification) within the study group, or where there is a preponderance
of male or females within the disease group. For example, many forms of autoimmune
disease are more common in women than in men and we could argue that this differ-
ence should be reflected in the control population. Body mass index may also be a factor.
Similar caveats apply when selecting controls for studies of diseases in children or diseases
in the elderly population. However, though matching controls and taking the age of onset
into account is desirable, it is not always practical, particularly in children.
Some studies, such as the WTCCC1 study, which is referred to throughout this book, used
a common control set. However, even this can be criticized because there is no consideration
of latent onset of disease in the control cohort. Indeed, we may question the whole idea of
a healthy control. It is quite common to find studies that have used a workplace population
as controls (Table 3.4). In these circumstances this population may be very different from
the general population, reflecting bias in terms of the socioeconomic group from which the
controls come. Finally, on some occasions there will be no healthy controls; instead, there
may be patient subgroups where each subgroup has a different disease phenotype.

Number of controls
General advice is that the number of controls should be at least the same as the number of
patients. If possible, the number of controls may exceed the number of patients provided

Table 3.4 Characteristics of matched controls.

Health Negative for the disease under study at time of recruitment; cannot
control for latent onset
Race/ethnicity Important to reduce effects of population stratification
Age of onset Age of onset is especially important if the illness is likely to have a
high rate of mortality and is common
Sex Male/female differences are common in many common diseases
Socioeconomic factors In some populations this is more important than others and
populations are stratified on this basis

2967_Ch03.indd 95 07/07/2015 10:32

 96 CHAPTER 3 How to Investigate Complex Disease Genetics

the number is not too excessive. As a rough guide, a control sample of up to but not
exceeding five times the patient sample size is considered acceptable. The upper limit is
important because the greater this number becomes, the greater the possibility of intro-
ducing unmatched controls and bias into the control sample.

Errors in the laboratory and in sample handling can also introduce

bias into a study
It is important to ensure that no handling errors occur during sample collection and
processing. There are ways of checking for errors through haplotype analysis. Where there
is an unexpectedly high level of genetic mixing, samples may be excluded and replaced
with fresh samples. Sample checking is performed as part of standard quality assessment
in some studies.

Statistical analysis is the key in any study of complex disease

Statistics are considered separately in Chapter 5. However, there is one important concept
that relates to study design to report here. If we test multiple markers, we must make some
allowance for this in the analysis. In the past we have applied Bonferroni’s correction
by multiplying the probability value by the number of tests performed. This is a crude
form of correction that reduces the likelihood of individuals reporting weak false-positive
associations. However, this correction is much criticized and may cause many weak asso-
ciations to be overlooked. As an alternative to this correction, the level at which a prob-
ability value is accepted as significant may be reduced, taking into account the number of
variables tested. Even for students with no interest in statistics it is essential that there is
an awareness of this practice.

SNP chip selection is an important factor to consider in study design

In science, as everywhere, new methods and ideas are greeted with great enthusiasm. GWAS
has been, and still is, the new boy in town, but we need to be cautious because GWAS have
limitations. All of the above apply to all methods discussed so far, but SNP chip selection
applies to GWAS and subsequent studies only. The selection of the genotyping arrays is very
important. There are a large number of different commercially available genotyping arrays. It
is essential to ensure that the array used is appropriate for the disease being studied. It is also
helpful when comparing studies that have used the same or similar arrays. This makes it pos-
sible to confirm initial findings from other research groups and it facilitates meta-analysis.
However, evolving technologies such as direct sequencing and cheaper SNP chips means
that this is no longer the major issue in these studies. This is discussed further in Chapter 12.

Allele frequency
When selecting arrays it is important to consider the frequency of the alleles under test.
Alleles that are present at low frequencies (minor alleles) are not informative in association
analysis and it is important to filter out any markers that are unlikely to be informative.
The reason for this is that the statistical power of a study depends partly on the allele fre-
quency and is lower when allele frequencies are low. Previously, the threshold chosen to
exclude SNPs and other markers was 5%, but in 2015 as much larger sample collections
have become available, SNP frequencies of 1% or lower are being included in studies. This
addresses some of the concerns that have been raised about excluding rare polymorphisms,
which suggested some associations may be missed and therefore exclusion of rare SNPs
may be counter-productive.

2967_Ch03.indd 96 07/07/2015 10:32

 Good and Bad Practice 97

Unfortunately publication bias does occur

All research work needs to be published and pass through peer review. The process can be
quite punishing and often we find our favorite manuscript bouncing around journals while,
for no apparent reason, manuscripts of lesser quality (or so we think) fly through the process.
However, time heals all and over time as our list of publications grows we find there is a level-
ing out process. It is important to be realistic about our studies and where possible publication
should be considered at the outset of a study and not post hoc. In exceptional circumstances
results that exceed the expectations of the research group may be sent to a higher-level journal
than initially planned, but it is important to be realistic to avoid disappointment.
Publication bias applies less often to papers with positive associations and more often to
those with negative associations. Essentially, failure to find statistically significant associa-
tions can mean that a manuscript will remain unpublished for some time. This then means
there is a bias towards papers with more positive associations. However, the production of
papers with positive results will eventually create openings for papers with negative results.
The problem with publication bias where it exists is that it induces the authors to over-
analyze data until a statistically positive observation is found. Good practice is to set the
aims of a study at the beginning with clear indications of the planned analysis (Box 3.2).
Post hoc subgroup analysis is considered bad practice and should be avoided. There has

BOX 3.2 BASIC GUIDELINES FOR GOOD PRACTICE IN

ASSOCIATION STUDIES

• Replication of associations with alleles should be normal practice prior to publication.

• Case selection should be based on well-defined criteria and unusual or uncer-
tain  cases should be excluded from the main analysis so as to reduce bias.
Variation between study groups must be taken into account when performing a
meta-analysis.
• The appropriate SNP chip should be selected in GWAS.
• The minimum threshold set for tagged SNPs and other markers should be based
on the number of SNPs being tested and size of the study population. Generally,
early GWAS used 5% as the lowest level for the least frequent variant, but threshold
limits of 1% or even lower are being accepted more recently due to the availability
of a greater number of SNPs (many with low frequencies) and larger sample sizes.
• Control groups should be selected from an appropriate population.
• Consideration should be given to race, ethnicity, and other factors.
• Close matching of controls to the disease group, with age of onset and sex matching
if possible.
• Sample size should be the maximum achievable within reasonable boundaries.
Consideration of collaboration to achieve appropriate numbers should be encour-
aged. Sample sizes should reflect any expectation of the likely size of the anticipated
genetic effect.
• The statistical analysis applied to the study should be clearly represented in the manu-
script. This should include initial power calculations performed prior to the study.

2967_Ch03.indd 97 07/07/2015 10:32

 98 CHAPTER 3 How to Investigate Complex Disease Genetics

BOX 3.2 BASIC GUIDELINES FOR GOOD PRACTICE IN

ASSOCIATION STUDIES (Continued)

Where there are multiple alleles tested, appropriate statistical correction (thresholds)
must be applied and the corrected probabilities (Pc) shown. All of the tests per-
formed should be included in the consideration of the appropriate correction factor.
Confidence intervals (CIs) should be shown as standard.
• If possible, closely linked loci should be investigated to provide haplotype data and
define regions of linkage disequilibrium. Linkage disequilibrium is a very useful tool
in confirming genotypes and haplotypes for genes with multiple polymorphisms,
such as within the MHC.
Guidance on strategy is not included here as a specific point. The choice we make when
planning these studies is made up of a cocktail of information. The plan (phase 2) is also
based on that information, but guided by some of the considerations in this box.

been considerable discussion of publication bias and some effort to encourage journals
to be more accepting of good quality studies that carry negative data. In addition, online
publication of such studies is becoming increasingly popular, creating a pathway through
which this bias can be reduced.

Replication in an independent sample is crucial for all association

studies, especially GWAS
Replication or repeating a study is primarily used to validate the initial findings and it is
essential in all association studies, but particularly in GWAS. However, because GWAS have
very stringent statistical parameters for acceptance of identified associations, some studies
have used the replication cohort as a tool to allow them to apply a lower cutoff point for sta-
tistical significance (significance threshold) in their first round. Using this plan, the study
group is divided into two: the first group is genotyped for all of the SNPs and markers, and
the second group is genotyped only for those SNPs and markers that are found to be signifi-
cantly associated with the disease in the first round. By applying a less stringent cutoff for
significance in the first round, it is hoped to reduce the likelihood that weak genetic associa-
tions will be missed (fewer false negatives). In practice, this may mean testing more SNPs
and markers in the second round than would be tested if a higher significance threshold
were set in the first round, but it does ensure that significant associations are not overlooked.
This has become one model for GWAS involving multiple rounds of analysis. Figure 3.11
shows a study design that employs this multistage approach for replication. In the first
phase, the entire genome is scanned for potential associations using approximately 500,000
SNPs. Two alternatives are offered for the second phase. Replication of initial findings at
a significance threshold of P < 10−7 or replication using a lower significance threshold
in the order of P < 10−3 to 10−5 focused on the identified alleles only. Whichever is used
replication of the initial findings is essential to validate them. In many cases a third-step
validation may be applied, focusing on either the same geographic or other geographic
populations. Follow-up studies and fine mapping of associations will also be performed
on specific chromosomal regions, haplotypes, and genes with potential causal links with
the disease. Causality can only be truly proven once functional studies linking phenotype
to genotype have been performed to consider the relationship between phenotypes and
specific systems that may explain the disease.

2967_Ch03.indd 98 07/07/2015 10:32

 Good and Bad Practice 99

Disease Control

Association analysis
round 1
(P<10-7or P<10- 5)

Replication
Alternative round 1
replication If round 1 used P<10-5
round 1 then P<10-7 would be
P<10-3 to10-5 preferred for replication
in round 2

Populations

Chromosomal regions – high resolution

Haplotypes – higher resolution

Genes – biological function – systems

Figure 3.11: Schematic illustration of the study design of a GWAS and the follow-up studies, including
fine mapping and functional investigations. The study employs a multistage approach for replication. First, the
entire genome is scanned for potential associations using approximately 500,000 SNPs and selected significance
thresholds (P < 10−5 or P < 10−7). Then either replication at a significance threshold of P < 10−7 or sometimes
between P < 10−3 and P <10 −5 can be applied. After re-testing, a second case control cohort focusing on the alleles
identified in the first analysis provides confirmation that the associations are valid or not. In most cases, third-
step validation is then applied focusing on other populations if possible. Follow-up studies include fine mapping,
focusing on chromosomal regions, haplotypes, and genes with potential causal links with the disease. The final step
once genes and regions have been identified is to perform functional studies linking phenotype to genotype, and
to determine the relationships between phenotypes and specific systems that may explain the disease.

2967_Ch03.indd 99 07/07/2015 10:32

 100 CHAPTER 3 How to Investigate Complex Disease Genetics

3.4 NEW TECHNOLOGIES AND THE FUTURE

GWAS is currently the primary phase 1 method for studying the genetics of any complex
disease. Even with the promise of multiple rounds of high-resolution genotyping we can-
not carry on doing this forever and at some point we need to focus on the individual genes
and haplotypes to identify the specific risk alleles. Direct sequencing is rapidly catching
up and will soon be the method of choice, replacing GWAS (see Chapter 12). The tech-
nology is not entirely new, but recent developments have enabled much longer sequences
to be determined with a higher degree of accuracy and at significantly lower cost. As a
consequence, it may be easier to abandon the use of tagged SNPs and high-resolution
genotyping for direct sequencing of identified candidate genes. More laboratories will be
encouraged to use this method as costs fall.

The technological advances of the past decade have had a major

impact on research into the genetics of complex disease and the rate
of change is going to increase
New information is constantly appearing. Recent advances in genotyping technology
have made it possible to genotype millions of SNPs in a high-throughput fashion with
lower and lower costs per genotype. An increasing number of databases and informa-
tion have been released on the web. In 2012 the 1000 Genomes Project published
information stating that they had identified and genotyped 38 million SNPs. HapMap
(http://hapmap.ncbi.nlm.nih.gov/) describes the haplotype map of the human genome
based on DNA sampled from eight different populations. HapMap phase 3 (2012)
described the human genetic variants in 1301 DNA samples. The project included
both SNPs as well as large differences from the loss, gain, or duplication of regions (i.e.
CNVs). These technological advances will add even more power to GWAS in complex
disease, but this is only one part of the story.

New developments will come from the ENCODE project, and will also
involve more epigenetics and imputation analysis
The ENCODE project was published in 2012 (http://www.genome.gov/10005107). The
project name refers to the Encyclopedia of DNA Elements and was started in 2003. It was
designed to identify all functional elements in the human genome. Most of the human
genome has been labeled as junk DNA, but this is not quite correct. ENCODE set out
to look at the functional elements only, thereby restricting the workload to manageable
proportions.
In today’s world, genotyping for all but the most complicated genetic systems is mostly
being outsourced. Few laboratories now perform all of their genotyping in situ and the
commercial cost of genotyping is falling. The 1000 Genomes Project is coming to a close
(with much of the preliminary data available now) and many centers will use whole-
genome sequencing in the future rather than SNP genotyping by GWAS, especially as
the cost is reduced. In addition, epigenetics offers a different look at complex disease
and imputation analysis offers an easier way to look at the genome at high resolution.
Currently, few studies have looked at systems, but the indication from some of the more
studied diseases is that systems analysis offers a way forward.
It is important to keep our feet on the ground in the face of these massive developments.
Studies still depend on good quality collections of patients and appropriate controls.

2967_Ch03.indd 100 07/07/2015 10:32

 Conclusions 101

However, everything else has changed or is changing. Increasingly, studies are collabora-
tive rather than being based on a single center. This has obvious benefits in terms of qual-
ity, but it can introduce problems too.

The real debate about the future of complex disease research lies not
in the genetics itself, but downstream from the genetics
Genetics only provides one piece of the jigsaw. The whole picture can only be achieved
by considering the genetics, the systems that risk alleles map to, and the environment.
Systems biology provides the basic science for the biological approach. The environment
is more difficult to tackle. Genes and the environment interact through systems. Consider,
for example, the impact of lipids in the diet and the way in which lipid genes may play a
role in obesity through regulating lipid metabolism. However, it is not so simple. Genes
interact, often in different directions. There is also redundancy in biology to allow for
biological failure or breakdown and also for the impact of the environment. Redundancy
exists in the environment too, for example there is more than one source of most essential
food elements.
A reductionist approach will not be sufficient to understand human disease, because
human biology is complex. There is interaction and redundancy within systems (as stated
above). Thus, deconstructing a genetic model can be interesting, but it does not necessar-
ily provide a clear indication of the relationship between genotype and phenotype.
Epistasis or gene–gene interaction is essential in biology and is increasingly spoken of in
the genetics of complex disease. Redundancy is one example of interaction (albeit nega-
tive) that illustrates this point. Redundancy can be seen in many systems through gene
duplication where multiple copies of genes exist or through more subtle dual coverage,
whereby there are two expressed isoforms of a gene with overlapping functions. It is pos-
sible that this redundancy or genetic compensation allows for some, but not all situations.
For example, the functionally expressed complement gene C4B may be able to compen-
sate for the deleted C4A gene except when the system is stressed, at which point the
absence of C4A may create a dysfunctional environment, permissive for the accumulation
and deposition of immune complexes in the small blood vessels, kidneys, and joints lead-
ing to intense outbreaks of inflammation as seen in SLE.

Conclusions
Following the completion of the Human Genome Mapping Project, HapMap, and SNP
Map, designing studies in complex disease has become relatively easy. However, despite all
of the technological advances, the current options in study design in complex disease are
still linkage and association analysis. The real difference between the past and the present
is one of scale. At one end, it is possible to concentrate on a single gene; at the other, it is
possible to search the whole genome. There are pros and cons for both of these options,
and a number of factors influence the choice of which to apply.
There are fewer restrictions now than there were in the past. In the post-genome era there
is a wealth of new information and technology to latch on to. Genome-wide association
analysis has become the most favored form of study for the past few years. Unrestricted
by prior knowledge and more-or-less hypothesis free, it has given researchers the free-
dom to roam across the genome screening for polymorphisms with relatively small effects.

2967_Ch03.indd 101 07/07/2015 10:32

 102 CHAPTER 3 How to Investigate Complex Disease Genetics

Multiple GWAS has led to accumulation of huge DNA banks and collaboration on a
global much greater scale as well as meta-analyses. This has produced some very significant
reproducible, if small associations. With the right information at the beginning, a good
study plan, and selection of the most appropriate method, which can be a low-technology
candidate gene association study or a full-blown GWAS, the results can be very promis-
ing. However, caution needs to be applied when designing studies. Good planning will
produce good quality results; poor planning leads to irreproducible results and wastes
resources. The lessons of the past are well worth consideration. The literature from the
1960s and 1970s onwards is full of unconfirmed associations. However, we should not
throw the baby out with the bathwater, as the saying goes. Some of the most important
genetic associations in complex disease also date back to 1960s and 1970s. Some would
even go so far as to argue that interest in the genetics of complex diseases, especially those
where there are no multiplex families, would never have been considered without these
early studies.
Careful consideration at the early stages of any research project will help to identify poten-
tial pitfalls and also identify potential new avenues for research. Resources, both financial
and material, are limited. It is therefore essential to make the most out of genetic studies.
The aim should be for high-quality, reproducible results. Choice of options depends on
indicators; of a genetic component in the disease (such as concordance in twins, sibling
relative risk), the available materials (mostly DNA banks), access to patients and their col-
laboration, prior knowledge of the disease pathology, and any results of previous studies.

2967_Ch03.indd 102 07/07/2015 10:32

 Further Reading 103

Further Reading
Books with a guide to and an illustration of the poten-
Frank L (2011) My Beautiful Genome: tial for understanding complex disease in the
Exposing our Genetic Future, One Quirk at a immediate post-genome period.
Time. Oneworld. Davies JL, Kawaguchi Y, Bennett ST et  al.
Spector T (2012) Identically Different: Why (1994) A genome-wide search for human
You Can Change Your Genes. Weidenfeld & type 1 diabetes susceptibility genes. Nature
Nicholson. 371:130–136. This is possibly the first genome-
wide search in complex disease and one which
Strachan T & Read AP (2011) Human used very different techniques compared with
Molecular Genetics, 4th ed. Garland. Chapter the large-scale GWAS of today.
16 deals with identifying human genes and sus-
ceptibility factors and provides excellent addi- Hallmayer J, Cleveland S, Torres A et al. (2011)
tional data for students wishing to delve deeper Genetic heritability and shared environmental
into medical genetics. There is also additional factors among twin pairs with autism. Arch Gen
relevant information in the other chapters. Psychiatry 68:1095–1102.
Hirschhorn JN & Daly MJ (2005) Genome-
wide association studies for common diseases
Articles and complex traits. Nat Rev Genet 6:95–108.
Brewerton DA, Hart FD, Nicholls A et  al. This review puts the idea of GWAS into con-
(1973) Ankylosing spondylitis and HL-A27. text, and provides useful insight into how such
Lancet 301:904–907. One of the earliest studies are planned and executed.
reported genetic associations with HLA— Johnson GCL & Todd JA (2000) Strategies
an association that has stood the test of time in complex disease mapping. Curr Opin Genet
despite changes in HLA phenotyping and Dev 10:330–334. This is an alternative review
genotyping methods and nomenclature (note of the issues in planning studies of genetic asso-
HL-A27 is now correctly referred to as HLA- ciations in complex disease. The paper provides
B*27). This illustrates the point that not all interesting insight into the future development
genetic associations are recent and not all early of GWAS, unachievable in 2002, but since
reports failed to be replicated. realized.
Bhugra D (2005) The global prevalence of Kieseppa T, Partonen T, Haukka J et al. (2004)
schizophrenia. PLoS Med 2:e151. High concordance of bipolar I disorder in a
Cardon LR & Bell JI (2001) Association study nationwide sample of twins. Am J Psychiatry
designs for complex diseases. Nat Rev Genet 161:1814–1821.
2:91–99. This is an excellent review of associa- Satsangi J, Parkes M, Louis E et al. (1996) Two-
tion studies of complex disease. The timing of stage genome-wide search for inflammatory
this review reflects the immediate post-genome bowel disease provides evidence of susceptibil-
era and pre-GWAS era, providing an interest- ity loci on chromosomes 3, 7 and 12. Nat Genet
ing historical insight into the genetics of com- 14:199–202. This paper describes one of the
plex disease. first genome-wide studies in complex disease.
Colhoun HM, McKeigue PM & Smith GD The 1000 Genomes Project Consortium
(2003) Problems of reporting genetic associa- (2010) A map of human genome variation
tions with complex outcomes. Lancet 361:865– from population-scale sequencing. Nature
872. This work provides an excellent guide and 467:1061–1073.
critique of the common problems in studies on The 1000 Genomes Project Consortium (2012)
the genetics of complex disease. It is an infor- An integrated map of genetic variation from
mative and well-scripted paper, and recom- 1,092 human genomes. Nature 491:56–65. The
mended for students wishing to develop their Human Genome (2001) Nature 409:813–958.
critical skills. The complete issue of Nature mostly dedicated
Collins FS & McKusick VA (2001) to the HGM with multiple papers, letters, and
Implications of the human genome project for editorial commentary from multiple authors. It is
medical science. JAMA 285:540–544. This is a full of useful insight and critical discussion. Any
very important review that provides the reader student of human genetics should read this issue.

2967_Ch03.indd 103 07/07/2015 10:32

 104 CHAPTER 3 How to Investigate Complex Disease Genetics

The International HapMap Consortium (2005) critical insight into the thinking behind this
A haplotype map of the human genome. Nature now widely adopted strategy.
437:1299–1320. Willer CJ, Dyment DA, Risch NJ et al. (2003)
The International HapMap 3 Consortium Twin concordance and sibling recurrence rates
(2010) Integrating common and rare genetic in multiple sclerosis. Proc Natl Acad Sci USA
variation in diverse human populations. Nature 100:12877–12882.
467:52–58.
The Wellcome Trust Case Control Consortium
(2007) Genome-wide association study of Online sources
14,000 cases of seven common diseases and http://hapmap.ncbi.nlm.nih.gov
3,000 shared controls. Nature 447:661–678. This site is a useful source for the results of the
This is another landmark paper highlighting International HapMap Project.
the development and application of new tech-
nologies (GWAS) in complex disease research, http://www.ncbi.nlm.nih.gov/omim
and it provides a mass of useful information for This is an essential site for updates and informa-
those studying complex disease. This study pro- tion on genetic disease of all types.
vides some of the basic information for students http://www.ncbi.nlm.nih.gov/projects/SNP/
studying Crohn’s disease as well as a further six snp_summary.cgi
diseases not discussed in this chapter. This site provides information about reference
Wang WY, Barratt BJ, Clayton DG & Todd SNP clusters in the genome.
JA (2005) Genome-wide association studies: http://www.ncbi.nlm.nih.gov/SNP
theoretical and practical concerns. Nat Rev
Genet 6:109–118. This early discussion of the This is an essential site for updates on SNPs.
practicalities of GWAS indicates the way in http://www.genome.gov/10005107
which studies of genetics of complex diseases The ENCODE project – a useful site from
were about to develop and provides some which information is freely available.

2967_Ch03.indd 104 07/07/2015 10:32

 CHAPTER
4
Why Investigate Complex
Disease Genetics?

We have already seen that complex diseases do not lend themselves to simple analysis, and
by definition we are dealing with mutations and polymorphisms that increase or reduce
the risk of a trait. It is also clear that there may be complex interactions between groups
of risk alleles (epistasis), and between risk alleles and the environment. All of the above
suggest that identifying genetic components in complex disease is and will be a difficult
task, and yet it is a task that has attracted a considerable amount of attention in biomedical
research. The potential rewards from this research are great and the technological advances
made over the past 20 years have been astounding.
In this chapter, we will consider the question of why we invest so much time and so much
of our resources to identify risk alleles in complex disease. We will use five key examples
of complex disease to illustrate these: ankylosing spondylitis, cardiovascular disease, hep-
atitis C virus (HCV), rheumatoid arthritis, and bipolar syndrome or bipolar disorder.
These five examples illustrate a host of key points, but they are selected here specifically
to illustrate the potential to use genetics in the differential diagnosis in disease (ankylos-
ing spondylitis), in understanding the response to infection and potentially planning for
patient management (HCV), and also to increase our knowledge of disease pathogenesis
(cardiovascular disease, rheumatoid arthritis, and bipolar disorder). These are complex
issues as the text will show, but the central principals are simple.

2967_Ch04.indd 105 07/07/2015 10:35

 106 CHAPTER 4 Why Investigate Complex Disease Genetics?

4.1 Why do we Investigate Complex Disease?

In considering the question we need to remind ourselves of a number of key points.
• Complex diseases do not conform to simple Mendelian patterns of inheritance.
• The potential clinical outcome from research in complex disease, sometimes referred to
as the promises of the Human Genome Project (HGP), include the ideas that knowl-
edge of the genetic background in complex disease:
• may be used to aid disease diagnosis.
• may be used to aid treatment/management and patient care.
• may help us to understand the pathogenesis of the disease, and help with the devel-
opment of novel therapeutic agents and strategies.
Each of these will be discussed in turn in this chapter.

Complex diseases do not conform to simple patterns of inheritance

By this stage in this book, the first point above is obvious, but it is worth reminding our-
selves that we are dealing with complex issues. When things become complex, we are often
forced to ask whether we should be spending vast amounts of time and effort to investigate
them. It should not surprise us that the questions we are trying to answer are difficult.
Science is full of questions that are difficult to answer. Yet we have to be cautious. To para-
phrase Hugo Mencken, the author and satirist, “there is always an easy solution for every
human problem (that is) neat, plausible, and wrong.” We are inclined, as Nobel Laureate
Herbert Simon suggests in his statement about the aims of science, to “find meaning-
ful simplicity in the midst of disordered complexity.” In other words, it is part of our
humanity that we should seek to find simple solutions to complex problems (Figure 4.1).
Mencken is telling us to beware, whereas Simon is encouraging us to tackle the issue and
not be put off. However, complex diseases are complex by nature, and therefore we should
not expect a neat and (simple) plausible answer. Once we agree with this principle then we

There has to be a
simple solution

Figure 4.1: Human desire to find simple solutions to complex

problems. It is our nature to seek simple solutions to complex problems.
However, we need to be aware of simple solutions, especially in the
genetics of complex disease. Are we looking for a neat and simple answer
that is wrong or are we over-complicating science? It is important to ask
these questions in science and reflect on the broad issues as well as the
pure facts.

2967_Ch04.indd 106 07/07/2015 10:35

 Why do we Investigate Complex Disease? 107

can proceed to answer the basic question. Why investigate the genetics of complex disease?
This is not pessimism, it is realism and that "realism" is essential in science.
Part of understanding the reasons for the mass of research in complex disease is born
out of an appreciation of the diseases themselves. Complex diseases are common, if we
include cancer in the count they may account for as much as 95% of all forms of human
disease. If we exclude cancer then the figure is likely to be closer to 75%. Not all cancers
are complex; some are Mendelian and some are mosaics that develop during the life of an
individual, that is they are not inherited from the parents. However, some mosaics can be
stimulated into production by inherited genetic variations. Cancer as we shall see later in
the book is a difficult area in complex disease. However, overall complex diseases whether
75% or 95% of all forms of human disease still have a major social and economic impact.
Understanding genetics helps us to understand disease. For example, we currently know
very little about the pathogenesis of many common diseases. This is because patients pres-
ent relatively late in the disease process, sometimes long after the disease-initiating pro-
cesses have done their work. Trying to understand disease pathogenesis by looking at late
stages of a disease can be like looking for fingerprints in a room where the criminal wore
gloves. However, genetics can help. As individuals, our whole genome does not change,
while the clinical symptoms of diseases do. There are some changes in specific cells due to
mutations during aging and in some cancers, but these are not present in all cells.

The HGP in research into genetically complex disease

Identifying risk alleles for common genetically complex disease has been set as one of
the major challenges of the post-genome era and indeed it was the potential to identify
such risk alleles that made the Human Genome Mapping Project (HGMP) such an
attractive proposal. The idea of finding risk alleles in complex disease is not new. However,
early studies were confined to selected candidate genes, the locations of which were
mostly unknown. The publication of the HGM, SNP Map (http://www.ncbi.nlm.nih.
gov/projects/SNP/snp_summary.cgi), and HapMap (http://hapmap.ncbi.nlm.nih.gov/)
enabled the door to understanding the genetics of complex disease to be fully opened
for the first time and now the race is on.
The promises of the HGP listed above and illustrated in Figure 4.2 will be used as the
basis for the discussion of the potential clinical outcomes and benefits of investigating

SNP Map HapMap

Human Genome
Project

Diagnosis Pathogenesis Figure 4.2: Clinical outcomes of the HGP. The figure illustrates the
potential clinical outcomes (promises) of the HGP in medical sciences
and the way in which the three main promises are interrelated. The
outcomes are: to aid diagnosis, to help with patient management and
care (including development of new therapies), and to improve the
Treatment understanding of disease pathology. All are achievable to some extent,
and care though some will take longer than others.

2967_Ch04.indd 107 07/07/2015 10:35

 108 CHAPTER 4 Why Investigate Complex Disease Genetics?

genetics in complex disease (below). The idea of individualized therapy and identifying
key pathways in disease pathogenesis are those that have been most achievable so far, but
ultimately all of these promises are linked. It is important to consider not only the data
that studies produce, but also the application of the data in terms of meeting these expec-
tations or promises.

4.2 Disease Diagnosis

It was suggested in 2001 when the draft HGP was published that genetic tests would
increasingly be used to predict disease and initiate preventative action. This is pre-­diagnostic
testing. Francis Bacon reminds us to be cautious; he said “the human understanding . . .
readily supposes a greater order and uniformity in things than it finds, (it) is infused by
desire and emotion, which gives rise to wishful science.” When we look at this promise in
particular it is important to consider whether the promise constitutes wishful science. The
extent to which identified risk alleles are currently used in disease diagnosis will provide a
clue. However, the best examples that are currently available are those that were identified
before the HGP began. This is by no means negative, as it shows what can be achieved.

Early studies on the genetics of ankylosing spondylitis indicated

what could be achieved in terms of differential diagnosis in the
post‑genome era
One of the best examples that has received a great amount of attention because of the
potential clinical value associated with it is the association between the rheumatic disease
ankylosing spondylitis and the HLA-B*27 allele family. The main ankylosing spondylitis
risk allele is located in the major histocompatibility complex (MHC). The MHC encodes
the genes for the human leukocyte antigens (HLA genes) and it is one of the most com-
plex genetic systems in the human genome. The MHC is discussed in detail in Chapter 6
and further information for this chapter is given in Box 4.1.
The pathogenesis of ankylosing spondylitis is unknown; however, it is a relatively com-
mon rheumatic disorder that occurs in up to 0.5% of the UK population. It is a debili-
tating disease of the lower spine that develops in early adulthood. Though the disease
usually involves the spine, it can cause other symptoms, such as abnormalities in the
aortic valve and pulmonary fibrosis. The genetic association with HLA-B*27 was first
described in 1973 by Brewerton et al. and confirmed by Schlosstein et al. in the same
year. Other clinical conditions that have similar symptoms have also been associated with
HLA-B*27, including reactive arthritis, uveitis, and sacroileitis in inflammatory bowel
disease, and psoriasis. However, none of these other conditions is as strongly associated
with HLA‑B*27 as is ankylosing spondylitis. Estimates for the strength of this association
between HLA-B*27 and ankylosing spondylitis vary. The gold standard study of ankylos-
ing spondylitis by Brown et al. published in 1996, suggested the odds ratio (OR) for anky-
losing spondylitis in those with HLA-B*27 may be as high as 171 (with 95% confidence
intervals of 135–218). This is one of the strongest genetic associations ever described
for an HLA allele family and clinically these data are of great potential significance. In
terms of diagnosis however simply being HLA-B*27-positive is not an indication that a
patient has ankylosing spondylitis or will develop the disease. Even in Brown’s study, 4%
of patients with ankylosing spondylitis were HLA-B*27-negative and approximately 9%
of the UK population are HLA-B*27-positive. If possession of HLA-B*27 were sufficient

2967_Ch04.indd 108 07/07/2015 10:35

 Disease Diagnosis 109

BOX 4.1 A SHORT HISTORY OF THE MHC

The problem with HLA genes is that they reside within the major histocompatibility
complex (MHC), which is so-called because of the role many of the genes encoded
there play in determining compatibility of tissues. A huge interest in HLA grew out of
the practical need to find matched organs for transplantation, particularly kidney trans-
plantation. In studying complex disease, the HLA genes are the most problematic. There
are two reasons for this: they are the most variable genes in the human genome and they
have the highest levels of linkage disequilibrium in the human genome. HLA antigens
are not expressed on red blood cells. HLA typing was initially performed using serum to
detect antigens on human white cells (leukocytes—hence the name “human leukocyte
antigens” for the genes). The HLA phenotypes, as they were correctly called, identified
a range of different antigens on four different HLA loci: HLA-A, HLA-B, and HLA-C
(all class I loci), and HLA-DR (a class II HLA gene).
The quality of early studies was based on the availability of banks of sera to identify a range
of antigens. Rare antigens were often missed or not tested and many remained undiscov-
ered until the 1990s. Later studies were able to capitalize on the introduction of molecular
genetics, first restriction fragment length polymorphism (RFLP) and then polymerase
chain reaction (PCR), to genotype the HLA genes. The introduction of molecular genet-
ics led to the discovery of other HLA genes, in particular the DQ and DP family.
As methods changed, so the names changed. In the 1970s and 1980s, the genes were
called HLA A, B, Cw, and DR, with the antigens identified in numbers without a space
(e.g. A1) and roman script used. In the 1990s, the names were changed again, e.g. HLA-
A, HLA-B, and HLA-C  (the “w” was dropped from the name for all HLA-C antigens and
genes). For each allele, the gene name is followed by an asterisk (*) and then a number
starting with 01, e.g. HLA-A*01. If a variant of this allele family is known, then this is
labeled with an additional colon (:) and two more numbers, e.g. HLA-A*01:01. The
alleles are all given in italics.
Two additional essential facts are that unlike HLA-A, -B, and -C, which encode only a
single α-polypeptide, the HLA class II antigens are the product of two expressed HLA
genes: an A gene and a B gene encoding an α-peptide and a β-peptide, respectively. Both
are found close together on chromosome 6p21.3. In all cases, these genes are polymor-
phic, but this polymorphism is limited in the case of the DRA gene.
One of the most difficult issues relating to the MHC is the extreme linkage disequilib-
rium between the alleles encoded within the region. This can make dissecting the details
about which candidate gene may be responsible for increasing the risk difficult. It is
also possible that the conserved combinations indicate multiple risk alleles on the same
haplotype. The MHC plays a central role in complex disease and cannot be dismissed.
More details about the MHC are given in Chapter 6 and associations with disease appear
throughout this book.

for ankylosing spondylitis to occur, then the disease would be found in a much greater
proportion of the population and all ankylosing spondylitis patients would be expected
to have HLA-B*27. In practice, only a small proportion of HLA-B*27-positives develop
ankylosing spondylitis; this is an example of incomplete penetrance (see Chapter 2). This

2967_Ch04.indd 109 07/07/2015 10:35

 110 CHAPTER 4 Why Investigate Complex Disease Genetics?

Diagnosis A Genetic test

Confirms
diagnosis A

Diagnosis B
Refutes
diagnosis B

Figure 4.3: Use of genetic tests in differential diagnosis in complex disease. The patient illustrated has two
potential diagnoses. For each diagnosis, genetic tests can help to determine which diagnosis is more likely to be
correct. This is called differential diagnosis and is widely used, and genetics is likely to play an increased role in
clinical practice. However, the genetic test is not being used here to diagnose the disease per se, but to add weight
to the likelihood of one or other diagnoses. Note this example is not a Mendelian disease, but a complex disease.

departure from the expected Mendelian norm is exactly what we expect to find in geneti-
cally complex disease, because complex diseases do not conform to Mendelian patterns of
inheritance. However, this does not mean that it is impossible to achieve the goal of using
risk alleles as a diagnostic tool. Associated alleles can be used provided the limitations are
correctly understood. In ankylosing spondylitis, HLA-B*27 testing can be helpful in mak-
ing a differential diagnosis as the simple illustration in Figure 4.3 shows.

Genetic associations in complex disease confer small risks

Most genetic associations found in complex diseases are with alleles that confer relatively
small risks. However, there are some exceptions. These exceptions include strong associa-
tions with mutations in the PTPN22 gene in rheumatoid arthritis (see below), and with
certain adverse drug reactions, such as the anti-human immunodeficiency virus (HIV)
drug abacavir and HLA-B*57:01 and the commonly used antibiotic flucloxacillin that is
also associated with HLA-B*57:01 (see Chapter 8).
Of course we cannot know what may arise in the future, just as the HGP could not predict
the success in each area. Current evidence suggests that most common diseases are likely
to be more complex than we have imagined and involve more risk alleles than we have
thought likely. For example, Crohn’s disease is associated with more than 163 risk alleles.
Without the complete picture of how risk alleles interact we cannot know whether indi-
viduals have to have a group of risk alleles (risk portfolio) for a specific disease (Figure 4.4)
or just one risk allele. If there are multiple interacting risk alleles, even though the indi-
vidual alleles may have a small impact in terms of risk, the combinations may have much
larger effects and identifying these risk portfolios may turn out to be useful in diagnosis.

4.3 Patient Treatment/Management and Care

As we search for alleles with small risk effects it will become increasingly difficult to use
the observations to support diagnosis in the classical sense, and yet it is possible that these
developments will help with planning patient treatment and care. Characteristics such as
disease progression, severity, and response to treatment are traits that may have genetic
associations.

2967_Ch04.indd 110 07/07/2015 10:35

 Patient Treatment/Management and Care 111

Risk alleles Risk portfolios

A B

1 5 1 3 2

5 8 4 7
2 6

C D
3 7
1 3 1 2
8

4 8 4 7 5 6

Figure 4.4: Risk portfolios in complex disease. The figure illustrates risk alleles numbered 1–8 in the circles
on the left and four cases in square boxes labeled A–D. Not all cases may have the same risk alleles. This figure
illustrates the idea that individual cases may have different risk portfolios, i.e. groups of risk alleles (shown in the
black boxes). Here, case A has 1–3–5–8, case B has 2–4–7, case C has 1–3–4–7, and case D has 1–2–5–6–8. Note
that each case has a different set of risk alleles and some cases have more risk alleles than others, e.g. B has three
alleles, D has five alleles, while A and C each have four alleles. Note that whatever the risk portfolio, the sum of
the risk is unlikely to be 1. These are complex diseases, thus being a carrier of a group of risk alleles will be neither
necessary nor sufficient for the disease to occur.

It is very easy in a research laboratory to forget that patients are individuals. In a labora-
tory, patients can become numbers within a population. We tend to lose sight of the fact
that each individual patient is different and we tend to assume that all cases with the same
disease are exactly alike. This is not true. A disease group is a collection of patients with
the same signs and symptoms. However, there is usually a great deal of variability in terms
of age of onset, severity, response to treatment, and prognosis.

Identifying risk alleles that predict onset of complex diseases may

enable patients to make beneficial lifestyle changes
In Alzheimer’s disease, there are early- and late-onset models (EOAD and LOAD; dis-
cussed in Chapter 2). The genetics of these two forms are different, though the signs and
symptoms are very similar. Identifying markers that will predict onset of complex disease
could be helpful if changes in lifestyle, for example diet, would prevent the disease or at
least extend the time to onset. In addition, genetic prediction of onset may be helpful if
there are preventative treatment options that can be applied. This may be particularly use-
ful in familial cases, but the same limitations apply to the use of identified risk alleles in
disease prevention as apply to their use in diagnosis. Possession of the risk allele does not
guarantee that the disease will occur. There is also no guarantee that lifestyle changes will
prevent the disease. It is all about balancing the risk.

Predicting disease severity through genetic analysis may have clinical

significance in terms of patient management
More promising are the possibilities to use identified risk alleles as markers of severity
and also response to treatment and prognosis. There have been several associations with
infectious disease and genetic variation in the human genome. Many of these concern the

2967_Ch04.indd 111 07/07/2015 10:35

 112 CHAPTER 4 Why Investigate Complex Disease Genetics?

outcome following HIV-1 infection and are discussed in Chapter 7. Response to HCV
has also been found to be strongly associated with polymorphism in the HLA‑DQB1
gene. This association is thought to be primarily with the DQB1*03:01 allele. The
DQB1*03:01 allele is in linkage disequilibrium with both DRB1*04 and DRB1*11. In
the UK DQB1*03:01 is most often found with DRB1*04 and in Southern France it is
most often found with DRB1*11. The DQB1*03:01 allele has been associated with a
higher incidence of self-limiting (acute) HCV infection compared to chronic HCV infec-
tion (Figure 4.5). Further studies have also shown that this polymorphism is associated
with a stronger host T cell response to synthetic HCV peptides. The clinical significance
of this observation is that HCV infection is very common and the majority of cases do
not spontaneously clear the virus. Most cases (50–80%) will develop a chronic infection
that over a long period of time can progress to liver fibrosis, liver failure, liver cancer, and
death. Interestingly, a significant proportion of cases do not respond to treatment when
treated. Patients with progressive disease may eventually require liver transplantation.
From a diagnostic perspective this association is of limited value. DQB1*03:01 is com-
mon (about 30% of the UK population have a DQB1*03:01 allele), but from a biological
perspective it is interesting. What is it about DQB1*03:01 that enables the majority of
those who carry this allele to clear the virus? In other words, what is DQB1*03:01 doing or
not doing? Information about this allele and how it works may be used in clinical practice
to develop a vaccine or to manipulate the host response to the virus in such a way that it is
spontaneously eliminated. At the time of writing, new treatments in HCV are having very
significant effects so the potential use of host genetics may change. More recently, there
have been quite a number of studies identifying genetic variations that are associated with
the outcome following HCV infection and response to treatment. These have identified a
number of interferon (IFN) genes, among others, that may prove important in develop-
ing new personalized therapy plans for patients with HCV and other infectious diseases.

70
Acute
60
Chronic
50
Frequency

40

30

20

10

UK DRB1*04 UK DQB1*03:01 France DQB1*03:01

Figure 4.5: HLA class II determines outcome following infection with HCV. The figure shows real data from
two different studies. Both studies reported genetic associations with HLA in two different patient groups: those
who developed chronic infection and those who had an acute (self-limiting) infection following HCV infection.
A group from the UK and one from France each reported a lower frequency of the DQB1*03:01 allele in patients
with chronic HCV infection. In the UK population, this allele is most often carried on the DRB1*04 haplotype
(data for DRB1*04 shown); in France, this allele is found more frequently with DRB1*11:01 (data not shown). The
important observation is that acute infection is associated with DQB1*03:01 in both populations.

2967_Ch04.indd 112 07/07/2015 10:35

 Patient Treatment/Management and Care 113

Common genetic variation may predict response to treatment

and be critical in patient care
Response to treatment is considered more extensively under the heading of pharmacoge-
netics in Chapter 8. This is a hot spot for research at present. Recently, there have been
major discoveries associating MHC genes with adverse reactions to commonly used phar-
macological agents. However, pharmacogenetics has a long history, with the first direct
study of pharmacogenetics beginning reported in 1932. Most of this data has yet to be
applied in a clinical setting. However, that is not to say the data could not be used. Genetic
variations in the response to the antiviral abacavir, are one such example.

Onset, severity, and response to treatment are all part of patient

management
All of the above are important in patient management. Deciding which drugs to use and
the correct dose is essential. Close monitoring of patient progression, and taking timely
and decisive action is essential. This does not only apply to pharmacogenetics, risk alleles
may also identify which patients within a group will progress more rapidly and require
more radical treatment. In the future, genetic risk assessment may be one means of ensur-
ing healthcare resources are used appropriately and with maximum efficiency.
Consider, for example, transplantation. At present, demand for transplants outweighs the
availability of donor organs (Figure 4.6). In the future, genetics could be used to ensure
resources, in this case donor organs, are used efficiently and appropriately. Consider the
example of two patients (A and B) with the same autoimmune liver disease (Figure 4.7).
Imagine that the disease is a progressive disorder for which there is no effective treatment.
Both patients will each eventually require a liver transplant. Both patients A and B present
clinically at the same age and the same stage, but only one can be listed for transplantation
due to a lack of donors. Presently this is a dilemma, but in the future it may be possible

UK Approximate
waiting number
list of donors

Heart 196 142

Lungs 227 188

Liver 462 775

Kidneys 6111 1750

Figure 4.6: Estimated donor organ availability versus demand. The figure illustrates the difference between
donor organ availability in the UK and need for four major organs in 2013. Note that the available figures are
constantly updated. These figures illustrate a major clinical issue, which is that in most years only the donor
numbers for livers seem to match their target. In some years there is even a deficit in liver donors. A large number
of the transplanted kidneys are from matched, related donors. Approximate donor figures are based on the
previous year’s transplant figures as there is no way to know the actual number of donors for the year ahead. On
a global scale, the donor deficits are much more pronounced and there is a serious debate over the advantages
of opt-in versus opt-out systems.

2967_Ch04.indd 113 07/07/2015 10:35

 114 CHAPTER 4 Why Investigate Complex Disease Genetics?

Patient A Risk portfolios Progress

Mild Moderate Severe

1 3

5 8 Rapid progression

Patient B

4 7 Slow progression

Figure 4.7: Disease progression and risk portfolios. This figure shows the potential to apply genetic testing to
identify patients at risk of rapid progression versus those with slow or moderate progression. Such information
could be used to determine the best course of treatment and management for individual cases. The use of
genetics in this way may also maximize the use of key resources, such as donor organs for transplantation.

to use genetic profiling to determine the risk portfolio for the two patients. This genetic
risk portfolio could then be used to determine the likely rate of progression for the two
patients and this information could be used to inform the selection about who (A or B)
to list for transplantation. This would lead to a more rational use of a precious resource
(transplant donors) and be of benefit to both patients as the unlisted patient could be
listed for transplantation later.

4.4 Disease Pathogenesis

One of the ultimate goals for the HGP is to map and understand the genetics of all human
disease. Understanding disease pathogenesis is the key element within this. The sad fact is
that we know too little about the pathogenesis of many common diseases. This means that
details of the early pathogenesis are lost to investigators and complicated by downstream
pathology. To understand what causes a disease we need to see it at the outset. We need to
monitor the processes and mechanisms that lead to the symptoms as pathology changes
with time. Systems fail due to the slow breakdown of intricate, interacting networks. This
breakdown almost certainly takes place over a considerable length of time, but in chronic
disease we most often see the end of the process only, making the process of understand-
ing disease pathology difficult to say the least. Fortunately, the genome of individuals does
not change, though there may be mutations in individual cells, and population genetics
evolves. Therefore, we can look at commonly inherited genetic variation to seek risk alleles
that may act as markers to outline disease pathology.
Unlike using risk alleles for diagnostic purposes, we are not restricted to the consider-
ation of alleles with large effects when studying pathology – we can use data on any allele

2967_Ch04.indd 114 07/07/2015 10:35

 Disease Pathogenesis 115

with a significant effect. This means that even markers for risk alleles associated with
small effects are of value. Consequently, it is in this arena that the majority of studies
have been focused and there have been some notable successes, and even though there is
still much, much more to come, we do not need to look back to the pre-genome era to
find good examples.
The Wellcome Trust Case Control Consortium (WTCCC1) study published in 2007
included seven diseases: Crohn’s disease (considered in Chapter 2), type 1 diabetes (T1D)
and type 2 diabetes (T2D) (both included in Chapter 10), rheumatoid arthritis, bipolar
disorder, coronary artery disease, and hypertension. There had been considerable work
prior to 2007 in each of these diseases but in each case the WTCCC1 study represented a
milestone in genetic investigation, though not necessarily the first genome-wide associa-
tion study (GWAS) in each case.
The WTCCC1 study confirmed strong associations, i.e. with a significance threshold of
P < 10−7 with the MHC region and PTPN22 in rheumatoid arthritis, and identified novel
significant associations with rs420259 on chromosome 16p12 in bipolar disorder and
with rs1333049 on chromosome 9p21 in cardiovascular disease. In addition, the study
identified nine moderate (i.e. significance threshold of P < 10−5) associations with rheu-
matoid arthritis, 13 with bipolar disorder, and six with hypertension.
In the remainder of the present chapter we will consider four diseases: ankylosing spon-
dylitis, rheumatoid arthritis, bipolar disorder, and cardiovascular disease. Three of these
are from this list of seven examples in the WTCCC1 study above, and will be considered
in detail to illustrate the concept that genetics can help us understand disease pathology.
Ankylosing spondylitis was not considered in the WTCCC1 study, but has been consid-
ered more recently in the WTCCC2 study. We will look at early pre-GWAS, early GWAS
(2007), and more recent studies for each disease. Regarding the other diseases in the
WTCCC1 study, Crohn’s disease is discussed in Chapter 2, T1D and T2D are discussed
in Chapter 10, and hypertension is not discussed, though many other diseases and traits
are discussed throughout the book.

Early studies offered potential insight into the biology of ankylosing

spondylitis
The HLA-B*27 association in ankylosing spondylitis has also been considered as having
a role in disease pathology. HLA-B*27 encodes a molecule with an unpaired cysteine
residue at position 67 that forms a disulfide bond with thiol-containing agents. It has
been suggested this results in an immune response to altered self-antigens, either by
antibodies or by natural killer (NK) cells (Figure 4.8). An alternative hypothesis is
that HLA-B molecules with cysteine 67 may have a different preference for certain ath-
ritogenic peptides leading to an increased likelihood of ankylosing spondylitis in those
with HLA-B*27. Crystallographic studies of B27 molecules with a variety of bound
peptides in the ­antigen-binding groove have shown arginine in the second position of
all bound peptides. Arginine carries a long side chain that engages a specific pocket in
the antigen-binding groove where cysteine 67 is also located. Subtyping HLA-B*27 sug-
gests not all HLA-B*27 alleles are associated with ankylosing spondylitis. In particular,
HLA-B*27:02, -B*27:04, and -B*27:05 are associated with ankylosing spondylitis, but
the association with HLA-B*27:03 is uncertain, and HLA-B*27:06 and -B*27:09 are
not associated. The hope for the future in ankylosing spondylitis is that such studies
will narrow the search for athritogenic peptides, leading eventually to novel and more
effective therapies.

2967_Ch04.indd 115 07/07/2015 10:35

 116 CHAPTER 4 Why Investigate Complex Disease Genetics?

Figure 4.8: Molecular mimicry in ankylosing spondylitis. The figure illustrates one of the major theories
behind the genetic association of ankylosing spondylitis with HLA-B*27. The pathogenic peptides (represented by
the dark gray smiley face with a hat) have sequences that are very similar to the host (represented by the smiley
white face without a hat).

Later GWAS have offered even further insight into the biology of
ankylosing spondylitis
The WTCCC2 study and others, including the International Genetics of Ankylosing
Spondylitis Consortium (IGAS), have identified at least 25 risk alleles for ankylosing spon-
dylitis, all of which are linked to immune function. The IGAS study alone found 25 risk
loci at a significance threshold of P < 5 × 10−7. Twelve of these had been previously identi-
fied between 2007 and 2011 (including ERAP1, IL12B, and IL23R). ERAP1 (also known
as ARTS) stands for endoplasmic reticulum aminopeptidase 1. In addition, in the Han
Chinese population, two other loci have been reported: HAPLN1–EDIL3 and ANO6.
A large number of the 25 identified genes produce proteins involved in immune inflam-
matory activity, e.g. IL6R and IL23R. The impact of the cytokines and their receptors has
received considerable attention with interest in interleukin (IL)-23 and IL-17 interaction
in pathogenesis, and the development of potential anti-IL-17 therapy. It is possible that in
ankylosing spondylitis we are now quite close to reaching the stage where understanding
the genetic basis of disease pathology will lead to the development of individualized treat-
ment and better diagnosis based on knowledge of each individual’s genetic risk portfolio.

Rheumatoid arthritis has many strong genetic associations, some of

which can be used to help us unravel the pathogenesis of this disease
Rheumatoid arthritis is a common disease affecting approximately 400,000 people in the
UK. Globally, the disease may affect as many as 1% of the adult population, it is more
common in women and in the aging population (Figure 4.9), and is characterized by
inflammation of the joints and by increased or abnormal levels of inflammatory activity
in the synovial fluid, leading to destruction of the synovial joints. Two-thirds of cases are
rheumatoid factor-positive, carrying antibodies against cyclic-citrullinated peptide (anti-
CCP). Generally, rheumatoid arthritis is thought to be an autoimmune disease and it
has many of the classical markers of such a disease, i.e. a female preponderance, auto-
antibodies, and genetic associations with risk markers in the MHC. There is an imbalance
in the normal immune regulatory procedures in rheumatoid arthritis, such that though

2967_Ch04.indd 116 07/07/2015 10:35

 Disease Pathogenesis 117

300,000
16 – 44

250,000 44 – 64

200,000 64+
Prevalence

150,000

100,000

50,000

Male Female

Figure 4.9: Prevalence of rheumatoid arthritis in the UK. The figure shows the prevalence (total number of
cases) in each age category for male and female patients from Symmons et al. (2002). Note that the prevalence
rises in each age group compared with the previous group and the greater prevalence in females compared with
males for all three age groups. (Data from Symmons D, Turner G, Webb R et al. [2002] Rheumatology 41:793–800.)

anti-inflammatory mediators are present and activated they fail to adequately downregu-
late the immune response. Genetic studies in seropositive European cases have identified a
large number of risk alleles for rheumatoid arthritis and these alleles are mostly associated
with immune or immune-related function. In the sections below we will briefly consider
some of the early studies of rheumatoid arthritis and then concentrate on three recent
studies, all based on GWAS and meta-analysis of GWAS data.
The early history of rheumatoid arthritis, the MHC, and PTPN22
Evidence of a genetic component in rheumatoid arthritis is based on familial occurrence
and elevated sibling relative risks (λs) of approximately 10. Early studies identified strong
reproducible associations with the HLA genes on chromosome 6p21.3 and also with the
PTPN22 gene on chromosome 1p13, which encodes the gene for the lymphoid-tyrosine
phosphatase (Lyp) protein tyrosine phosphatase non-receptor type 22.
The MHC in rheumatoid arthritis
The WTCCC1 study identified HLA as the major risk region for rheumatoid arthritis. This
finding was not a surprise as the association had been described long before this. The genetic
association between rheumatoid arthritis and HLA is very neatly summarized by Gerald
Nepom in his chapter in the book HLA in Health and Disease (Lechler and Warrens, 2000)
(summarized here in Table 4.1). The primary associations in each case are with alleles from
the DRB1*04 family, with the exception of Native Americans, where the primary associa-
tion is with the DRB1*14:02 allele. All of these susceptibility alleles either encode the amino
acid sequence QKRAA or QRRAA (Q = glutamine; K = lysine; R = arginine; A = alanine)
at positions 70–74 of the DRβ polypeptide of the HLA DR molecule. This shared epitope
is within the HLA-binding groove and a site where antigenic side chains engage with the
HLA molecule. At the molecular level, the crucial consideration is the e­ lectrostatic charge
in and around the binding pocket. The electrostatic charge at each pocket and within the
groove will determine which peptide antigens are preferentially bound and presented to the
T cell receptor (see Chapter 6). Allelic variation in the molecular structure of the expressed

2967_Ch04.indd 117 07/07/2015 10:35

 118 CHAPTER 4 Why Investigate Complex Disease Genetics?

Table 4.1 HLA and shared epitopes associated with rheumatoid arthritis.

HLA class II allele Relative risk Shared epitope (DRβ-70–74)

DRB1*04:01 6–11 QKRAA
DRB1*04:04 5–14 QRRAA
DRB1*04:05 6–10 QRRAA
DRB1*14:02 1–2 QRRAA

Lechler R & Warrens A (eds) (2000) HLA in Health and Disease, 2nd ed. Elsevier. With permission from Elsevier.

HLA molecule encoded by different HLA alleles may explain the genetic association in
rheumatoid arthritis, providing a platform for further work to identify athritogenic pep-
tides and potentially develop novel therapies for the disease.
In addition, further work in rheumatoid arthritis suggests that there may be a gene dose effect.
Not all DRB1*04 family members carry the QKRAA or QRRAA sequence. In those that do
carry this sequence at the DRB1 70-74 position, especially: DRB1*04:01, DRB1*04:04, and
DRB1*04:05, all have a greater risk of RA. This is higher in those who carry two copies of
this sequence (that is they are homozygous), than in those who carry only one copy (that is
they are heterozygous). Despite the strong association between HLA and rheumatoid arthri-
tis, the risk ratios listed indicate that HLA-DR genotyping would be of very little value in the
diagnosis of rheumatoid arthritis. However, it may be of some use as a prognostic indicator.
The DRB1*04 alleles are associated with more progressive severe erosive disease.
The story of rheumatoid arthritis is more complex than first thought
It is always nice to see a clear explanation for HLA associations with disease, such as the
shared epitope hypothesis above for rheumatoid arthritis. However, recent GWAS indicate
a more complex situation in rheumatoid arthritis with at least three different HLA genes
involved. DRB1 remains the strongest, but independent associations with HLA‑B*08 and
DPB1 have also been identified. In this analysis some elements of the shared epitope hypoth-
esis for DRB1 still stand, but the amino acids at positions 11, 71, and 74 are highlighted
as the most important, and potentially the only DRB1 amino acids with causal effects in
rheumatoid arthritis. This idea is supported by observations of significant, if weaker associa-
tions with amino acids at position 9 in the binding grooves of both HLA‑B*08 and DPB1.
This is a departure from the idea of a string of amino acids and focuses more on specific
isolated amino acids only. These studies involved a complex analysis with a mix of GWAS
using 3117 single nucleotide polymorphisms (SNPs) across the MHC in over 5000 cases
with seropositive rheumatoid arthritis and close to 15,000 controls. Imputation analysis
was used to identify the HLA alleles and haplotypes. Rheumatoid arthritis is not a rare ill-
ness and therefore these findings are likely to be either confirmed or challenged by follow-
up studies. It will also be interesting to see how direct sequencing impacts on these findings
and how imputation analysis compares with direct sequencing.

PTPN22 in rheumatoid arthritis

In mouse models, mutations in the PTPN22 gene have been shown to be associated
with an enlarged thymus and spleen, and with increased T cell, B cell, and dendritic cell
activity. The exact mechanism for this was until recently unknown. Recent data suggests
PTPN22 is involved in calpain-mediated degradation of Lyp. In mice, polymorphisms

2967_Ch04.indd 118 07/07/2015 10:35

 Disease Pathogenesis 119

in the PTPN22 gene leads to lower Lyp expression, and thus to T, B, and dendritic cell
hyper-responsiveness, such as that seen in rheumatoid arthritis and T1D.
The genetic associations with the MHC and PTPN22 in rheumatoid arthritis are not
surprising as both are associated with a wide variety of other autoimmune diseases, includ-
ing T1D, systemic lupus erythematosus, and multiple sclerosis. In rheumatoid arthritis,
estimates suggest that together the MHC and PTPN22 account for as much as 50% of the
familial genetic risk of the disease.

A sample of recent genome-wide studies in rheumatoid arthritis

The 2007 WTCCC1 GWAS identified both the MHC and PTPN22 as major susceptibil-
ity loci in rheumatoid arthritis with probability values P < 7.5 × 10−27 and P < 5.6 × 10−25
respectively (Table 4.2). In addition, a further nine SNPs with moderate P values map-
ping to loci not previously associated with rheumatoid arthritis were identified. Among
these are SNPs that map close to the α- and β-chains of the IL-2 receptor (IL2RA and
IL2RB). The IL-2 receptor is very important in T cell stimulation and a number of SNPs
in this region have also been associated with T1D.
Seven new associations were reported in a more recent GWAS meta-analysis by Stahl et al.
(2010) involving 5539 cases and 20,169 controls, and a replication set of 6768 cases and
8806 controls. The new associations were with SNPs close to the genes encoding IL6ST
(IL-6 signal transduction), SPRED2 (sprout-related EVA domain containing 2), RBPJ
(regulation signal-binding protein for immunoglobulin κJ region), CCR6 (C-chemokine
receptor 6), IRF5 (IFN regulating factor 5), C5orf30 (chromosome 5 open reading frame
30), and PXK (PX domain containing serine/threonine kinase). The study also confirmed
the previously identified associations at 27 different loci including IL2RA (IL-2 receptor A),
CCL21 (C-chemokine ligand 21) and AFF3 (AF4/FMR2 family number 2). Interestingly,
the probability values associated with many of the risk loci in this study were higher than
those usually found in many GWAS (i.e. P > 10−7 or 10−8). For example the P values for
HLA-DRB1 and PTPN22 were P < 10−299 and P < 9.1 × 10−74 respectively. These higher P
values for these well known, strong associations reflect the larger numbers in the Stahl et al.
study. All of the associations were confirmed in the replication cohort. The most significant
associations from the WTCCC1 study and the Stahl study are also shown in Table 4.2.
Stahl et al. also listed a further 10 moderate risk alleles for rheumatoid arthritis (Table 4.3).
This latter group all have P > 5 × 10−7, i.e. they are outside the normal significance thresh-
old (hence the term moderate associations). The major gene groups identified in this
study are all associated with immune regulation. In addition, many of the identified SNPs
mapped to loci associated with other autoimmune disorders, suggesting common path-
ways in the genesis of rheumatoid arthritis (Figure 4.10).
New susceptibility alleles are continually being added to the list for rheumatoid arthritis.
A meta-analysis of six GWAS in 2011 identified seven new rheumatoid arthritis loci and
listed a total of more than 37 risk loci for rheumatoid arthritis. With at a least a further
nine risk loci being suggestive (moderate associations) from the three studies above, and
a very large number of confirmed and unconfirmed loci listed on the OMIM (Online
Mendelian Inheritance in Man) database, this brings the current total of risk alleles for
rheumatoid arthritis to more than 48.
The problem with information supplied in Tables 4.2 and 4.3 is the OR (risk) values for
the highly significant associations are in all but two obvious exceptions less than 1.5 (see
Figure 4.11). The critical question is how are these data to be used? Or perhaps even how
useful are these data? The answer is quite simple – they cannot at this stage be used to aid

2967_Ch04.indd 119 07/07/2015 10:35

 120 CHAPTER 4 Why Investigate Complex Disease Genetics?

Table 4.2 A short list of selected major risk genes for rheumatoid arthritis.

Gene/allele Location P value

HLA-DRB1 6p21 7.5 × 10−27
Increases to maximum 10−299 (OR >2.5)
PTPN22 1p13 5.6 × 10−25
Increases to maximum 10−74 (OR > 1.5)
IRF5 7q32 2.65 × 10−6
4.2 × 10−11 (OR = 1.25)
IL2RA 10p15 5.92 × 10−8
Increases to 1.4 × 10−11 (OR = 1.11)
TNFA1P3 (rs6920220) 6q23 1.58 × 10−5
Increases to 8.9 × 10−13 (OR = 1.4)
IL2RB 22q13 1.15 × 10−6
CTLA4 2q33 6.25 × 10−9 (OR = 0.85)
Lower in Stahl 1.2 × 10−8 (OR = 0.87)
STAT4 2q32 2.9 × 10−7 (OR = 1.16)
TRAF1/C5 9q33 2.1 × 10−7 (OR = 1.13)
CD40 20q13 2.8 × 10−9 (OR = 0,85)
CCL21 9p13 3.9 × 10−10 (OR = 0.87)
AFF3 2q11 1.0 × 10−14
REL 2p13 6.01 × 10−10
Lower in Stahl 7.9 × 10−7 (OR = 1.13)
BKL 8p23 5.69 × 10−9
Lower in Stahl 1.5 × 10−5 (OR = 1.12)
SPRED2 2p14 5.3 × 10−10 (OR = 1.13)
IL6ST/ANKRD55 5q11 9.6 × 10−12 (OR = 0.85)
C5orf30 5q21 4.1 × 10−8 (OR = 0.93)
PXK 3p14 4.6 × 10−8 (OR = 1.13)
RBPJ 4p15 1.06 × 10−16 (OR = 1.18)
CCR6 6q27 1.5 × 10−11 (OR = 1.11)

Where two sets of figures are presented, the top figures are from the WTCCC1 (2007) study (not italic) and the
lower figures are from Stahl et al. (2010) (italic). Where only one set is represented for a gene, not italic or italic
indicates the source as WTCCC1 or Stahl et al. respectively.

disease diagnosis and patient management, but they do point at the disease pathology. As
almost all of the risk alleles point to the immune system, this suggests that using genetic
associations to target and unpick specific immune pathways and/or correct defects in spe-
cific pathways may be the way forward. There are two elements to this: (1) identifying
pathways helps us understand the pathogenesis of this disease, and (2) by improving our
understanding of the disease, we may be able to produce better treatment and manage-
ment plans for patients. For example, we could consider the IL4–STAT6 pathway. These

2967_Ch04.indd 120 07/07/2015 10:35

 Disease Pathogenesis 121

Table 4.3 Weak (suggestive) genetic associations with rheumatoid arthritis from Stahl (2010) and listed
on OMIM.

Gene Location Probability (OR)

Other suggestive risk alleles (from Stahl et al. 2010)
IL6R 1q21 7.9 × 10−5 (OR = 1.13)
CD247 1q24 3.6 × 10−5 (OR = 0.9)
IL2/IL21 4q26 1.0 × 10−3 (OR = 0.89)
ZEB1 10p11 2.0 × 10−3 (OR = 0.93)
SH2B3 12q24 4.0 × 10−3 (OR = 0.93)
BATF 14q24 1.0 × 10−5 (OR = 1.16)
CD19/NFATC21P 16p11 5.3 × 10−5 (OR = 1.14)
IKZF3 17q12 4.7 × 10−5 (OR = 1.1)
UBE2L3 22q11 7.0 × 10−4 (OR = 1.1)
Listed on OMIM
CCP Not given
IL1A 2q13 Not given
NFKBIL1 6p21.33 Not given

two genes interact. IL-4 activates STAT-6, a signal transducer and activator of transcrip-
tion (type 6). This pathway has been found to be activated in patients with both short-
and long-term rheumatoid arthritis. If those carrying risk alleles could be given a dietary
supplement to enhance or reduce the activity in this pathway, then maybe it would be
possible to reduce or even eradicate rheumatoid arthritis. This may seem overly optimistic
considering the statement that Hugo Mencken applied to seeking solutions to complex
problems, and maybe we are over-inclined to seek simple solutions, but equally the best
solutions often turn out to be simple.

MHC
T1D
MS
SLE
PTPN22
CD6
T1D
MS
Crohn’s
Figure 4.10: Overlap between risk alleles of specific
Rheumatoid genes and different diseases focusing on rheumatoid
arthritis arthritis. This diagram illustrates some of the overlap
between associated genetic markers in rheumatoid
IL2RA arthritis and other diseases. Overlap is common among
IRF5
T1D complex diseases and it is helpful in identifying common
SLE
MS pathways in pathogenesis. Each of the major risk alleles
or genetic regions identified in the circles is associated
IL2/IL21 with rheumatoid arthritis. In each case, other diseases
Celiac that share some degree of overlap are illustrated, e.g.
IRF5 and systemic lupus erythematosus (SLE), and CD6
and multiple sclerosis (MS).

2967_Ch04.indd 121 07/07/2015 10:35

 122 CHAPTER 4 Why Investigate Complex Disease Genetics?

3.0

2.5

2.0

OR 1.5

1.0

0.5

MHC PTPN22 IRF IL2RA TNFA1P3 IL2RB

Figure 4.11: ORs for six of the most strongly associated genetic risk markers for rheumatoid arthritis. The
figure shows the small size of the effect for most risk alleles as determined by the OR. Figures vary from study
to study, and indeed even within studies, and therefore it is advisable to consider these as approximations or
estimates rather than absolute values, see Box 2.2.

The story of rheumatoid arthritis so far tells us that there are difficulties in reaching our
goals, but we should be optimistic about the possibilities of achieving at least one of the
aims of the HGP with regard to this disease, that of understanding pathways that lead to
pathogenesis. The problem is that we have more data than we can handle at present. In
one respect, it is reassuring that there are so many identified alleles because this will enable
us to explore multiple pathways and interactive networks. However, this takes time and
resources, and pathways and networks are complex. We need time to work on the data,
but more genetic data is piling up all around us as we work and it can be overwhelming.
The answer is currently out of our reach, but as we shall see in some of the other chapters
in this book, it is getting nearer by the day. It will take time to complete this exercise in
rheumatoid arthritis, but hopefully the final outcomes will be well worth the wait and the
investment in time and money.

Bipolar disease is a disease for which there are many weak genetic
associations, but few strong consistent associations
Bipolar disease is a manic depressive illness characterized by recurrent episodes of distur-
bance in mood, ranging from extreme elation or mania to severe depression. The pathogen-
esis of this disorder is poorly understood. There is clear evidence of a heritable component
with a sibling relative risk of 7–10 and heritability rated at 80–90%. Twins studies have
shown variable rates of concordance with values from 33–75% for monozygotic and
0–13% for dizygotic twins. Differences in the diagnostic criteria applied, a­ scertainment
bias, and selection of controls all help to account for this variation. However, heritability
is not just about genetics, it involves many other factors and thus high heritability does
not mean strong genetic associations.
Regions identified in early linkage and the 2007 WTCCC1 study
Several genome regions have been implicated in linkage studies (Table 4.4), and there is
clear evidence of overlap with genetic susceptibility to schizophrenia and bipolar disorder.

2967_Ch04.indd 122 07/07/2015 10:35

 Disease Pathogenesis 123

Table 4.4 Linkage studies for risk loci in bipolar disorder.

Gene/allele Location
MAFD1 18p
MAFD5 2q22–q24
MHW1 4p
MHW2 4q
MADF4 (region includes PALB2, NDUFABI, 16p12
and PCTN5)

MHW indicates mental health and welfare. MHW1 and MHW2 are based on linkage studies in the Amish. MAFD
is the name for bipolar disorder, based on manic affective disorder, then a number. Interestingly, OMIM reports
linkage on chromosomes 5, 11, and 18, which have not been identified in association studies.

Associations with DAOA (d-amino acid oxidase activator), DISC1 (disrupted in schizo-
phrenia 1), NRG (neuregulin 1), and DTNBP1 (dystrobrevin binding protein 1) have all
been reported. In the 2007 WTCCC1 study, the strongest signal was on chromosome
16p12 (Table 4.5). Genes close to this region with potential clinical relevance include:
• PALB2 (partner and localizer of BRCA2), which is involved in key structures within
cells including chromatin.
• NDUFAB1 [NADH dehydrogenase (ubiquinone) 1, α/β complex 1], which encodes a
mitochondrial respiratory chain subunit.
• DCTN5 (dynactin 5), which encodes a protein involved in intracellular transport that
is known to interact with the DISC1 protein, the gene for which is already implicated
in susceptibility to both bipolar disorder and schizophrenia.
Expanded analysis in the WTCCC1 study found strong associations with four regions
of the genome, one of which is close to the KCNC2 gene that encodes the Shaw-related
voltage-gated potassium channel. Changes in the dynamics of ion channel activity are
known to cause episodes of central nervous system disorders, including seizures, ataxia,
and paralysis. This may include the extreme mood changes seen in bipolar disorder.
Other highly ranked SNP signals indicated the importance of the γ-aminobutyric acid
(GABA) neurotransmission pathway identifying three signals in particular rs7680321,
rs1485171, and rs11089599, which are close to or within the genes for GABRB1 (encod-
ing the GABA A receptor β1), GRM7 (glutamate receptor, metabotropic 7), and SYN3
(synapsin III).
More recent GWAS
Following on from the 2007 WTCCC1 GWAS, additional studies of bipolar disorder
have identified novel/different SNP associations that include:
• DGKH (diacyclglycerol kinase).
• MYO5B (myosin 5B).
• NCAN (neurocan), which is an extracellular matrix glycoprotein involved in cell adhe-
sion and migration.
• MAD1L1 (meiotic arrest deficient like 1), which is important in cell division.

2967_Ch04.indd 123 07/07/2015 10:35

 124 CHAPTER 4 Why Investigate Complex Disease Genetics?

Table 4.5 The main findings from GWAS in bipolar disorder.

Gene(s) Location(s) P value (if available)

Major observations of bipolar disorder genes from 2007 WTCCC1 GWAS
MADF4 (region includes PALB2, 16p12 <5.14 × 10−8
NDUFABI, and PCTN5)
KCNC2 12q21 <2.18 × 10−7
Most likely SYN3 22q12 <1.61 × 10−7
GABRB1 4p12 Not listed
GRM7 3p25–26 Possibly <7 × 10−6 in WTCCC study
NRG1 8p12 <6.86 × 10−6
Genes identified in other GWAS of bipolar disorder
DGKH 14q11 Not listed
MYOB 2q32.3 Not listed – this could be due to linkage
disequilibrium with risk allele(s) in 2q33
– WTCCC1 study above identified tagged
SNP <5.4 × 10−5 as closest.
HLA C*01:02 6p21.3 Not listed
TRIM26 6p22.1 <5 × 10−16 in schizophrenia
Identified from meta-analysis of GWAS
CACNAIC 12p13.3 <7 × 10−8
ANK3 10q21.2 <9.1 × 10−9
JAM3 11q25 <1 × 10−6
SLC39A3 19p13.3 <5 × 10−6
NCAN 19p13.11 <2.14 × 10−9 (OR = 1.17)
MAD1L1 7p22.3 <1.28 × 10−7 (OR = 1.6)
PBRM1 3p21.3 <2.68 × 10−9 (OR = 0.875)
ODZ4 11q14.1 <4.4 × 10−8 (OR = 0.89)

OMIM reports linkage on chromosomes 5, 11, and 18 which have not been identified in association studies. The
WTCCC1 study data can be read in several ways; there are major associations and moderate associations, and
associations only reported in the supplementary texts. In this chapter, the major associations are listed together
with some of those from the supplementary text that could be identified by gene name, this leaves 13 listed as
moderate unaccounted for in T4.4 (these are on chromosomes 1p31, 2p25, 2q12, 2q14, 2q31, 2q37, 3q37,
6p21, 6q22, 9q32, 14q22, 14q32, and 20p13). It is assumed that the tagged SNPs for 3p23, 8p12, and 16p12 are
markers for GRM3, NRG1, and PALB2, respectively. DAOA, DTNBP1, and DISC are also not listed in T4.3 as
they fail to reach minimum values for moderate association P < 10−5.

The DGKH and MY05B genes are not currently listed as potential candidates in bipolar
disorder, and interestingly these post-WTCCC1 studies did not immediately replicate the
findings of the WTCCC1 study itself. Altogether, it is quite challenging trying to find a
consistent message when reading through all of the papers on genetic association studies
in bipolar disorder. On first review of the data there appears to be no consistency between

2967_Ch04.indd 124 07/07/2015 10:35

 Disease Pathogenesis 125

them. However, further analysis, including meta-analysis, has revealed some important
new and consistent associations. These include novel associations with the CACNA1C
(calcium channel, voltage dependent, L-type α 1C subunit), and ANK3 [ankryin 3, node
of Ranvier (ankryin G)] genes, and confirmation of the NCAN association.
There are many reasons for these reported differences between initial studies and the
meta-analysis studies. Bipolar disorder is not easily diagnosed, the genetic associations
with the disease are quite weak, and some of the studies have been based on quite small
case cohorts. Meta-analysis can iron out some of these wrinkles. One particular study
that stands out and illustrates this well is that of Ferreira et al. (2008). In a GWAS of
1098 cases, Ferreira et al. first reported no major associations with bipolar disorder. The
authors decided that this inconsistency may have been due to the small number in their
own study cohort or the use of different tagged SNPs for their analysis. However, the
authors were not discouraged. They decided to perform further analysis focusing on areas
of known association from other studies, especially the recently identified weak associa-
tion with the CACNA1C gene that had been identified by combining two data sets in a
meta-analysis. When the authors fed their own data into a meta-analysis with the two
previous studies, there was a weak association with the CACNA1C gene. In addition, by
referencing the existing data to HapMap they were able to impute the genotypes for each
case using PLINK software. This software uses the expected pattern of linkage disequilib-
rium to assign haplotypes and increases the genome coverage—in this case from approxi-
mately 350,000 to over 1.5 million SNPs (see Chapter 12). Imputation is increasingly
used on GWAS, provided there is an adequate database of fully genotyped individuals
on which to make the haplotype assumptions for the test sample. In this case, assigning
non-genotyped alleles to the test samples on the basis of known linkage disequilibrium
is a very profitable exercise, reducing cost and increasing efficiency. However, the quality
of the study relies on the control data bank. This analysis identified the ANK3 gene on
chromosome 10q21 as the major susceptibility gene in bipolar disorder, with CACNA1C
as the second strongest.
It must be stressed that many of these associations with bipolar disorder did not reach
the significance threshold of P < 10−7 in any of the initial analyses or for any individual
group. Only the combination of the data provided P values of the required acceptable
magnitude.
The future of genetic studies in bipolar disorder
Clearly, much more needs to be done in bipolar disorder, but taken together the observa-
tions identify a number of potentially important pathogenic pathways in bipolar disor-
der, particularly the GABA neurotransmission pathway. These studies also indicate the
potential for polymorphisms in the genes involved in calcium channel activity essential
for neurotransmitter release and genes from the solute carrier family (CACNA1C and
SLC39A3, respectively) to be functionally associated with bipolar disorder. The SLC39A3
gene encodes a solute carrier family 39 (zinc transporter) member 3. The association with
GRM7 illustrates the possibilities of unraveling the complexities of genetic links in bipolar
disorder. GRM7 encodes a member of a family of glutamate receptors. l-Glutamate is a
major excitatory neurotransmitter in the central nervous system. l-Glutamate activates
glutamate receptors such as GRM7 and is active during most normal brain activities.
GRM7 is one of a family of metabotropic G-protein-coupled receptors that has been
linked to the inhibition of the cyclic AMP cascade. It is quite easy to consider how poly-
morphisms that regulate this receptor could have functional effects in terms of neurotrans-
mission and how the downstream consequences of such interpersonal variation may lead
to a trait such as bipolar disorder (Table 4.6).

2967_Ch04.indd 125 07/07/2015 10:35

 126 CHAPTER 4 Why Investigate Complex Disease Genetics?

Table 4.6 The function of selected genes associated with increased risk of bipolar disorder.

Gene/allele Function
PALB2 Partner and localizer for stabilization of BRCA2. Acts as a molecular
scaffold attracting BRCA1 and RAD51 ensuring stabilization of the
BRCA1–PALB2–BRCA2 complex, which is required for homologous
recombination.
KCNC2 Potassium-gated voltage channel Shaw-related family member 2. As the
name implies, this mediates potassium ion permeability across the cell
membrane.
NRG1 Glial cell growth factor interacting with tyrosine kinase receptors to recruit
ERBB1 and ERBB2 co-receptors resulting in ligand-stimulated tyrosine
phosphorylation and activation of ERBB receptors. Involved in induction
of growth of glial and neuronal cells, among other cells Hence, early
identification as a glial cell growth factor.
GABRB1 γ-Aminobutyric acid A receptor β1. Encodes a subunit of a chloride
channel that mediates rapid inhibition of synaptic transmission in the
central nervous system. This gene is strongly associated with schizophrenia.
GRM7 Metabotrophic G-coupled-protein receptor for glutamate associated with
inhibition of the cyclic AMP cascade.
SYN3 May be involved in the regulation of neurotransmitter release and genesis
of synapse.
CACNAIC Calcium channel voltage-dependent type 1 α1 C subunit gene. Mediates
influx of calcium into cells and is involved in neurotransmission.
ANK3 Ankyrin G is specifically found at a neuronal junction or axonal segment
known as the node of Ranvier in the central and peripheral nervous
systems. It is one of a group of proteins thought to be associated with
various functions in the cytoskeleton.
JAM3 Junctional adhesion molecule 3. Promotes cell–cell adhesion.
SLC39A3 Solute carrier family member 39 family 3. Zinc transporter. Another
member of the SLC family, SLC6A4, may be important in serotonin
release.
NCAN Neurocan. A chondroitin sulfate proteoglycan that is thought to be
involved in control of cell migration and adhesion. Studies show that it is
expressed in the hippocampus of mice and humans.
MAD1L1 Meiotic arrest deficient-like 1. Important in cell division and regulation of
processes in meiosis.
PBRM1 Polybromo-1. Involved in transcriptional activation and repression of select
genes by chromatin remodeling. It is thought to act as a negative regulator
of cell proliferation.
ODZ4 The ODZ4 gene encodes a human homolog of the Drosophila
pair‑rule gene Tenm (odz). The gene product may function as a signal
transducer.

2967_Ch04.indd 126 07/07/2015 10:35

 Disease Pathogenesis 127

Functional studies on human tissue samples have shown that the NCAN and MAD1L1
genes are both expressed in the hippocampus, and studies on mice show NCAN to be
present in the cortical and hippocampal areas of the brain that are involved in cognition
and regulation of emotions.
The study examples above also indicate that similar diseases can share risk alleles, but they
also illustrate some of the difficulties of working with disorders that are not easily defined,
such as bipolar disorder, compared with working with a disease with more physical mani-
festations, such as rheumatoid arthritis. Altogether, these studies serve as reminders about
study planning, design, and the limitations of association studies (discussed in Chapter
3 and 5). Finally, these findings also illustrate the importance of taking a holistic view
and not dismissing non-genetic factors. Many factors are known to contribute to bipolar
disorder, not all of them are genetic (Figure 4.12).

Coronary artery disease is the most common cause of death in the

developed world
The environment is known to play a major role in coronary artery disease, but genetic
variation is also thought to have an impact. It is the most common cause of death in the
developed world and second most common cause in medium- to low-income countries.
Most patients develop atheromatous plaques on the walls of the coronary arteries, and
this causes angina and myocardial infarction (heart attack). The plaques are made up of
a lipid and fibrous matrix. The pathogenesis of the disease is complex with several known
contributors, including endothelial cell dysfunction, oxidative stress, and inflammation, as
well as environmental factors such as smoking and diet. The sibling relative risk for cardio-
vascular disease is quite low, between 2 and 7, reflecting the strong environmental input
into the disease pathogenesis. In contrast to this, data from a study of 20,000 twins in the
Swedish Twin Registry suggest that heritability for angina in men may be as high as 39%
and death from coronary disease as high as 57%, with figures of 43 and 38% for women.
This suggests that heritability is important and contributes to cardiovascular disease, but it
is not as clear-cut as seen in other diseases. Once again this reminds us that the term herita-
bility has multiple uses and does not simply refer to our genes. Heritability can include the

Bipolar
disorder

Family life Nutrition Stress Genetics

Figure 4.12: Different potential causes of bipolar disorder. The figure illustrates probable contributing
factors to bipolar disorder. They may act individually or in combination. There is no restriction on the number
of elements active. For example, stress in a genetically susceptible individual may trigger bipolar disorder, but
nutritional status may also be a factor. However, it must be noted these are only possible relationships and not
definite relationships.

2967_Ch04.indd 127 07/07/2015 10:35

 128 CHAPTER 4 Why Investigate Complex Disease Genetics?

CHD Cancer

CVD Other

Men Stroke Women

Figure 4.13: Impact of coronary artery disease based on figures from the British Heart Foundation. These
figures for the UK population in 2008 show morbidity for coronary heart disease, cardiovascular disease, stroke,
and cancer against other diseases and pathologies in men and women. It is quite clear that coronary artery
disease-related conditions [coronary heart disease (CHD), cardiovascular disease (CVD), and stroke] account for
a significant portion of deaths considering that coronary heart disease, cardiovascular disease, and stroke may
all be the consequences of coronary artery disease. Compare the figures given for these three problems with
diabetes, which is considered a major scourge and accounted for approximately 1% of deaths in 2008 in the UK.
In comparison, the figures for coronary heart disease, cardiovascular disease, and stroke were 18, 8, and 7% in
men (total 33%) and 13, 12, and 8% in women (total also 33%) that same year.

environment. As cardiovascular disease is a common cause of mortality (Figure 4.13), the

search for risk alleles remains as intense as that for diseases with a greater and more obvious
genetic component.
Genome-wide studies in cardiovascular disease can be divided into those produced prior
to the publication of the WTCCC1 GWAS in 2007 and those that followed.
Studies of cardiovascular disease prior to 2007
Though the potential promise of GWAS in cardiovascular disease was well established
before 2007, there were only two GWAS performed prior to 2007 that identified putative
candidates. These two candidate genes were ALOX5AP on 13q12 [arachidonate 5-lipoxy-
genase-activating protein (also known as FLAP), with probability values around 2 × 10−6
for males and 24 × 10−4 for females] and LTA4H on 12q22–q23 (leukotriene A4 hydro-
lase, with OR 1.16 in Europeans and 3.57 in African-Americans) genes. The product of
the ALOX5AP gene is important in leukotriene synthesis. Leukotrienes are involved in
immune inflammation and vascular adhesion. Leukotriene A4 hydroxylase is an epoxide
hydrolase that catalyses the final step in the biosynthesis of the pro-inflammatory media-
tor leukotriene B4—a leukotriene involved in cell adhesion via the STB4 biosynthesis
pathway. Both of these genes are plausible candidates as risk markers for cardiovascular
disease. However, the WTCCC1 report did not find any strong associations with these
genes and nor did any other later GWAS report associations at these sites (Table 4.7).

2967_Ch04.indd 128 07/07/2015 10:35

 Disease Pathogenesis 129

Table 4.7 Risk loci for coronary artery disease identified in the 2007 WTCCC1 GWAS.

Gene Location P value and Odds Ratio (OR)

CDKNA2/CDKN2B 9p21.3 1.79 × 10−14 (OR = 1.47)
(close to MTAP) (rs1333049) (9p21.3) (1.75 × 10−16)
ADAMTS17 15q24–q26 1.1 × 10−4 (1.6 × 10−7)
(close to ADAMTS7 at 15q25.1 rs19994016)
MTHFD1L (rs6922269) 6q25.1 6.33 × 10−6 (OR = 1.17)
(rs17672135) 1q43 2.35 × 10−6
(close to MIA3 at 1q41 rs17465637)
(rs383830) 5q23 1.34 × 10−5
(note association with 5q31 C4D group)
(rs8055236) 16q23 5.6 × 10−6
(rs7250581) 19q12 2.5 × 10−5
(note APOE at 19q13)
(rs688034) 22q12 3.75 × 10−6

The lower case “rs” denotes the tagged SNP for the region. Not all potential regions identified were associated
with specific genes in the WTCCC1 study. Many potential candidates were suggested following further analysis.
Note association with ALOX5AP and LTA4H identified as candidates prior to 2007 were not confirmed in the
WTCCC1 study. The values quoted depend on the reading of the paper and which table they were extracted from,
as a range of possibilities are given in the different analyses. Note that not all associations were replicated in later
studies.

The WTCCC1 study was also unable to replicate previously reported associations with the
APOE gene. This problem may have been due to the fact that the tagged SNPs on the
Affymetrix gene chip that was used in the study were not good markers for identifying
APOE compared with direct genotyping. Interestingly, this explanation appears to be sound
because studies in 2011 confirmed the association with APOE in cardiovascular disease.
Considering the function of APOE, which mediates the binding, internalization, and catab-
olism of lipoprotein particles, this gene makes a very likely candidate. The APOE protein
can serve as a ligand for the low-density lipoprotein receptor as well as interacting with the
APOE receptor.
The 2007 WTCCC1 study and cardiovascular disease
The most important single observation on cardiovascular disease in the WTCCC1
study was a new and strong (P < 10−14) association with a SNP on chromosome 9p21.3.
The region contains genes for two cyclin-dependent kinase inhibitors (CDKN2A and
CDKN2B). Both genes are widely expressed and are involved in the regulation of the cell
cycle. CDKN2B is expressed in macrophages but not smooth muscle cells with fibrous
lipid lesions. CDNK2B expression is induced by transforming growth factor-β, a signal-
ing system that has been associated with cardiovascular disease. In addition to the two
CDNK genes, the SNP on 9p21.3 maps close to the MTAP gene, coding for a methyl-
thioadenosine phosphorylase enzyme that contributes to processes involved in adenine
and methionine salvage. This is a ubiquitously expressed gene present in the cardiovascular
system and it is also listed as a candidate for cardiovascular disease. Genetic variation in
any one or all of these genes could contribute to the risk of cardiovascular disease. The

2967_Ch04.indd 129 07/07/2015 10:35

 130 CHAPTER 4 Why Investigate Complex Disease Genetics?

association with CDNK2A–CDNK2B remains the strongest reported and most consistent
association for cardiovascular disease.
The other associations with cardiovascular disease reported in the 2007 WTCCC1 study
include modest associations with a SNP (rs6922269) within the MTHFD1L gene [meth-
ylene tetrahydrofolate dehydrogenase (NADP+-dependent) 1-like]. This gene encodes the
mitochondrial isozyme of C1 tetrahydrofolate (THF) synthase, which converts the single
carbon units carried in folic acid. C1 THF synthase is used in several biological processes,
including purine synthesis. Cardiovascular disease has also been associated with mutations
in another THF enzyme, i.e. THF reductase (encoded by MTHFR). Interestingly, both
MTHFD1L and MTHFR activity can influence plasma levels of homocysteine, indicating
a common pathway for cardiovascular disease. One other gene identified in the expanded
analysis as having a major risk for cardiovascular disease is the ADAMTS17 (a disintegrin-
like and metalloproteinase with thrombospondin type 1 motif 17). This gene encodes a
protein involved in vascular extracellular matrix degradation and remodeling—a process
central to atherosclerosis.
GWAS in cardiovascular disease beyond 2007
Since the publication of the WTCCC1 GWAS there have been many other studies of car-
diovascular disease using GWAS. Two studies published in 2011 stand out. The first study
was a meta-analysis of data from 14 GWAS comprising 22,233 cases and 64,762 controls
with a replication cohort of 56,682 cases performed by the CARDIoGRAM Consortium.
In this analysis, 13 loci were identified for the first time as risk loci for cardiovascular dis-
ease and 10 of 12 previously reported risk loci were confirmed. The data from this study
are summarized in Table 4.8. The authors of this study found that only three of the new
loci were associated with traditional markers for cardiovascular disease, with the remain-
der being in gene regions not previously implicated in the pathogenesis of cardiovascular
disease. The report also highlights the overlap between risk markers as five of the new loci
have strong associations with other diseases or traits.
The second study in 2011 (also summarized in Table 4.8) used the data from the study
above and a second data set produced by the C4D Consortium based on approximately
40,000 cases. This combined report identified 35 common gene variants at 34 loci as risk
factors for cardiovascular disease and suggested that these common variants account for
only 13% of the overall genetic risk of cardiovascular disease (Table 4.9).
The reports themselves are too detailed to reiterate every fact in this chapter, but they raise
a number of important issues for anyone looking at published data or about to set out on
a GWAS.
Many of the associations were not replicated
A review of the report of the 2007 WTCCC1 study in the light of the 2011 studies
shows that only one of the associations for cardiovascular disease reported in 2007 has
been replicated and all of the others are not confirmed (ADAMTS7, MTHFD1L, and
MTHFR in particular) at a significance threshold of P > 5 × 10−8. One explanation for
these failures in replication is the application of different gene chips or the closer asso-
ciation with tagged SNPs in a different location which may be identified in subsequent
studies, but not identified (or perhaps even tested) in the initial study. However, all is not
lost as some of this discordance may be explained through linkage disequilibrium whereby
SNPs associated with CAD in original studies are in close linkage with other loci that have
themselves been claimed as risk genes for cardiovascular disease. For example: MIA3 (mel-
anoma inhibitory activity family 3) on 1q41, APOE on 19q13, and ADAMTS17 (ADAM

2967_Ch04.indd 130 07/07/2015 10:35

 Disease Pathogenesis 131

Table 4.8 Risk loci identified in the 2011 CARDIoGRAM study.

Gene(s) Location P value (OR)

Confirmed
PCSK9 1p32.3 9.1 × 10−8 (OR = 1.15)
SORT1 1p13.3 2.89 × 10−10 (OR = 1.29)
MIA3 1q41 1.36 × 10−8 (OR = 1.2)
WDR12 2q33.1 1.12 × 10−9 (OR = 1.16)
MRAS 3q22.3 3.34 × 10−8 (OR = 1.15)
PHACTR1 6p24.1 1.15 × 10−9 (OR = 1.13)
LPA 6q25.3 3.0 × 10−11 (OR = 1.92)
CDKN2A/CDKN2B 9p21.3 1.35 × 10−22 (OR = 1.25/1.37)
Identified in WTCCC
CXCL12 10q11.21 2.93 × 10−10 (OR = 1.33)
Identified in MIGen Consortium
SH2B3 12q24.12 6.35 × 10−6 (OR = 1.13)
LDLR 19p13.2 9.73 × 10−10 (OR = 1.14)
Identified in MIGen Consortium
MRPS6 21q22.11 4.22 × 10−10 (OR = 1.19)
Novel associations
PPAP2B 1p32.2 3.81 × 10−19 (OR = 1.17)
ANKS1A 6p21.31 1.36 × 10−8 (OR = 1.07)
TCF21 6q23.2 1.07 × 10−12 (OR = 1.08)
ZC3HC1 7q32.2 9.18 × 10−18 (OR = 1.09)
ABO 9q34.2 4.08 × 10−14 (OR = 1.10)
CYP17A1/CNNM2/NT5C2 10q24.32 1.03 × 10−9 (OR = 1.12)
ZNF259/APOA5-A4-C3-A1 11q23.3 1.02 × 10−17 (OR = 1.13)
COL4A1/COL4A2 13q34 3.84 × 10−9 (OR = 1.07)
HHIPL1 14q32.2 1.14 × 10−10 (OR = 1.07)
ADAMTS7 15q25.1 1.07 × 10−12 (OR = 1.08)
SMG6/SSR 17p13.3 1.15 × 10−9 (OR = 1.07)
RASD1/SMCR3/PEMT 17p11.2 4.45 × 10−10 (OR = 1.07)
UBESZ/GIP/ATP5G1/SNF8 17q21.32 1.81 × 10−8 (OR = 1.06)

The CARDIoGRAM study data presented here indicates confirmation of 12 locations and 13 novel genes with
P <10−8, with one exception (SH2B3 at 6.35 × 10−6).

2967_Ch04.indd 131 07/07/2015 10:35

 132 CHAPTER 4 Why Investigate Complex Disease Genetics?

Table 4.9 Risk loci identified in other GWAS (excluding WTCCC and CARDIoGRAM reports shown
in T4.7 and 4.8 respectively), but included in the list of 35 loci reported by Peden and Farrall
(2011).

Locus Location OR (Source)

ABCG8 2p21 OR = 1.09 (MIGen Consortium)
IL5 5q31 OR = 1.02 (C4D Consortium 2011)
7q22 7q22 OR = 1.08 (C4D Consortium 2011)
TRIB1 8q24.13 OR = 1.04 (C4D Consortium 2011)
KIAA1462 10p11.23 OR = 1.07 (C4D Consortium 2011)
LIPA 10q23.31 OR = 1.08 (C4D Consortium 2011)
PDGFD 11q22.3 OR = 1.08 (C4D Consortium 2011)
APOE 19q13 OR = 1.14 (C4D Consortium 2011)
MRPS6 21q22.11 OR = 1.18 (MIGen Consortium)

The data presented here represent those loci not listed in T4.7 and 4.8. Interestingly, the list includes APOE,
which was not reported by the WTCCC1 2007 study, and IL5. Both of these gene loci have been associated with
susceptibility to other diseases.

metalloproteinase with thrombospondin type 1 motif 17) on 15q25. Interestingly,

ADAMTS7 and ADAMTS17 map close to each other, and it would be easy to make a false
assumption that an association with a SNP on chromosome 15q25.1 was due to linkage
with ADAMTS7 rather than with ADAMTS17. This illustrates the importance of search-
ing an area thoroughly and in fine detail. Weak or moderate associations are still at a risk
of being overlooked, but it also is possible to make false assumptions through accepting
weak associations.
There is at least one very strong and reproducible genetic association with cardiovascular
­disease—that with chromosome 9p21.3 and the CDKN2A–CDKN2B region. Here, the
probability may be as high as 1.35 × 10−22. However, the quoted probability values vary from
study to study and even within studies, especially where numerous analyses are performed.
In summary:
• The impact of these genetic polymorphisms is small. All of the OR values indicate
small effects. It is important to remember the role of environmental factors in cardio-
vascular disease and consider OR values for gene polymorphisms in that context.
• There is now a mass of data available and not all of the genes identified indicate obvi-
ous candidates for cardiovascular disease. This is important because identifying novel
candidates may help to unravel the pathogenesis of cardiovascular disease.
• All of the studies above relate to European patients. However, it is important to con-
sider other populations.
Genetic studies of cardiovascular disease in different populations
As with the European population, there have been many studies of the genetics of car-
diovascular disease in populations outside Europe. One study in particular stands out.
Published in 2012, a study of the Han Chinese population based on 1515 cases and 5159
controls with a replication cohort of 15,460 cases and 11,472 controls identified four

2967_Ch04.indd 132 07/07/2015 10:35

 Disease Pathogenesis 133

new loci at the 5 × 10−8 threshold of significance and confirmed the associations with
CDKN2A–CDKN2B reported in the European population, as well as the associations
with PHATCTR1 (phosphatase and actin regulator 1), TCF21 (transcription factor 21),
and C12orf51 (chromosome 12 open reading frame 51).
On close reading, the Chinese study illustrates a number of important issues in this area
of research. Most important among these is the observation that not all of the associations
reported in Europeans are replicated in the Chinese population (Table 4.10). This popula-
tion variation is well demonstrated by the SNPs that identify common genetic variations
in the 12q24 region. The variations in this region that are common in Europeans are not

Table 4.10 Risk loci for coronary artery disease reported in the Han Chinese population.

Gene(s) and tagged SNP Location P value (OR)

Han Chinese 2012
TTC32–WDR35 2p24.1 6.83 × 1−11 (OR = 1.12)
(rs2123536)
GUY1A3 4q32.1 1.26 × 10−11 (OR = 1.14)
(re1842896)
6orf10–BTNL2 6p23.32 2.77 × 10−15 (OR = 1.16)
(rs9268402)
APT2B1 12q21.33 5.68 × 10−10 (OR = 1.11)
(rs7136259)
Other Chinese studies
C6orf105 6p24.1 Androgen-dependent TFPI-
(rs6903956) regulating protein
Strong association reported in Europeans and confirmed in the 2012 Han Chinese Study
PHACRT1 6p24.1
TCF21 6q23
CDKN2A–CDKN2B 9p21.3
C12orf51 12q24.13
Weak but consistent other European (confirmed in the 2012 Han Chinese study)
PPAP2B 1p32.2
MIA3 1q41
LIPA 10q21.31
CYP17A1/CNNM2/NT5C2 10q24.32
ZNF259/APOA5-A4-C3-A1 11q22.3
ADAMTS7 15q25.1
SMG6/SSR 17p13.3

This table illustrates the importance of studying several different populations. The listed risk loci for
coronary artery disease in the Han Chinese population includes some loci also identified as risk loci in
Europeans and some only found in the Han Chinese. Not all the loci identified in Europeans are confirmed in the
Han Chinese.

2967_Ch04.indd 133 07/07/2015 10:35

 134 CHAPTER 4 Why Investigate Complex Disease Genetics?

found in the Chinese (i.e. the Han population appears to be monomorphic with respect
to these particular variants), whereas all of the variants at 12q24 reported in the Han
Chinese appear to be monomorphic in the European population. Population variation is
a major factor in genetic association studies. Even where there is general agreement about
associations, there are often subtle differences between populations. This can be seen with
respect to the CDNK2A–CDNK2B loci, where the pattern of linkage disequilibrium var-
ies between the Han Chinese and the European populations even though both popula-
tions carry genetic associations with the CDNK2A–CDNK2B haplotype. However, not all
associations are restricted to single populations and there is considerable overlap here. In
addition to the four confirmed associations reported, the Chinese study investigated 29
different cardiovascular disease associated loci identified from European studies and found
a further seven with consistent nominal associations (1p32.2, 1q41, 10q23.31, 10q24.32,
11q22.3, 15q25.1, and 17p13.3). Furthermore, 11 of the SNPs (at 10 loci) associated
with cardiovascular disease in Europeans were monomorphic in the Han Chinese popula-
tion, and use of proxy SNPs as substitutes for the monomorphic SNPs identified three
further associations at 3q22.3, 6p26, and 17p11.2. In their conclusion, the authors stated
that both shared and unique genetic associations occur in this complex disease. This idea
of a mixture of shared and unique susceptibility alleles in different populations is funda-
mental in complex disease. Comparison of populations may be very fruitful in helping us
to understand diseases pathogenesis. In this compare and contrast model, comparing
and contrasting data may pull out the real plums.

Genetic polymorphism in cardiovascular disease and disease pathogenesis

What does all this information (so far) tell us about the pathogenesis of cardiovascular
disease and how does it direct functional research? It is quite interesting to create pictorial
views of how these genes may interact to produce a disease portfolio for cardiovascular dis-
ease. We can do this by looking at the function of the different genes associated with cardio-
vascular disease listed in Tables 4.7–4.10. There are obvious groups of genes (Figure 4.14):
• Genes associated with the immune response through inflammation and vascular adhe-
sion and cell migration, e.g. CXCL12, SH2B3, IL5, PDGFD, and 6orf10–BTNL2.
• Genes associated with lipid metabolism and cholesterol homeostasis, e.g. PCSK9,
LIPA, LDLR, APOA5-A4-C3-A1, ABCG8, APOE, and APT2B1.
• Genes associated with cell growth, e.g.CDKN2A and CDKN2B.
• Genes associated with fibrosis and regeneration, e.g. MIA3, COL4A1, COL4A2, and
ADAMTS7.
• Genes associated with coagulation and anticoagulation activity, e.g. C6orf105 and
ABO.
• Genes associated with other activities, including formation of cellular particles
(WDR12), different kinase pathways (MRAS), phosphatase activity (PHACTR1),
mitochondria (MRPS6), cell differentiation (TCF21), a number of zinc finger genes
(ZC3HC1 and ZNF259), a number of cytochromes (CYP17A1), ATP synthatase
(ATP5G1), and nitric oxide signaling (GUY1A3).
The associations that have been reported in cardiovascular disease are mostly weak. On
a good day it is easy to look at these lists and identify obvious targets for further genetic
studies, and in some cases functional studies, links with lipid metabolism, and choles-
terol homeostasis, for example. However, it is important to be mindful of seeking simple
solutions to complex problems. There is no doubt that strong associations that have been

2967_Ch04.indd 134 07/07/2015 10:35

 Disease Pathogenesis 135

APOE
IL5 MIA3
APOA5-A1
CXCL12 ADAMTS7
PSK ABO
SH2B3 COL4A1
LIPA
PDGFD COL4A2
LDLR

Cholesterol Immune
and lipid regulation Coagulation Fibrosis
cells T cells

CAD
Figure 4.14: Systems and pathways in coronary artery disease (CAD). This simple picture illustrates the
broad concept that identifying risk genes may help us understand disease pathology. Some of the candidate
alleles identified (e.g. cholesterol and lipid) are quite obvious candidates for coronary artery disease. It is not
possible to list all of the susceptibility alleles nor is it yet possible to make a clear link between some of the
identified associations and disease pathology. Genes that are associated with cell growth (e.g. CDKN2A and
CDKN2B) are not included here nor are genes that are associated with other activities, including formation of
cellular particles (WDR12), different kinase pathways (MRAS), phosphatase activity (PHACTR1), mitochondria
(MRPS6), cell differentiation (TCF21), a number of zinc finger genes (ZC3HC1 and ZNF259), a number of
cytochromes (CYP17A1), ATP synthetase (ATP5G1), and nitric oxide signaling (GUY1A3). Finally, this illustration
does not take account of the strong environmental impact in coronary artery disease.

confirmed are the best place to start when we begin to construct a picture of the disease
pathology. These genetic associations are after all the most reliable and most frequently
confirmed, and they also tend to be those which are seen in multiple populations. It is
more difficult to link function to other genes (e.g. the zinc finger genes) and cardiovas-
cular disease, but these too are also associated with an increased risk of cardiovascular
disease and therefore a better understanding of genotype–phenotype relationships is
required.
Shared risk alleles between cardiovascular disease and other diseases
We should also not overlook the genetic similarities between diseases. Autoimmune dis-
eases share risk alleles. Genetic variants associated with the immune response play a major
role in many diseases and often the same allele is listed in two or more. With reference to
cardiovascular disease, it is tempting to look for links with T2D (Chapter 10). However,
despite there being considerable overlap between cardiovascular disease and T2D, studies
so far have identified very few shared risk alleles. The two genes that have been identified
in both T2D and cardiovascular disease are CDKN2A, which encodes a cyclin-dependent
kinase inhibitor on chromosome 9p21.3, and SH2B3 on chromosome 12q13. CDKN2A
is one of two cyclin-dependent kinase inhibitors associated with CAD with cardiovascular
disease. SH2B3 maps close to a tyrosine kinase precursor gene ERBB3. A functional link
is yet to be established between these genes and either of the two diseases, but associa-
tions shared between diseases are always interesting because they may indicate a common
functional link in the disease pathology. Looking at cardiovascular disease and T2D, it is

2967_Ch04.indd 135 07/07/2015 10:35

 136 CHAPTER 4 Why Investigate Complex Disease Genetics?

surprising to see so few shared associations. However, it is possible that there are shared
associations out there, and with larger studies and higher-resolution genotyping or even
direct sequence analysis they will be cataloged eventually.

4.5 WHAT ABOUT THE OTHER DISEASES?

It would be wrong to give the impression that the five disease examples in this chapter are
the only diseases that have been studied in this way. As the book shows us, there are many
more diseases under investigation, though even this is not enough. In 2012, 1350 GWAS
had been published reporting data on at least 421 different traits. In the last quarter of
2013 the total number of published GWAS reports was 1960. This list is updated regularly
by the National Human Genome Research Institute at the National Institutes of Health
website (http://www.genome.gov/gwasstudies). Some of the more interesting associations
described include those we have discussed already (Alzheimer’s disease and Crohn’s dis-
ease, Chapter 2) or will be discussing later in this book (cancer in Chapter 9 and diabetes
in Chapter 10). However, we cannot consider 421 or more different diseases here or even
in the whole book. That is not our purpose and therefore our consideration will be limited
to selected examples. These will be used to illustrate concepts and ideas as well as provide
knowledge on specific diseases. This discussion and the lists that accompany the examples
are intended to be a starting point for further exploration by students of this topic and are
by no means comprehensive. The intention is to provide examples that illustrate concepts
that are transferable from one disease to another. It is also important to consider the over-
lap between disease and between populations, where possible (Figure 4.15). Associations
that persist in different populations can be more informative than those that are popula-
tion specific, but not always.

Population 1 Population 2

A C B

Figure 4.15: Similarities and differences in risk alleles in different populations with the same diseases.
The figure illustrates the genetic risk portfolio for a disease in two populations: population 1 and population 2.
The smaller oval shape illustrates the disease population for both populations 1 and 2. The oval is subdivided into
three regions. Those in areas A and B share no risk alleles between the two populations, while those in areas C
share risk alleles.

2967_Ch04.indd 136 07/07/2015 10:35

 Conclusions 137

Conclusions
Considerable scientific resources, both financial and intellectual, have been ploughed into
studying the genetics of complex disease. One may ask the question why? After all, com-
plex diseases are not simple, they do not lend themselves to simple analysis, and they do
not conform to Mendelian patterns of inheritance. The answer lies with the promises
of the HGP. Under this banner, three promises can be considered: (1) identifying risk
alleles will aid disease diagnosis, (2) identifying risk alleles will aid patient treatment/
management and care, and (3) identifying risk alleles will inform the debate on disease
pathogenesis and may ultimately lead to the development of new therapies. Clearly these
outcomes are all linked and though at present there is less potential to use risk alleles in
disease diagnosis, there is great potential to use the new genetics in patient management,
including the potential for developing personalized therapy. However, so far, the greatest
success has been with the third promise.
This is illustrated through the selected disease examples discussed above: ankylosing
spondylitis, HCV, rheumatoid arthritis, bipolar disorder, and cardiovascular disease. In
ankylosing spondylitis, the idea of molecular mimicry based on the shared sequence of
self-peptides and those of pathogens was proposed long before the publication of the
map of the human genome, but recent GWAS have led to the identification of mul-
tiple immunoregulatory genes, all of which helps to support the concept of this being an
immune-mediated or autoimmune disease and will enable us to dissect the pathogenesis
of this disease. In fact, significant progress is being made in ankylosing spondylitis fol-
lowing recent GWAS. As with ankylosing spondylitis, in HCV the association with HLA
DQB1*03:01 has been linked to T cell function, though this was also identified prior
to the publication of the first draft of the map of the human genome. However, once
again, subsequent genetic studies have identified several other risk immune alleles for
HCV, some of which may interact with DQB1*03:01 and others that may influence viral
clearance independently. These include IFNL3 (previously IL28) and IFNL4. In rheuma-
toid arthritis, multiple risk alleles mostly associated with immunity have been identified,
including some cytokines and some chemokine receptors (e.g. IL6 and CCR6). In bipolar
disorder, the two major genes for focus are ANK3 and CACNA1C. It is easy to see the
potential for polymorphisms in the genes involved in calcium channel activity (such as
CACNA1C) that is essential for neurotransmitter release being important in bipolar disor-
der. In cardiovascular disease, it is easy to see the potential for genes in lipid metabolism,
cell growth, fibrosis, and regeneration as potential risk genes.
However, this will all mean nothing unless we can link genotype to phenotype—the next
piece in the jigsaw. It is OK to be selective in terms of deciding which genes to look at as
long as all of the potential candidates are considered at some point. Otherwise, the geno-
typing will have been a waste of time and the phenotyping work will fall into the same
trap that applied to early genetic association studies, i.e. rejection of positive associations,
but not based on numbers or study design (as in the past), but this time based on absence
of knowledge or absence of an obvious (simple) link.
Together, these examples illustrate the potential of this work. There is clear progress and
there is a great deal to be gained from this work. However, science is a slow process and
not all studies are easily replicated. This is most strongly illustrated in studies of bipolar
disorder. Overall, we must be sensible and proceed with caution because there is still a
great deal of work to be done.

2967_Ch04.indd 137 07/07/2015 10:35

 138 CHAPTER 4 Why Investigate Complex Disease Genetics?

Further Reading
Books Collins FS & McKusick VA (2001) Implications
Lechler R & Warrens A (eds) (2000) HLA in of the human genome project for medical sci-
Health and Disease, 2nd ed. Elsevier. ence. JAMA 285:540–544. This is a very impor-
tant review that provides the reader with a guide
Strachan T & Read AP (2011) Human to and an illustration of the potential for under-
Molecular Genetics, 4th ed. Garland Science. standing complex disease in the immediate
Chapter 16 deals with identifying human genes post-genome period.
and susceptibility factors, and provides excel-
lent additional data for students wishing to Ferreira MA, O’Donovan MC, Meng YA
delve deeper into medical genetics. There is also et al. (2008) Collaborative genome-wide asso-
additional relevant information in the other ciation analysis supports a role for ANK3 and
chapters. CACNA1C in bipolar disorder. Nat Genet
40:1056–1058.
Tiwari JL & Terasaki PI (1985) HLA and
Disease Associations. Springer. Chapter 8 on Helgadottir A, Manolescu A, Thorleisfsson G
Neurology covers early studies of multiple et  al. (2004) The gene encoding 5-lipoxygen-
sclerosis and narcolepsy plus other neuro- ase activating protein confers risk of myocardial
logical conditions. One of the authors of the infarction and stroke. Nat Genet 36:233–239.
1985 book, Paul Terasaki, is considered as one Helgadottir A, Manolescu A, Helgason A et al.
of the founding fathers of HLA typing and (2006) A variant of the gene encoding leukotri-
immunogenetics. ene A4 hydrolase confers ethnicity-specific risk
of myocardial infarction. Nat Genet 38:68–74.
Hirschhorn JN, Lohmueller K, Byrne E &
Articles Hirschhorn K (2002) A comprehensive review
Barrett JC (2012) From HLA association to of genetic association studies. Genet Med 4:45–
function. Nat Genet 44:235–236. Review of the 61. This paper provides an interesting pre-
paper by Raychaudhuri et al. (2012). GWAS view of association studies.
Baum AE, Akula N, Cabanero M et al. (2008) Lees CW, Barrett JC, Parkes M & Satsangi J
A genome-wide association study implicates (2011) New IBD genetics: common pathways
diacylglycerol kinase eta (DGKH) and several with other diseases. Gut 60:1739–1753. This
other genes in the etiology of bipolar disorder. paper provides a current summary of the genet-
Mol Psychiatry 13:197–207. ics of Crohn’s disease and also highlights the
possibility of shared pathways.
Brewerton DA, Hart FD, Nicholls A et  al. Lu X, Wang L, Chen S et al. (2012) Genome-
(1973) Ankylosing spondylitis and HL-A 27. wide association study in Han Chinese identifies
Lancet 301:904–907. This is the original iden- four new susceptibility loci for coronary artery
tification of association between ankylosing disease. Nat Genet 44:890–894. This study not
spondylitis and HLA-B*27; see also Schlosstein only identifies association with cardiovascular
et al. (1973). disease specific to the Han Chinese population,
Brown MA, Pile KD, Kennedy LG et al. (1996) but also confirms some shared associations with
HLA class I associations of ankylosing spon- the European population. Data from this study
dylitis in the white population in the United are summarized in Table 4.10.
Kingdom. Ann Rheum Dis 55:268–270. This Peden JF & Farrall M (2011) Thirty-five com-
paper is the gold standard paper for ankylosing mon variants for coronary artery disease: the
spondylitis and HLA-B*27. fruits of much collaborative labour. Hum Mol
Cichon S, Muhleisen TW, Degenhardt FA Genet 20:R198–R205. This review paper sum-
et  al. (2011) Genome-wide association study marizes the findings of two major groups (the
identifies genetic variation in neurocan as a CARDIoGRAM Consortium and the C4D
susceptibility factor for bipolar disorder. Am J Consortium) and identifies a total of 35 asso-
Hum Genet 88:372–381. This article is a very ciated common variants. The data from this
good overview of the genetics of bipolar disor- paper are summarized in Tables 4.8 and 4.9.
der with original data and some good examples The paper provides a useful discussion of the
on how to plan GWAS. differences between the two studies.

2967_Ch04.indd 138 07/07/2015 10:35

 Further Reading 139

Raychaudhuri S, Sandor C, Stahl EA et  al. United Kingdom: new estimates for a new cen-
(2012) Five amino acids in three HLA pro- tury. Rheumatology 41:793–800.
teins explain most of the association between The 1000 Genomes Project Consortium (2010)
MHC and seropositive rheumatoid arthritis. A map of human genome variation from popula-
Nat Genet 44:291–296. This original paper is tion-scale sequencing. Nature 467:1061–1073.
reviewed by Barrett (2012)—reading both is
useful to give a full understanding of the paper The Human Genome (2001) Nature 409:813–
and the meaning of the results. 958. The complete issue of Nature mostly
Schlosstein L, Terasaki PI, Bluestone R & dedicated to the HGM with multiple papers,
Pearson CM (1973) High association of an letters, and editorial commentary from multi-
HL-A antigen W27 with ankylosing spondyli- ple authors. It is full of useful insight and criti-
tis. N Engl J Med 288:704–706. One of two cal discussion. Any student of human genetics
original reports (see also Brewerton et al. 1973) should read this issue.
of the genetic association of HLA-B*27 with The International HapMap Consortium (2005)
ankylosing spondylitis. This example illustrates A haplotype map of the human genome. Nature
how not all findings from pre-genome and pre- 437:1299–1320.
GWAS studies failed to be replicated. Notice
in both cases the HLA nomenclature differs The International HapMap Consortium (2010)
from present. Standard nomenclature was only Integrating common and rare genetic variation in
just coming into use at this time and in retro- diverse human populations. Nature 467:52–58.
spect these differences can cause some confu- The Wellcome Trust Case Control Consortium
sion when looking at early texts. However, this (2007). Genome-wide association study of
allele family is now labelled HLA-B*27. It is 14,000 cases of seven common diseases and
important to note the quality of phenotype for 3,000 shared controls. Nature 447:661–678.
B antigens in 1973 was reasonable for common This is another landmark paper highlighting
antigens such as B27, but poor for rare antigens the development and application of new tech-
and variants. nologies (GWAS) in complex disease research,
Schunkert H, Konig IR, Kathiresan S et  al. and it provides a mass of useful information for
(2011) Large scale association analysis identi- those studying complex disease.
fies 13 new susceptibility loci for coronary heart Vassos E, Steinberg S, Cichon S et  al. (2012)
disease. Nat Genet 43:333–340. This paper Replication study and meta-analysis in
describes a large-scale study on cardiovascular European samples supports association of
disease based on a meta-analysis of 14 GWAS. the 3p21.1 locus with bipolar disorder. Biol
The data are presented in Table 4.8. Psychiatry 72:645–650. This paper describes a
Sklar P, Smoller JW, Fan J et al. (2008) Whole recent GWAS and meta-analysis study of bipo-
genome association study of bipolar disorder. lar disorder that illustrates some of the prob-
Mol Psychiatry 13:558–569. lems and uses of this type of analysis.
Stahl EA, Raychaudhuri S, Remmers EF et al. Zhang J, Zahir N, Jiang Q et  al. (2011) The
(2010) Genome-wide association study meta- autoimmune disease-associated PTPN22 vari-
analysis identifies seven new rheumatoid arthri- ant promotes calpain-mediated Lyp/Pep deg-
tis risk loci. Nat Genet 42:508–515. This is radation associated with lymphocyte and
an excellent review of genetics of rheumatoid dendritic cell hyperresponsiveness. Nat Genet
arthritis and provides a good starting point for 43:902–907.
any student looking at this disease.
Stranger BE, Stahl EA & Raj T (2011) Progress
and promise of genome-wide association stud-
ies for human complex trait genetics. Genetics Online sources
187:367–383. This is an excellent review of http://www.wtccc.org.uk/ccc2
the planning and how to do GWAS with data
on rheumatoid arthritis to illustrate the ideas. http://www.ncbi.nlm.nih.gov/SNP
There is fresh data in this paper to add to the This an essential site for updates on SNPs.
paper by Stahl et al. (2010). http://www.ncbi.nlm.nih.gov/omim
Symmons D, Turner G, Webb R et al. (2002) This is an essential site for updates and informa-
The prevalence of rheumatoid arthritis in the tion on genetic disease of all types.

2967_Ch04.indd 139 07/07/2015 10:35

 140 CHAPTER 4 Why Investigate Complex Disease Genetics?

http://www.ncbi.nlm.nih.gov/projects/SNP/ used with existing commercial packages such as

snp_summary.cgi SPSS or SAS.
This site provides information on reference http://www.genecards.org
SNP clusters in the genome.
This site is sponsored by the Weizmann Institute
http://hapmap.ncbi.nlm.nih.gov of Science and is free for use by academic and
This is the site of the results of the International non-profit institutions. Good for gene identifi-
HapMap project. cation, location, and function.
http://pngu.mgh.harvard.edu/~purcell/plink http://www.genome.gov/gwastudies
PLINK is a freely available, open source, whole A catalog of published GWAS. This website pro-
genome association analysis toolset designed vides a catalog of listed associations for GWAS
for quality control and analysis of GWAS data.
with SNP numbers greater than 100,000 and
http://www.r-project.org associations with a significance threshold cut of
This is another freely available and excellent 1.0 × 10−5.
package for statistical computing that can be

2967_Ch04.indd 140 07/07/2015 10:35

 CHAPTER
5
Statistical Analysis in Complex
Disease: Study Planning and
Data Handling

Now that we have considered why we are interested in the genetics of complex disease and
how to design studies, it is important to consider how to handle the data from these stud-
ies. This is done through the application of complex statistical analyses. No longer is statis-
tics a minor consideration in genetics; statistical genetics has become a science in itself and
statistical geneticists have become major players in complex disease. The reasons behind
this are clearly illustrated in Chapter 3. In the past, without good statistical advice, studies
were often badly designed and this led to the production of numerous false-positive and
false-negative genetic associations. However, this is not a statistics textbook, but given the
importance of this subject it is essential to consider the basic principles that are currently
applied in complex disease. Therefore, this chapter will consider some of the different
methods available to test statistical significance in both linkage and association analysis. It
will start with some basic methods applied to early studies, but will concentrate mostly on
statistics in genome-wide association analysis.
As stated in Chapter 3, there are two types of study in complex disease: linkage and asso-
ciation. These have been discussed at some length in previous chapters and they will only
be considered here from a statistical viewpoint.

2967_Ch05.indd 141 07/07/2015 10:38

 142 CHAPTER 5 Statistical Analysis in Complex Disease

5.1 Linkage Analysis

Linkage analysis is based on the detection of a physical link between a genetic marker
and a disease or disease subgroup. Linkage analysis can only be performed in families
and sibling pairs. Linkage analysis is most often performed in Mendelian diseases where
familial occurrence is common. Linkage analysis is less useful in complex disease. To illus-
trate linkage we can consider the example of nail–patella syndrome shown in Figure 5.1.
Nail–patella syndrome is an autosomal dominant disorder characterized by small poorly
developed nails and primitive knee caps. There is close linkage between this syndrome and
the genes encoding the ABO blood group on chromosome 9. In this illustration the ABO
alleles are IA for blood group A, IB for blood group B, and i for blood group O. The A and
B phenotypes are co-dominant, and the alleles IA and IB are dominant over i.
In Figure 5.1, the ABO genotypes can be inferred from the phenotypes of the offspring
produced. Female I-2 (i.e. generation I, person 2), for example, has blood group B,
which has two possible genotypes: IBIB or IBi. Looking at her offspring, one is blood
group O and must therefore have inherited an i allele from each parent. Consequently,
the genotype of female I-2 must be IBi. Similarly, because O requires i from both parents,
then the genotype of male I-1 (who displays the A blood group) must be IAi. Otherwise
his offspring would have blood group A because A is dominant over O. Nail–patella

A B
1 2 A: I A I A or I A i
B: I B I B or I B i
IA n IB n
O: ii
i N i n

B A O B A A
1 2 3 4 5 6
IB n IA n i n IB n IA n IA n

i N i n i N i N i n i n

A A B B O
1 2 3 4 5
IA n IA n IB n IB n i n

i N i n i n i n i n

Figure 5.1: Linkage between ABO blood group and nail–patella syndrome. The figure illustrates three
generations of a family, and shows the blood group and phenotype for nail–patella syndrome (a Mendelian
autosomal dominant disorder). The ABO blood type is indicated in each circle or square (IAIA, IBIB, and ii). The
genotype inferred from the phenotype is given below each circle or square. There are two possibilities for the
nail–patella genes: n for wild-type and N for mutant. Circles indicate female family members and squares indicate
male family members. In this figure the grandfather, generation I, person 1 (I-1), is blood group A heterozygous IAi
and is heterozygous for the nail–patella trait (nN), and will display the trait (indicated here by black symbols). The
grandmother, generation 1 person 2 (I-2), is blood group B heterozygous and carries two copies of the wild-type
nail–patella gene, and is therefore unaffected (indicated in white). The grandfather’s haplotypes are IAn and iN.
Those who inherit the paternal i allele in this family also inherit the N nail–patella gene mutation (all shown as black
squares or circles). Not all O blood group individuals have nail–patella syndrome (it is quite a rare trait); however, this
example indicates extreme linkage disequilibrium and suggests that the two genes are not independently assorted.

2967_Ch05.indd 142 07/07/2015 10:38

 Linkage Analysis 143

syndrome is illustrated in Figure 5.1 by either n (wild-type) or N (mutant). The N allele

is the autosomal dominant disease-causing allele. In this model we assume male I-1 is
heterozygous for the nail–patella trait (nN) and therefore affected. In contrast, female
I-2 is homozygous wild-type (nn) and therefore unaffected by this autosomal dominant
disease. The genotype of the generation I is:

I A i Nn × I Bi nn
In the second generation we find that only those offspring who inherit the paternal i allele
have the nail–patella trait and offspring with blood group A do not. This suggests that the
genes for the nail–patella syndrome are not independently assorted. If these two genes
were independently assorted, then some of the children with the paternal A blood group
could inherit the nail–patella syndrome from their father. This outcome is really impor-
tant as we need this information to determine whether the nail–patella gene in the father
is on the same chromosome as the marker allele (IA) or not. This is called determining the
phase of these two loci. We can establish the phase of the above loci in the parents as:

I A n I Bn
×
iN in
Further examination of the pedigree in Figure 5.1 shows that while there is no recombi-
nation in the offspring of generation I, there are two instances of recombination in the
offspring of generation II. If we focus on the couple II-1 and II-2 with the following
genotypes:

I Bn I A n
×
iN in
we can see that one of the children (male III-2) is unaffected and has blood group A with
the following genotype:

I An
in
This child must have inherited the i and n alleles from his father and not the i and N
combination that were expected, and because these alleles are on different chromosomes
in the father a crossover must have taken place. Similarly, crossover must have occurred
in child III-3.
Individuals III-2 and III-3 are examples of recombination between the two loci for blood
type (in this case the marker locus) and disease, and from this information we can cal-
culate the recombination fraction (θ) for this family. The distance between two loci is
defined in terms in units of centimorgans (named after geneticist Thomas Hunt Morgan
and denoted cM). If two loci are 1 cM apart, then there is a 1% chance of recombination
between the two loci as the chromosome is passed from one generation to the next.
It is essential to have informative families in linkage analysis, preferably with a large number
of offspring and several generations with both affected and unaffected members in each.
When small numbers are included there is a greater risk of false-positive and false-negative
results. Recombination is more frequent in cases where the trait and marker locus are not
close together, and the evidence for linkage can be quite low. Linkage analysis needs to

2967_Ch05.indd 143 07/07/2015 10:38

 144 CHAPTER 5 Statistical Analysis in Complex Disease

take into account these possibilities. The LOD score (logarithm of odds) is a measure of
the significance of linkage.

The LOD score is a measure of significance of linkage between

a trait and a marker allele
The LOD score compares the likelihood of obtaining the results if the two loci are linked
versus the likelihood of observing the same data purely by chance. The likelihood of link-
age depends on the frequency of recombination and recombination depends on the genetic
distance between the genes. In 1956, Morton published a model for the calculation of the
LOD score symbolized at the time as the Z score. The calculation is based on standard
pedigrees and is calculated for a range of recombination fraction values (θ). In this model,
the highest LOD score identifies the recombination fraction for the marker and the trait,
and is presented as a logarithm in base 10. A positive LOD score indicates the presence of
linkage, whereas negative LOD scores indicate that linkage is less likely. By convention,
a LOD score of 3 or greater is considered as strong evidence of linkage between two loci.
A LOD score of −2 or less is taken as evidence of no linkage. A LOD score between −2
and 3 is a gray area, with some statisticians arguing the upper limit should be lowered to
2 from 3 and others suggesting that even 3 is too low. Currently, it is reasonable to say a
value in-between the two above values is inconclusive. It is important to keep in mind that
when using a logarithm to base 10, a value of 3 indicates 1000:1 odds for linkage, suggest-
ing that the likelihood (L) that the observed linkage occurs by chance is less than 0.001.
An example of how LOD score is calculated is shown in Box 5.1. Finally, linkage analysis
can be either two-point mapping or multipoint (locus) mapping. Multipoint mapping is
considered more useful and is most frequently used in the present era.

box 5.1 THE LOD SCORE METHOD FOR LINKAGE

The LOD score indicates the logarithm of the odds to base 10 and it is calculated using
the recombination fraction (distance) between two loci on the same chromosome, which
is denoted by the Greek letter θ. θ is the likelihood (L) of a recombination event between
two loci and it is a function of the distance between the two loci. Two unlinked loci have
θ = 0.5; the closer they are, the lower the recombination fraction becomes. We use the
likelihood ratio or OR (the ratio between two probabilities) to estimate the probability
that two loci are linked such as:

L(θ) likelihood of an offspring’s sequence with recombinaation frequency θ

=
L(θ = 0.5) likelihood of an offspring’s sequence if loci are unlinked

If loci were unlinked, the most likely recombination frequency would be 1/2 or 0.5, and
in this case the numerator and the denominator of the ratio would be the same, and thus
the OR will be 1. The easiest way to calculate the above likelihood ratio is to employ the
logarithms (logarithm of odds to the base 10 LOD score) such as:

L(θ = θ )
LOD score = Z (θ) = log10
L(θ = 0.5)

2967_Ch05.indd 144 07/07/2015 10:38

 The Basic Statistical Concepts of Association Analysis 145

BOX 5.1 THE LOD SCORE METHOD FOR LINKAGE (Continued)

where θ  is the calculated recombination rate expressed as the distance between two loci.
A likelihood ratio of 1:1 equals to a LOD score of 0, a ratio of 10:1 equals of LOD score
of 1, a ratio of 100:1 equals a LOD score of 2, etc. A way of calculating LOD scores
using the above equation is to consider the number of births in a family (offspring) and
the number of recombinants as:

(1 − θ)n − r θr
LOD score = Z (θ) = log10
(0.5)n

where n is the total number of births and r is the number of recombinant types. If we
assume a test family with 18 non-recombinant offspring and two recombinant offspring
between two loci A and B, we need first to know the distance between the two loci.
Distance is measured in centimorgans (cM) and in this example is 10 cM, describing a
recombination frequency of 1% and estimates the value of θ. We will have:
n = 18 + 2 = 20
r = 2
θ = (10/100) = 0.1

(1 − 0.1)20 − 2 0.12
Z (0.1) = log10 = 3.2
(0.5)20

The resulting LOD score of 3.2 indicates that the two tested loci A and B show a strong
linkage.
By convention a Z (θ) greater than or equal to 3 (likelihood ratio = 1000:1, in favour of
linkage) is considered to be proof of linkage between two loci. A Z (θ) of −2 (likelihood
ratio = 0.01:1) or less instead, is taken as proof of not linkage. A Z (θ) between −2.0 and
3.0 is considered inconclusive and warrants for further studies. Some authors, however,
consider a Z (θ) of 2 (likelihood ratio = 100:1, in favour of linkage) as strong evidence
of linkage between two loci.

5.2 The Basic Statistical Concepts of

Association Analysis And Their
Application In Study Design
Association studies are not the same as linkage analyses. Different study plans have been
discussed at some length in previous chapters, but it is important to remember this.
Association studies seek to find association between alleles at one or more loci and a dis-
ease, but they do not establish a physical link. LOD scores do not apply in this type of
study. Association studies are mostly performed as case control studies, but the transmis-
sion disequilibrium test (TDT) and FBAT (family-based association test) are also associa-
tion-based methods (though they use families).

2967_Ch05.indd 145 16/07/2015 10:53

 146 CHAPTER 5 Statistical Analysis in Complex Disease

In statistical terms, there are two different hypotheses to consider

in the analysis of genetic association studies: a null hypothesis
and an alternative hypothesis
Differences between two groups in case control analyses are assayed using statistical tests.
Two possibilities are considered: the null hypothesis H0 and the alternative hypothesis
H1. These two possibilities are tested to make decisions about differences between the
two groups. The null hypothesis H0 assumes that there is no difference between the two
groups and that all samples are drawn at random from the same population. If we accept
the null hypothesis, then any differences between the two groups are considered to be due
to chance. The alternative hypothesis assumes the opposite and is incompatible with the
null hypothesis.
When statistical testing is applied to a study, if the probability that the observed differ-
ences are due to chance variation is sufficiently small (usually less than 1:20, P < 0.05),
then the null hypothesis is rejected and we conclude that there is a genuine difference
between the groups. In other words the sample populations are genetically different. In
this case we reject the null hypothesis and thus accept the alternative one. If we use regres-
sion analysis (as demonstrated below) we can conclude that the independent variable x has
some effect on the dependent variable ( y).

5.3 Statistical Error, Power, and P Values

Making the right decision and avoiding errors in the hypothesis testing
Before we address the concept of probability values (P) we must first consider differ-
ent types of statistical error and the meaning of statistical power. Though we think we
will achieve the right conclusion after we run an experiment, we may in fact make the
wrong conclusion. Whenever we decide—either to reject or accept a null hypothesis—
we could make mistakes, which are described as type I and type II statistical errors
(Figure 5.2).
We may either accept or reject the null hypothesis. If we reject the null hypothesis, we are
saying there is a difference between the two groups and that the null hypothesis is false.
If we accept (or fail to reject) the null hypothesis, we are saying there is no difference
between the groups and that the null hypothesis is true. However, in both instances we
can be wrong, making statistical errors as described below.
A type I statistical error is the incorrect rejection of the null hypothesis when it is in fact
true. This generates false-positive associations. The probability of making a type I error is

Null hypothesis (H0) is true Null hypothesis (H0) is false

Reject null hypothesis (H0) Type I error α Correct outcome

False positive True positive

Accept null hypothesis (H0) Correct outcome Type II error β

True negative False negative

Figure 5.2: Type I and type II statistical errors. The figure shows a number of different possibilities arising from
statistical errors. The null hypothesis (H0) states that there is no statistical difference between two groups.

2967_Ch05.indd 146 07/07/2015 10:38

 Statistical Error, Power, and P Values 147

denoted α and is the same α as the significance level of a test; we reject the null hypothesis
if the inferred P value is less than the significance level (or threshold) α. We need to estab-
lish the value of α before we run the analysis. A conventional P value of 0.05 is commonly
assigned, though we may choose a more restrictive value such as 0.01, 0.001, or even
5 × 10−7 depending on the nature of the study and the number of tests being run. For a
given null hypothesis H0, the type I error rate P is shown as:

P(reject H0 H0 is true ) ≤ α

where P is the probability of rejecting the null hypothesis when it should instead be
accepted, i.e. a false positive.
Type II statistical errors arise when we accept the null hypothesis when it is in fact false,
giving rise to false negative associations. A type II error is denoted by β and this corresponds
to the probability of not rejecting the null hypothesis when it is false; 1 – β represents the
power of the test. For a given null hypothesis H0, the type II error rate is shown as:

P(not reject H0 H0 is false ) ≤ β

The way to avoid these problems is outlined further in this chapter, and includes good
study planning, and large-scale studies with clear statistical goals and significance thresh-
olds. In many clinical situations the type I error is potentially serious and a procedure able
to minimize the probability of a type I error is essential. The significance level of α is cru-
cial in reducing the likelihood of a type I error. If you set α at 0.05, it means that you want
the probability of a type I error to be no more than 0.05 or no more than approximately
1:20 cases to be a false positive.

The likelihood of detecting a significant difference in an association

study is directly related to sample size
The power of a test (1 – β) is the probability of correctly rejecting the null hypothesis
when it is false (i.e. H1 is true) for a given significance threshold α. Statistical power is an
important concept in association studies, especially because it is possible to miss impor-
tant statistical differences if a study is under-powered. It is essential therefore that we have
information on the statistical power of a proposed investigation before we embark on a
study. A number of factors have a direct influence on statistical power, such as sample
size, the expected size of the impact in terms of risk in the disease population (or disease
subgroup), and the significance threshold (α), which is set prior to the study. The power
increases if we enlarge the sample size and also with increases in the size of risk expected
to be found. A study design aimed at decreasing the variability of the observations also
produces more power. The power is higher with less stringent significance levels. However,
this also results in an increased likelihood of a type I error. Thus, we are more likely to
detect an effect if we decide to use a threshold of α = 0.05 rather than α = 0.01, but the
effect detected is also more likely to be false.
One way to maximize the power is to increase the sample size, other methods are available
related to study design, and sample and genotype selection, but sample size is the only
method discussed in detail here. Differences in sample size may account for the differences
in reported genetic associations in complex disease, especially in early pre-genome-wide
association studies (GWAS) (see Chapter 3).

2967_Ch05.indd 147 07/07/2015 10:38

 148 CHAPTER 5 Statistical Analysis in Complex Disease

Probability (P) values are simply statements of the probability that the
observed differences between two groups could have arisen by chance
As the P value falls, the evidence against the null hypothesis in favor of the alternative
hypothesis becomes stronger. This also increases the likelihood of obtaining the same or
similar results in a second (validation) cohort. The lower the P value, the greater the con-
fidence one can have in the results obtained.

Large-scale studies are also performed to increase statistical confidence

In large-scale studies, especially GWAS, the threshold for significance (P value) varies
from the above norms depending on the study size. The Wellcome Trust Case Control
Consortium (WTCCC1) GWAS published in 2007 recommended a minimum value
of 5 × 10−7, whereas Dudbridge and Gusnanto (2008) suggested an even lower value
of 7.2 × 10−8 for the UK population. The reason for this higher level of stringency in
the acceptable threshold is that an average GWAS investigates at least 500,000 single
nucleotide polymorphisms (SNPs) and in such a study we would expect approximately
25,000 false-positive associations if the conventional significance threshold of P < 0.05
was applied. As 500,000 SNPs are tested, 500,000 statistical tests are being performed
(multiple testing) and therefore correction for multiple testing should be applied in order
to avoid large numbers of false positives. One option is to use Bonferroni’s correction,
which calculates a new significance cut-off according to the number of SNPs or alleles
tested.

Bonferroni’s correction
Bonferroni’s correction sets the new significance cut-off at α/n where α is the significance
level (usually α = 0.05) and n represents the number of SNPs assayed (or the number of
tests carried out). For example, in the above investigation of 500,000 SNPs, Bonferroni’s
correction sets the new threshold at 5 × 10−7.
Under Bonferroni’s correction many authors prefer to adjust each single P value instead
of the α significance level. Each P value is multiplied by the number of SNPs tested and
called significant if the corrected P value is still under 0.05.
Corrected P value = P value × n (number of SNPs in test) < 0.05
Bonferroni’s correction assumes there is independence between markers, and thus inde-
pendence between each SNP (or other genetic variant) and the trait. In the context of
genetic association studies, as many of the SNPs under investigation are located on the
same chromosome and linked through varying degrees of linkage disequilibrium, this
correction has been criticized as excessively conservative, which may lead, especially for
tightly linked SNPs, to loss of power and thus to an increase in the likelihood of a type
II error. For this reason many GWAS studies are performed in stages, with preliminary
studies applying less stringent P values in the first round aimed at identifying potential
loci of interest that can then be confirmed in subsequent rounds. This type of multistage
planning is discussed in Chapter 3.
Alternative methods have been proposed in order to avoid the high penalty applied by
using Bonferroni’s correction and to correct for multiple testing. A practical alternative,
for example, is to approximate the type I error rate using a permutation procedure. This
procedure aims to calculate the approximate false-positive rate that can be then be used as
the threshold for the P value in the data analysis. The method is conceptually simple, but
is computationally demanding and is beyond the scope of this book, and therefore will not
be discussed further. Other approaches have been proposed, including an estimation of

2967_Ch05.indd 148 07/07/2015 10:38

 Statistical Error, Power, and P Values 149

the false discovery rate and a Bayesian approach. The false discovery rate is the propor-
tion of positive tests that are false positive and often leads to low threshold levels, while the
Bayesian approach incorporates the prior probability of association.
Sample relatedness
It is assumed in all case control association studies that cases and controls are unrelated.
Alleles can be identical by state (IBS), i.e. they are the same allele having the same pheno-
type. Alternatively, they can be identical by descent (IBD), i.e. they are the same and share
a common ancestor, and thus have a shared haplotype from one of the parents (Figure 5.3).
In order to evaluate the degree of relatedness of a sample, pair-wise probability (of IBD)
between every individual in the study is calculated. As GWAS uses dense SNP arrays this
makes it is easy to compute pair-wise kinship estimates. In practice, there is no need to
perform analysis of the whole genome. A GWAS SNP data set from only 100,000 mark-
ers will yield stable estimates of kinship coefficients. On average, siblings share zero, one,
and two alleles IBD at 25, 50, and 25% for each gene, respectively. Unrelated individuals
do not share alleles IBD. The most commonly used approach to incorporate locus-specific

Family A

A1 A2 A1 A3

A1 A3 A1 A2

Family B

A1 A2 A2 A3

A1 A3 A1 A2

Figure 5.3: Identity by state (IBS) versus identity by descent (IBD). The figure illustrates two families: A and B.
Consider a single gene (A) with three possible alleles A1, A2, and A3. The father’s alleles are shown in gray boxes.
In family 1, the paternal genotype is A1/A2 and the maternal genotype is A1/A3. The children both inherit the A1
allele. However, this is inherited from the father in the male child and the mother in the female child. The male
child receives the maternal A3 allele and the daughter receives the paternal A2 allele. Therefore, in family 1 the
male child has the genotype A1/A3 and the female child has the genotype A1/A2. Though both children carry
the A1 allele, they are IBS for this allele and as a consequence may carry different alleles in linkage disequilibrium
with the A1 allele. In family 2, the pattern of inheritance is subtly different. The father has the same genotype as
family 1, but the mother carries the A2 and A3 alleles only. Both children inherit A1 from their father. Thus, with
respect to the A1 allele they are IBD. The two children do not, however, have the same genotype as they inherit
different maternal alleles, creating A1/A2 in the male child and A1/A3 in the female child.

2967_Ch05.indd 149 07/07/2015 10:38

 150 CHAPTER 5 Statistical Analysis in Complex Disease

IBD sharing probabilities into analysis is the pi-hat approach, which is implemented in
the PLINK whole-genome association analysis software. PLINK is a single-syllable term
for a free, open-source whole-genome association analysis toolset, designed to perform a
range of basic, large-scale analyses in a computationally efficient manner. PLINK can be
used for a range of tests and analyses. Using the – –genome option in PLINK, it is possible
to estimate pair-wise IBD to detect pairs of individuals who look more or less different
from each other than we would expect in a random sample. As a rule of thumb a pi-hat
value of 0.95 or greater resulting from pair-wise IBD comparisons is taken as individual
relatedness. Detailed instructions on producing this data can be found on the PLINK web
page (http://pngu.mgh.harvard.edu/~purcell/plink/).

Genotyping efficiency and sample call rate

Genotyping efficiency per sample, or sample call rate, is an informative indicator of sam-
ple quality. The call rate per sample represents the percentage of SNPs genotyped in each
individual. If a large proportion of SNPs assayed on an individual DNA sample fails to
be genotyped it may be indicative of a poor quality DNA sample, which could lead to
abnormal genotype calling. False positives will arise if DNA quality differs with phenotype,
as the frequency of called genotypes will differ. False negatives will also arise by reduc-
ing the statistical power as a result of reduced sample size. Samples with low genotyping
efficiency should be removed from further analysis. Acceptable call rate thresholds vary
from study to study depending on the genotyping platform used, but they are usually
set at 98–99%. Some manuscripts or books may refer to sample genotyping efficiency
as individual ­missingness, which, for each subject, is the fraction of missing SNPs (not
genotypes) and it ranges from 0 (all SNPs are genotyped) to 1 (all SNPs failed). Genotyping
efficiency can be checked using the -missing option in PLINK. This will produce a file
showing the genotype missingness rate (1 – efficiency) for each individual (proportion of
SNPs that failed on each sample) and for each SNP (proportion of individuals for which
no genotype was called). Samples below a desired threshold should be eliminated from any
further (downstream) analyses.

SNP quality and SNP call rate

One of the determinants of failure or individual missingness is SNP quality. SNP assays
that fail on a large number of samples are poor assays with a low call rate and are likely
to result in spurious results. As with individual missingness, the suggested threshold for
removing SNPs with a low call rate is approximately 98–99%, though this threshold may
vary from study to study. The higher the threshold, the lower the proportion of miscalled
data, but this increases the proportion of missing SNPs and leads to a loss of potentially
useful information. This can generate a sort of false-negative result whereby potentially
informative tagged SNPs are removed from the test set before the association analysis is run.
Some studies have found that call rates may be lower where there are higher proportions of
heterozygotes and this may introduce bias into a study. Differential rates of SNP missing-
ness between cases and controls may also be a problem. Quality control of SNP genotyping
should be performed before any quality control on sample genotyping because in this way
fewer samples will be dropped from the analysis as they were genotyped with tagged SNP
that had poor performance. In PLINK, markers can be removed based on call rate by using
the – –geno option, followed by a threshold for a lower limit of missingness (e.g. – –geno
0.02 would remove SNPs with more than 2% missing, i.e. less than a 98% call rate).
Issues such as sample relatedness, sample call rate, and SNP call rate can all be built into the
study design, but a certain amount of prior information is needed for each and therefore
these are often dealt with at the analysis stage, especially in the first round of GWAS.

2967_Ch05.indd 150 07/07/2015 10:38

 The Basic Statistical Considerations 151

Data storage format and software: GWAS and other studies

The most commonly used format to store and analyze association data (including GWAS
and genome-wide linkage analysis) is the pedigree file format. This file is matrix-like,
where each individual is in a row, and phenotype and genotype information are stored
in columns. The first six columns are identifying information (family ID, individual ID,
father ID, mother ID, sex, phenotype), and the remaining columns (column 7 onwards)
are genotypes (two columns per genotype; one for each allele, white-space delimited).
There are several variations of this format, including transposed (long) formats (typed)
and compressed (binary) formats. Descriptions of these file formats can be found on the
PLINK homepage (see above). PLINK is frequently used for handling GWAS data and
quality control analysis; it is specifically designed and optimized for GWAS and is compu-
tationally efficient. PLINK is only one package that allows quality control for genome-wide
associations and there are other possibilities, including those writing custom codes using
the freely available and excellent R package for statistical computing (www.r-project.org/).

5.4 The Basic Statistical Considerations for

Analysis Of Case Control Association
Studies And Their Application to Data
Collection And Analysis
Checking for Hardy–Weinberg equilibrium (HWE) is one step in the quality control
analysis of markers in any association analysis including GWAS data. HWE is discussed
at some length in Chapter 1; however, as a focal principle, it is important to remind our-
selves of the role and importance of HWE. Under HWE, alleles segregate randomly in
the population, allowing expected genotype frequencies to be calculated from observed
frequencies. Departure from this equilibrium can be indicative of potential genotyping
errors, population stratification, or even actual association with the trait under study.
A simple comparison of the expected and observed genotype frequencies provides a test of
HWE and departure from HWE is generally tested for by using Pearson’s χ2 test. This test
evaluates goodness of fit between the observed genotype counts and expected genotypes
under HWE, and is illustrated in Table 5.1.

Departures from HWE can have different causes

Departures from HWE may be caused by technical errors, e.g. in genotype assignment or
detection (most frequently this is due to missing heterozygotes), but there may also be bio-
logical explanations, such as non-random mating, population stratification, and selection.
In addition, some cases of departure from HWE may occur as a result of a genotyped SNP
having a strong influence on disease susceptibility. Genotypes for a disease mutation will
only be in HWE if the genetic model is multiplicative (see below). Occasionally, one allele
is strongly associated with disease in a near-Mendelian pattern. In these cases, HWE may
appear to have been broken down. Disease-free controls by contrast should follow HWE
more closely. No standard guidelines for rejecting SNPs that depart from HWE have been
developed. In practice, all SNPs for which the P values obtained by Pearson’s χ2 test are
below a predetermined threshold should be checked manually for genotyping quality. In
Haploview, a software program that is used extensively with HapMap data, the default
significance threshold for departure from HWE is set at 0.001.

2967_Ch05.indd 151 07/07/2015 10:38

 152 CHAPTER 5 Statistical Analysis in Complex Disease

Table 5.1 Testing for departure from HWE.

Genotype count Estimated frequency Estimated frequency

AA AC CC Total of A allele of C allele
General
Observed a b c N 2a + b 2c + b
p= q =
Expected pN 2
2pqN qN 2 2N 2N
Pearson’s χ 2
3
(Oi − Ei )2
χ2 = ∑
i =1
Ei

Data set
Observed 50 30 20 100 2 × 50 + 30 2 × 20 + 30
p= q =
Expected 42.2 45.5 12.2 200 200
= 0.65 = 0.35
Pearson’s χ 2
11.6
P value 0.0008
a, b, and c represent numbers of AA, AC, and CC genotypes, respectively. There is no “b” allele in this illustration;
this represents the AC heterozygotes in the general section of this table only.

Analyzing the data

In a simple single-gene locus study with a single polymorphism under consideration,
genotype data are usually illustrated in a 2 × 3 contingency table, where rows represent
the disease status and columns represent the genotype counts. Allele frequencies are pre-
sented in a 2 × 2 contingency table with a similar structure. Different statistical analysis
methods can be applied to these types of tables. In the section below we will consider
a number of different statistical methods, starting with Pearson’s χ2 test then moving
on to the Fisher’s exact probability test (which in theory is simpler, but requires more
computational capacity), and the more advanced Cochran–Armitage and logistic regres-
sion tests.

Pearson’s χ2 and Fisher’s exact test are used to assess the departure
from the null hypothesis
To consider Pearson’s χ2 test, let us use the data illustrated in the 2 × 3 table in Table 5.2,
which represents data from a single SNP for a standard case control association study. The
total number of genotyped individuals n is represented as ncases and ncontrols for cases and
controls, respectively. The table represents a single SNP with alleles A and C, where nAA
is the total number of AA genotypes observed, nAC is the total number of AC genotypes
observed, and nCC is the total number of CC genotypes observed.
Pearson’s χ2 test is used to assess departure from the null hypothesis, which states that cases
and controls have the same distribution of alleles and of genotypes. In this example we will
use this statistical test to determine the χ2 distribution with 2 degrees of freedom (d.f.)
from the 2 × 3 contingency table. In this instance, 2 d.f. is selected because there are two

2967_Ch05.indd 152 07/07/2015 10:38

 The Basic Statistical Considerations 153

Table 5.2 Simplified 2 × 3 for Pearson’s χ2 test.

AA AC CC Total
Cases a b c ncases
Controls d e f ncontrols
Total nAA nAC nCC n

possibilities A or C. The null hypothesis being tested is that there is no significant differ-
ence in genotype distribution between cases and controls. In order to understand this test,
consider Figure 5.4 and focus on the observed value for the AA genotype (OAA = a). When
we apply Pearson’s χ2 test, this observed value is compared with its expected value EAA cal-
culated from an equation involving the total number of cases (ncases), the total number of
AA genotypes (nAA), and the total number of individuals (n):

n AA × ncases
E1 =
n
The above equation is used to calculate each expected genotype i as:

ni × ncc
Ei =
n
where ni = ni cases + ni controls is the observed number of each genotype given by the sum of
i cases plus i controls, ncc is the total number of cases (ncases) or controls (ncontrols), and n is
the total number of individuals genotyped (ncases + ncontrols). Each observed genotype is then
compared with each expected value and the full statistical test is applied:
6
(Oi − Ei )2
χ2 = ∑
i =1
Ei

where the summation is over all six cells in the table and Oi are the observed values (a, b,
c, d, e, and f in Table 5.2) in each cell, while Ei is the expected value calculated according
to the equation above for each genotype in cases and controls.

Allele A Allele C Totals

Cases a b a+b

Controls c d c+d

Totals a+c b+d n=a + b + c + d

Figure 5.4: A simplified form of the χ2 test. This simplified version of the χ2 test is often applied in association
studies. χ2 is calculated from a 2 × 2 table above using values for alleles in boxes for cases and controls allele A
and allele C above (a, b, c, and d here) and the data from sum boxes (containing a + b, c + d, a + c, b + d, and n
here). The 2 × 2 table can also be used to calculate the OR using the values in the boxes containing a, b, c, and d.

2967_Ch05.indd 153 07/07/2015 10:38

 154 CHAPTER 5 Statistical Analysis in Complex Disease

Pearson’s χ2 test compares the observed and expected genotypes in cases and controls
assuming both cases and controls have the same frequency of genotypes. The test is asymp-
totic, implying that the analysis becomes more accurate with larger data sets. A low count
in any of the cells of the table can violate the above assumption. In practice, an expected
value of at least five observations in each cell is regarded as a minimum number needed in
order to apply Pearson’s χ2 test. Fisher’s exact test is recommended if any of the cells have
values below five.
When using χ2 it is important to use real numbers and not percentages or mean values as
some authors have done in error. It should also be noted that some calculations use a 2 × 2
table in a simplified version of Pearson’s χ2 test (Figure 5.4). The principles are the same in
this analysis; however, allele frequencies are counted rather than genotypes. In complex sys-
tems where genes have multiple alleles it is often easier to count individuals as either positive
or negative for an allele of interest rather than count genotypes. Homozygosity is ignored
in this model and no assumptions are made about homozygous or heterozygous advantage.
Data from 2 × 2 tables illustrated in Figure 5.4 can be used to calculate χ2 using the simple
formula:

n( ad − bc )2
χ2 =
( a + c )(b + d )( a + b )(c + d )

The value of this calculation is that it makes no assumptions about the impact of hetero-
zygotes versus homozygotes—the calculation is simply based on the number of alleles in
the population. Homozygotes are counted once only. This same formula can be used to
calculate the odds ratio (OR) as:

a×d
OR =
c ×b
These two calculations (simplified χ2 and OR) are among the most well-known and fre-
quently used in association studies, especially in the early phases of data analysis.

Fisher’s exact test calculates the exact probability (P) of observing the
distribution seen in the contingency table
Fisher’s exact test is computationally more demanding than Pearson’s χ2 test and requires
factorial calculations. The problem of working with factorials is that they generate very
large values quickly, consequently where it is necessary to use them most students will use
computer statistical packages such as R for this type of analysis. However, it is important
to understand how this test works. Therefore, for the sake of clarity, the formula to calcu-
late Fisher’s exact test from a standard 2 × 2 contingency table is calculated as:

r1 ! r2 ! c1 ! c 2 !
P =
n ! a !b ! c ! d !
where P is Fisher’s exact probability. The symbol “!” refers to the factorial for the cells and
the marginal values that are identified in the 2 × 2 Punnett Square in Figure 5.4. In this
table, a, b, c, and d are the observed values and r1, r2, c1, and c2 are the marginal values
(computed as the sum of each column or row depending on position). The letter n denotes
the total number.

2967_Ch05.indd 154 07/07/2015 10:38

 The Basic Statistical Considerations 155

An exhaustive explanation of this test is beyond the scope of this book and we advise the
reader to refer to a good statistics book.
Figure 5.4 shows a 2 × 2 contingency table with two alleles A and C in two groups: cases
and controls. Allele frequencies for the two groups are represented by the values a, b, c, and
d. There is no assumption about the status of homozygous genotypes in this calculation.
The values a + c and b + d are the sum values for the two columns (c1 and c2) and a + b
and c + d are the sum values for the two rows (r1 and r2). The value n is the total for the
whole cohort. The marginal values used in Fisher’s test are c1, c2, r1, and r2 corresponding
to a + c, b + d, a + b, and c + d above, respectively.

Allele A Allele C Totals

Cases a b a + b = r1
Controls c d c + d = r2
Totals a + c = c1 b + d = c2 n

The Cochran–Armitage test looks for a trend for a difference between

cases and controls across the ordered genotypes in the table
The Cochran–Armitage test was developed specifically to detect a linear trend in propor-
tions over levels of exposure for a risk variable; when applied to genetic association analy-
ses it tests whether the probability of disease increases in a linear pattern with the three
corresponding genotypes and the binary trait (the latter meaning cases or controls in this
instance). The Cochran–Armitage test begins by taking into account the ordering of the
columns of the contingency table (shown in Figure 5.5). In Figure 5.6, the column order
is AA, AC, and CC. Cochran–Armitage tests the null hypothesis of zero slope for a line of
best fit for the three genotypic risk estimates. We can consider the following example to
illustrate the Cochran–Armitage equation for genotypes.
The hypothetical counts of genotypes in a 3 × 3 contingency table for each marker of
hypothetical cases and controls are N1· = n11 + n12 + n13 and n·1 = n11 + n12, etc..

Genotype Totals
AA Aa aa
Cases n11 n12 n13 N1·
Controls n21 n22 n23 N2·
Totals n·1 n·2 n·3 N

The Cochran–Armitage statistic is calculated as:

2
 3 
∑
 i =1
wi (n1i n2 • − n2i n1• 

T2 =
n1•n2 •  
3 2 3

n 

∑ wi2 n•i (n − n•i ) − 2 ∑∑ wi w j n•i n• j 

 i =1 i =1 j = i +1 

2967_Ch05.indd 155 07/07/2015 10:38

 156 CHAPTER 5 Statistical Analysis in Complex Disease

(b) Dominant model: allele C (c) Recessive model: two copies of allele
increases the risk C are necessary to increase the risk

AA AC + CC AA + AC CC

Cases a b+c Cases a+b c

Controls d e+f Controls d+e f

(a) Full data no model

AA AC CC

Cases a b c

Controls d e f

(d) Multiplicative model: the disease (e) Additive model: the risk of developing the disease
risk increases r-fold. The heterozygous is r-fold for the AC genotypes, and 2r-fold for the CC
AC carriers risk r; the homozygous genotypes. The disease risk increases r-fold.
carriers risk 2r. Analyzed by allele. The heterozygous AC carriers risk r; the homozygous
carriers risk 2r. Analyzed by allele.
A C
AA AC CC
Cases 2a + b b + 2c
Cases a b c
Controls 2d + e e + 2f
Controls d e f

(f) Cochran-Armitage test is additive model when

weights are set w = 0, 1, 2

AA AC CC

Cases a b c

Controls d e f

Figure 5.5: Cochran–Armitage test for trend. The figure illustrates several different options for analysis of data
from single SNP association studies. Several different models are illustrated. Here, allele C is assumed to increase
the risk for the disease. The models applied to (a) are: dominant (b), recessive (c), multiplicative (d), and additive
(e). In the additive model, the risk of the trait is greater than two-fold in homozygotes, whereas in the multiplicative
model the risk is two-fold in homozygotes compared with heterozygotes. (Adapted from Lewis CM & Knight J
[2012] Cold Spring Harbor Protoc 2012:297–306. With permission from Cold Spring Harbor Laboratory Press.)

where wi = (w1, w2, w3) weights are set to detect particular types of association. In genetic
association studies and GWAS, the Cochran–Armitage test is mostly used to examine the
potential additive effect of the alleles. To do this the weights are set as w = (0, 1, 2), which
correspond to Figure 5.5, for the additive effect of the allele C in developing the risk of
the disease.
In addition, the Cochran–Armitage test can also be used to test for a potential domi-
nant (w = 0, 1, 1) or recessive effect (w = 0, 0, 1) effect of the C allele in Figure 5.5.

2967_Ch05.indd 156 07/07/2015 10:38

 The Basic Statistical Considerations 157

0.64

Case/case + controls
0.62

0.60

0.58

0.56

0 1 2
Genotype score

Figure 5.6: Cochran–Armitage test for trend in a single SNP case control association study. In this case
we are looking at a single gene with two alleles (alleles 1 and 2). The genotypes are scored as 0, 1, and 2 for
homozygotes allele 1, heterozygotes alleles 1 and 2, and homozygotes allele 2, respectively. The circles indicate
the genotype score for the cases and controls combined together with their least-squares line. Here, the line fits
the data reasonably well as the heterozygote risk estimate is intermediate between the two homozygote risk
estimates, corresponding to an additive genotype risk. (Adapted from Balding DJ [2006] Nat Rev Genet 7:781–791.
With permission from Macmillan Publishers Ltd.)

T 2 has a χ2 distribution with 1 d.f. under the null hypothesis of no association. The
Cochran–Armitage test for trend performs with robust power in the case of an additive
risk of disease developing. Conversely, the power is reduced by deviations from an addi-
tive model. Cochran–Armitage is a more conservative test, minimizing the probability of
incorrectly rejecting the null hypothesis and thus reducing the risk of a type I statistical
error. Cochran–Armitage is also more robust when it comes to departures from HWE
than either Pearson’s χ2 or Fisher’s exact probability tests.

There is no simple answer to the question of which test to choose

There is no easy answer about which statistical test to use for case control association stud-
ies. Adopting the Cochran–Armitage test can mean sacrificing statistical power if the risks
for each genotype do not conform to an additive model. Using Pearson’s χ2 or Fisher’s test
spreads the analyses over the full range of risk models, and this inevitably means investing
less effort in the detection of additive risks. Pearson’s and Fisher’s tests have reasonable
power regardless of the underlying risks, but when the genotype risks become additive,
these tests lose their power. It is also worth remembering that Sasieni (1997) has shown
that when HWE is violated, an allele-based test is inappropriate and the frequency of gen-
otypes rather than alleles should be compared using the Cochran–Armitage test for trend.

Data may also be analyzed assuming a predefined genetic model

Genetic contributions to disease risk for complex traits are complex and do not fit easily into
such simple models. However, it is possible to identify patterns looking at individual loci.
Dominant, recessive, and additive (or over-dominant) models
In Figure 5.5, if we assume that allele C is a high-risk allele for the disease being tested,
then we can test to see if the effect is dominant (Figure 5.5b) or recessive (Figure 5.5c).
The model assumes that carrying allele C increases disease risk, and to test it AC and CC
genotypes are pooled together to give a 2 × 2 table. This is particularly relevant when allele

2967_Ch05.indd 157 07/07/2015 10:38

 158 CHAPTER 5 Statistical Analysis in Complex Disease

C is rare, with few CC observations in cases and controls. Alternatively, under a recessive
model for allele C, genotypes AA and AC are pooled together (Figure 5.5c). This model
assumes that two copies of allele C are required for increased risk. However, in some cases
risk is additive, meaning that the risk in homozygotes is higher than that for heterozy-
gotes. In an additive model if we assume allele C is the risk allele, the risk of developing
the disease is r for the AC genotypes and 2r for the CC genotypes (Figure 5.5e).
Though the described models and tests are valid methods for analysis of association stud-
ies, these methods should be applied according to a predefined study plan because a ran-
dom application of such tests increases the probability of a false-positive result. These
models are rarely used in the primary analysis of complex genetic disorders, but they may
be employed as a secondary tool to explore the potential mode of inheritance of an associ-
ated SNP or to test a predefined hypothesis.
Multiplicative risk
Under this model, if C is the risk allele, the risk of developing a disease is modeled by
increasing risk factor r from r for heterozygotes (AC) to r2 for homozygotes (CC). Under
the multiplicative model the allele count is then the most powerful method of testing.
The total number of A and C alleles in cases and controls are compared, regardless of the
genotypes from which these alleles are constructed (Figure 5.5d).
Finally, we must remember that in complex diseases we anticipate the involvement of sev-
eral genetic risk factors that may interact in complex models. We must be cautious about
simple patterns. In fact, most single-gene studies count alleles and make no assumptions
about homozygotes or models. This is especially true for human leukocyte antigen (HLA)
studies where the number of alleles at each locus is extreme.
Logistic regression
Regression is a statistical method employed to determine the dependence between a
response variable (y), known as the dependent variable, and one or several predictors (x),
called the independent variables. In the simplest case, y (the dependent variable) is predicted
by only one x (independent variable) by a linear equation or the equation of a straight line:
y = α + βx + ε
where α is the intercept, β is the regression coefficient (i.e. the slope of the line), and ε
represents the error (i.e. the differences between predicted and observed y values). The
above equation illustrates a simple linear regression model where a line fits all values of x
with y. Where there are infinite lines we need to find the linear equation able to best fit all
observed values of x. To achieve this aim we use the least-squares criterion, which states
that the line of best fit for the data is the line where the sum of the squares of the verti-
cal distances from the observed points to the line are as small as possible. This criterion
is illustrated in Figure 5.7 with an example taken from population ecology regarding the
distribution of a predator population (e.g. wolves) and their prey (e.g. hares), where each
dot represents an observed value and x is the number of prey observed within an area and
y the corresponding number of predators. In this equation, d represents the difference
between the y coordinate of the data point and the corresponding y coordinate on the line.
The best line of best fit (Smin) is the line which minimizes the sum of the d squares (d 2):

Smin = ∑d
i
i
2

2967_Ch05.indd 158 07/07/2015 10:38

 The Basic Statistical Considerations 159

y
250
d

200
d d
Predators

d
150
d d

100
d
x
1000 2000 3000 4000
Prey

Figure 5.7: Linear regression using a population ecology model. An example from population ecology
that represents the relationship between prey and predator numbers within a studied area, where each dot
represents an observed value; x is the number of prey observed within an area and y is the number of predators,
d represents the difference between the y coordinate of the data point and the corresponding y coordinate
on the line. In this case, an increase in the prey population corresponds with an increase in the number of
predators.

The distance of each point in the scatter from the regression line is known as the residual
or error, denoted d. When all of these residuals are squared and then added together they
are given the term ∑ d 2 . The “best” straight line is the one with the lowest value for ∑ d 2
hence the name ordinary “least squares.” Another example is shown within Figure 5.8,
where a scatter plot is plotted to illustrate the relationship between body mass index versus
hip circumference for a sample of 1500 women in a diet and health cohort study.

y
40

30

BMI
20

10

x
100 110 120 130 140 150
Hip circumference (cm)

Figure 5.8: A scatter plot of body mass index (BMI) versus hip circumference. The figure is a scatter plot
based on data from a sample of 1500 women in a diet and health cohort study. The scatter illustrates the
relationship between BMI and hip circumference. The scatter of values appears to be distributed around a
straight line, i.e. the relationship between these two variables appears to be almost linear.

2967_Ch05.indd 159 07/07/2015 10:38

 160 CHAPTER 5 Statistical Analysis in Complex Disease

A simple linear regression model is one with only one independent variable. When we
have more than one independent variable the regression model is called a multiple linear
regression model. This model is an extension of simple regression where the dependent
variable y is predicted by several independent variables (xi) simultaneously:
y = α + β1x1 + β2x2 + … + ε
This multiple linear regression allows us to include more predictors or covariates in the
analysis. This will reduce the residual variance (d ) of y.
In genetic association studies, however, a logistic regression model is preferred compared
with a linear regression model.

Logistic regression is frequently used in association studies

There are two regression models: linear regression, which may be either simple or mul-
tiple, and logistic regression. A logistic regression model is preferred in case control genetic
association studies where we have a binary outcome of interest (e.g. presence/absence of
a disease) and a number of explanatory variables. In studies of quantitative traits logistic
regression would not be appropriate. In order to have a valid analysis we need to satisfy the
assumptions of logistic regression or we may derive invalid statistical inferences. Before we
can use our model and make any statistical inference, we need to check that our logistic
model fits the assumptions underlying logistic regression:
• The observations are independent. Observations are independent if there is one pair of
observations on each individual.
• The dependent variable must be a dichotomy (i.e. there are two categories).
• No important variables are omitted.
• No extraneous variables are included.
• The independent variables are measured without error.
Furthermore, a logistic regression does not require assumptions of normality, linearity,
and homoscedasticity (homogeneity of variance) usually needed in linear regression. In
summary, in logistic regression:
• The independent variables do not need to be normally distributed, though normality
yields a more stable solution. Also, the residuals do not need to be normally distributed.
• Assumption of a linear relationship between dependent and independent variables is
not needed. As logistic regression applies a non-linear log transformation to the pre-
dicted OR (see below) it can deal with all sorts of relationships.
• Assumption of homoscedasticity (equal variance) of the residuals is also not needed.
We can start to make a logistic regression model by creating a binary variable to represent
two possible outcomes of the dependent variable, such as 1 for having the disease and 0
for not having the disease. This binary variable cannot be used as the dependent variable
because we cannot explain predicted values that are not 0 or 1 and after all such a variable
is also not normally distributed. However, if we consider the probability P of an individual
being classified into the highest coded category (i.e. having the disease) as the depen-
dent variable we can overcome the above problem. In order to predict this probability,
logistic regression employs maximum-likelihood estimation where the outcome is the
probability of belonging to a condition of y (i.e. having the disease), which can take any
value between 0 and 1. To overcome mathematical difficulties we need to transform the

2967_Ch05.indd 160 07/07/2015 10:38

 The Basic Statistical Considerations 161

probability P to a logarithmic distribution. This transformation normalizes the distribu-

tion and creates a link with the linear regression equation. The logarithmic distribution (or
logistic transformation of P) is also called the logit of P or logit(P ) or log odds. Logit(P)
is the natural logarithm (log to base e) of the odds ratio and it is given by the formula:

 P 
logit(P ) = log e 
 1 − P 

According to this formula, loge is a natural logarithm and the ratio (P/1 − P) is an odd
(odds ratio) of the event occurring and can range from 0 to infinity. Odds values tell you
how much more likely it is that an observation is a member of the target group rather than
a member of the other group. For example, if the probability P is 0.80, the odds are 0.80/
(1 – 0.80) or 0.8/0.2 or 4 to 1. The probability P can only range from 0 to 1, but logit(P)
can range from negative infinity to positive infinity. We can also express the above equa-
tion using a linear regression equation in which the logit of P or y (dependent variable) is
determined by a linear function of the independent variables xi (where P is a function of x
corresponding to the dependent variable y):

 P 
logit( y ) = log e 
 1 − P 
= β0 + β1 x1 + β2 x 2 +  + βn xn + ε

The terms y and x represent dependent and independent variables, respectively. Each βi
coefficient measures each xi partial contribution to variation in y. Logistic regression uses
a maximum likelihood method, which maximizes the probability of getting the observed
results given the fitted regression coefficients.
We can use a more complicated logistic regression model when additional covariates may
affect the onset of a disease. Examples of this are situations in which environmental effects
such as epidemiological risk factors (e.g. smoking and gender), clinical variables (e.g. dis-
ease severity and age at onset), population stratification, and other marker loci have inter-
active effects on disease risk (and gene–gene interaction or epistasis).
Fortunately, we do not have to calculate any of these mathematical equations by hand.
Computer software is easily available, but, once again, it is important to understand the
principles and how we get from a regression formula to a line to the logistic analysis.
Odds Ratio (OR)
In order to interpret the logit coefficients, however, we need to introduce the concept of
the OR. The OR is a measure of the strength of an association between two variables and
is widely used in case control studies. In a genetic association study, OR values give us use-
ful information for understanding the relationship between alleles and disease risk, and we
can define the OR as the ratio of the odds of bearing an allele in the group with the disease
(cases) to the odds of carrying the same allele in the group without the disease (controls).
In other words, OR estimates the risk of a disease or trait for a specific allele.
In a standard 2 × 2 table, the OR can be expressed as the cross-products ratio (ad/bc), as
seen in Table 5.3 (also discussed in Chapter 3). In this calculation a, b, c, and d are the
frequencies of the alleles in cases and controls. OR values range from 0 to infinity. OR
values close to 1 indicate no relationship between the allele and the trait, while OR values
of less than 1 suggest a protective effect and values greater than 1 suggest an increased risk.

2967_Ch05.indd 161 07/07/2015 10:38

 162 CHAPTER 5 Statistical Analysis in Complex Disease

Table 5.3 A standard 2 × 2 plot with allele frequencies shown as a, b, c, and d (this table would be
used for calculation of simple χ2 or OR).

Cases Controls
Risk allele a b
Other allele c d

However, causality cannot be established from an association because associations may

arise as a result of linkage with a true disease-causing mutation. In other words, association
analysis detects associations only and these associations may reflect linkage with hitherto
uninvestigated disease-causing genes or variants.
The OR is closely connected to logistic regression by modeling the natural logarithm of
the OR as a linear function of independent variables. This is a powerful and very common
method, and it is calculated using the exponential function eβ and the regression coeffi-
cients from one of the two following equations:

OR = eβ0 + β1x1 + β2 x2 + + βn xn + ε
or

OR = eβ0 eβ1x1 eβ2 x2  eβn xn e ε

In a simple example (logit(y) = β0 + βx + e), a value of β = 2 means that e2 ~ 7.4, which

means that when x increases by one unit, the odds of being affected (y = 1) increase by a
factor of 7.4 when other variables are not changed.

The pitfalls and problems of GWAS

A major challenge in GWAS is false-positive associations typically resulting from type I
statistical errors and/or population structures. In consideration of the fact that GWAS
may have clinical implications, false-positive results are an important issue and it is vital
to reduce the incidence of them. We have already discussed quality control as an essential
step to reduce the number of false-positive findings and we have introduced the impor-
tance of correcting for multiple testing across the number of SNPs. In this section, we
will discuss the impact of some of problems associated with study design and population
structure, its implications, and how to deal with this phenomenon.
Study heterogeneity
One problem in GWAS is that heterogeneity between and across the studies may lead to
equally valid but different conclusions about the role of an allele in disease risk.
Different studies of the same disease may use different clinical criteria reflecting sub-
tle ­differences in disease severity, disease subtype, age of diagnosis, or duration of the
­disease—all of which can influence the outcome of genetic studies.
Population heterogeneity
In addition, there are variations between human populations that may lead to differ-
ent associations in different populations. Evolutionary forces such as random drift, novel

2967_Ch05.indd 162 07/07/2015 10:38

 The Basic Statistical Considerations 163

mutations, and natural selection (less common) act on human populations, and thus vari-
ations in SNP frequencies are seen between the major population groups (this is discussed
in Chapter 1 in some detail). Association tests are statistical assays that detect differences
in SNP or other genetic marker frequencies between cases and controls. Variation within
and between populations may lead to false-positive associations. An example is a SNP in
the complement factor H gene (CFH) which alters the susceptibility to age-related macu-
lar degeneration (AMD). The frequency of the AMD protective A allele in the Yoruba
population of sub-Saharan Africa is four-fold higher than in Europeans. Similarly, another
SNP in the complement system genes, the factor B gene (CFB), shows a higher frequency
in Africans than Europeans. These differences may account for differences in susceptibility
to AMD in these populations.
Recent associations reported an increased risk of Alzheimer’s disease associated with
genetic variation in two SNPs (rs11136000 and rs3818361) of the complement pathway
genes for clusterin (CLU) and for complement receptor 1 (CR1). The CR1 risk allele is
frequent in the European population, but rare in African and Asian populations, while
the CLU risk allele is much more frequent within Asian than African populations. It is
not known whether this variation is a consequence of genetic drift or natural selection. A
meta-analysis including white, African-American, Israeli–Arab, and Caribbean Hispanic
individuals found an association between polymorphisms in CR1 and CLU in populations
with a European ancestry only. In addition, some mutations may be completely absent
from some population groups, e.g. the CARD15 mutations that are present in more than
30% of European Crohn’s disease patients are more or less absent in Asian populations.
Meta-analysis
Meta-analysis is a means of compiling data from previously published studies to focus
on a specific question. Recently, it has been applied to studies of the genetics of common
disease and will be used more often as the number of GWAS grows. Meta-analysis has the
advantage of being able to identify and confirm weak associations because with the vast
numbers included the statistical power of these studies is considerably higher than any
other individual center or single collaborative study. Meta-analysis is frequently used in
clinical drug trials and in the analysis of psychological disorders.

Population structure or population stratification

Population stratification occurs when the study samples consist of several discrete subgroups
of individuals who differ in their ancestry. In the presence of stratification, spurious asso-
ciations can be reported. Thus, checking for population stratification in the study samples
is important to avoid false-positive or false-negative associations. Statistical methods have
been developed and implemented into software to control for population stratification.
Several methods have been proposed for case control studies, including genomic control,
structured association, principal components analysis (PCA), and the ­stratification
score analysis. Genomic control and the structured association methods both use SNPs
that have not been associated with the disease to control for population stratification.

Genomic control
In the presence of population stratification, regular methods for testing association, such
as Pearson’s χ2 method and the Cochran–Armitage test for a trend could produce excess
false positives compared with the nominal significance levels. A popular method used in
the presence of population stratification is genomic control. This method aims to control
for population stratification by first estimating an inflation factor and then adjusting all of
the test statistics. The logic behind genomic control is that the test statistic for association

2967_Ch05.indd 163 07/07/2015 10:38

 164 CHAPTER 5 Statistical Analysis in Complex Disease

is inflated by a multiplicative factor λ, which is proportional to the level of population sub-

structure present. λ is estimated using SNPs that are believed not to be in association with
the phenotype tested (affection status) and are therefore expected to be random samples
from the χ2 distribution with 1 d.f.. One common solution to estimate genomic control is to
align the median of Pearson’s χ2 or Cochran–Armitage test distributions with that of the χ2
distribution with 1 d.f., which is about 0.456. A robust estimate of λ is therefore given by:

median(χ2 )
λ=
0.456
An inflation factor λ of 1 indicates that no population stratification is detected within
the samples. We can use this value of λ to rescale all of the test statistics in a downward
direction. In practice, the test statistics of each SNP are divided by the λ value then the
corresponding adjusted P value can be re-estimated:

χ2
χ2adjusted =
λ
It is important to note that λ denoted above is not same as the λ value used in calculating
disease incidence and prevalence (sibling relative risk).
Structured association
SNP genotypes that are not associated with disease can also be used in structured asso­
ciation analysis. This method has been implemented in STRUCTURE software
(http://pritchardlab.stanford.edu/structure.html) and employs genotype data from
­
the study to determine population structure in a similar method to that used in the
genomic control analysis above. Predictive SNPs are those known to be associated with
a specific ethnic population or other population subgroup. The software then performs
tests for associations within each inferred subpopulation. STRUCTURE may also be
used to identify individuals who do not cluster with the majority of the samples. The
latter samples can then be eliminated from the analysis.
PCA
Another analytical methodology often incorporated into GWAS able to scan for popula-
tion structures is called PCA. The basic idea behind PCA is to measure the data as prin-
cipal components rather than on a normal x–y axis. Principal components are structures
or directions where the data are most spread out and thus they indicate where the most
variance is. If we are able to find those directions where most of the variance is, we can
recognize structures in the data. Though we mainly use math to find the principal com-
ponent throughout eigenvectors and eigenvalues, we can imagine plotting our data into
a scatter plot and using PCA to find the direction where there is most variance. The PCA
will first determine the straight line through the points that are able to catch the most
variance among the data in a manner similar to linear regression. This inferred line is then
used as the first principal axis or principal component to generate a second principal com-
ponent. This process occurs in multidimensional space with one dimension for each of
the variables (i.e. SNPs) included and the lines inferred through the points. At the end of
the process PCA will synthesize the data from a mass of variables into a set of compound
axes. The first axis will explain the most variation, then the second, and so on. Once the
principal components are inferred, individuals can be plotted according to their principal
components and the possible structures in the dataset will then be recognized.

2967_Ch05.indd 164 07/07/2015 10:38

 How To Interpret a GWAS 165

PCA can be also be used in a principal components adjustment analysis to control for type
I errors. Association tests performed on data affected by population stratification can be
adjusted by performing PCA on different subsets of variants (or SNPs). We can first carry
out a PCA using only common variants (set at less than 5%) with minor allele frequencies
(set at 5%), or only low-frequency variants (minor allele frequencies between 1 and 5%) or
only rare variants (minor allele frequencies of less than 1%) or a combination of the above.
Then we can test if population stratification is present or not and eventually perform asso-
ciation inferences on the subset. PCA based on either common variants or low-frequency
variants seems to provide effective control for population stratification, while rare vari-
ants performs less well. PCA based on low-frequency variants seems to adjust better than
common variants, but the use of the former could result in over-adjustment producing a
substantial loss of the power, especially in the absence of population stratification.

5.5 How To Interpret a GWAS

The ultimate goal for all this effort is to identify genes associated with or linked to increased
risk of disease. However, interpreting the results of these studies is a potential minefield.

There are several ways to interpret statistically significant

genetic associations
Significant P values resulting from genetic association studies can indicate one of the
following:
• A direct association, whereby there is a true causal relationship between the allele and
the disease.
• An indirect association, whereby the identified allele is in linkage disequilibrium with
the true causal allele on the same chromosome.
• A false positive result, whereby the association is simply a chance observation, per-
haps resulting from some error in the study design, sample bias, or from population
stratification.
Distinguishing between direct and indirect associations is challenging, and only sequencing
the candidate gene region or dense fine-map genotyping of all available SNPs may resolve
this issue. The detection of associations with a non-synonymous SNP (e.g. a SNP where
variation results in an amino acid change) may indicate that it is the true causal variant in
disease susceptibility. Functional studies, however, should be attempted in order to provide
evidence before assigning a functional relationship to any polymorphism identified.

There are several diagnostic plots that can be used for the
visualization of genome-wide association results
Quantile–quantile plots
In statistics, a quantile–quantile (Q–Q) plot is a plot for comparing probability distri-
butions by plotting their quantile values. It is an effective graphical method to determine
if the two distributions come from the same or two different statistical populations. This
statistical tool is widely used in GWAS to assess the significance of observed associations
(y-axis) compared with the expectations under no association (x-axis).

2967_Ch05.indd 165 07/07/2015 10:38

 166 CHAPTER 5 Statistical Analysis in Complex Disease

(a) (b)

50 50

40 40
Observed χ2

Observed χ2
30 30

20 20

10 10

0 0
0 10 20 30 40 50 0 10 20 30 40 50
Expected χ2 Expected χ2

Figure 5.9: Hypothetical Q–Q plots in GWAS. The figure illustrates two different Q–Q plots. The plot compares
probability distributions using their quantile values. This statistical tool is widely used in GWAS to assess the
significance of observed associations comparing observed values (on the y-axis) and expected values (on the x-axis).
Each dot represents a hypothetical SNP while the blank line is the expected null distribution. Plot (a) illustrates
strong deviation from the expected values and therefore strong genetic association of the trait under study with
SNPs in a heavily genotypes region or spurious associations. Plot (b) indicates very little deviation from the expected
values and may suggest cryptic errors in the study population, in the genotyping, or either true association.

In Figure 5.9(a and b), the observed association statistics such as χ2 or the calculated
–log10P values for each SNP are ranked from smallest to largest and plotted against the
distribution that would be expected under the null hypothesis for no association. If the
two compared distributions are similar, the points in the Q–Q plot will approximately lie
on the null or identity line (denoted y = x), indicating that no association is detected for
each SNP. These plots help to indicate whether the study has generated more significant
results than expected by chance. Deviations from the identity line indicate either some
bias in the observed (assumed) distribution or strong true associations.
As the underlying assumption in GWAS is that the vast majority of assayed SNPs are
not associated with the trait, strong deviations from the null hypothesis suggest either a
strongly associated locus, a heavily genotyped locus (Figure 5.9a) (i.e. an associated gene
with many genotyped SNPs), significant undetected differences in the population struc-
ture (Figure 5.9b), cryptic relatedness in the study population, or errors in genotyping. A
clean Q–Q plot (Figure 5.9a and b) should show a solid distribution matching the y = x
line until it sharply curves at the end, which represents the small number of true associa-
tions among the many thousands of genotyped SNPs.
Manhattan plots
Named after the Manhattan skyline, the plots are widely used to visualize data from
GWAS studies. This type of scatter plot is particularly useful to display data with a large
number of data points. A GWAS Manhattan plot outlines the −log10P value of each SNP
on the y-axis against the genomic coordinates on the x-axis (Figure 5.10). Each dot in the
plot represents a different SNP, with the alternating bands of shading representing differ-
ent chromosomes plotted on the x-axis. The y-axis indicates the strength of the association
between a SNP and the trait under study. As the strongest association carries the smallest
P value, it will be the highest ranked and will be visualized at the top of the plot. A good
Manhattan plot visualizing a robust association study should show the highest ranked
SNPs rising with a column of other associated SNPs symbolized as dots from the same

2967_Ch05.indd 166 07/07/2015 10:38

 How To Interpret a GWAS 167

15

10
–log10P

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 17 19 21 X
16 18 20 22
Chromosome

Figure 5.10: Manhattan plot displaying findings from a GWAS study. This Manhattan plot presents the
findings of a GWAS. Each chromosome can be identified sequentially (1–22, then X and Y) in alternating light and
dark gray coloring. The y-axis shows the –log10P. The significance threshold is set at 10−5 for moderate associations.
Each tagged SNP is represented by a dot distributed according to the –log10P along the 22 autosomes and the X.
P is for 1 d.f. because with each SNP tested individually there are two possibilities. Points above the dotted line
indicate SNPs with a P < 10−5. P < 10−5 or P < 10−7 are significant; the greater the digit after the value 10, the more
significant the observations are. (From The Wellcome Trust Case Control Consortium [2007] Nature 447:661–678.
With permission from Macmillan Publishers Ltd.)

chromosomal location and haplotype. This is the result of linkage between nearby SNPs,
all of which will have the same or similar association signals. A single dot visualizing as
the highest ranked SNPs spread over the whole plot in isolation shows a pattern typical of
genotyping errors and this may bring into question any reported associations.

Linkage disequilibrium is a useful tool in association studies

provided you know how to handle it
Linkage disequilibrium and recombination are discussed in detail in Chapters 1 and 2,
and therefore this subsection will be brief. Linkage refers to a situation where alleles or
SNPs at specific loci are inherited together as a consequence of their physical proximity on
a single chromosome. When two loci are close together, genetic recombination will occur
less frequently than when they are further apart. Assuming recombination is a random
event, all alleles would be independently assorted and all alleles would be in linkage equi-
librium. However, recombination is not random. There are recombination hot spots and
cold spots. Therefore, some alleles are found together more often than they would be by
chance. These alleles are “out of equilibrium” and they are therefore said to be in “linkage
disequilibrium”. The level of linkage disequilibrium depends on the difference between
observed (non-random) and expected (assuming random distributions) frequencies. By
exploiting this feature of non-random association of alleles it is possible to detect genetic
associations even if the causal polymorphism is not genotyped. Linkage disequilibrium is
both a blessing and a curse. While it can help to identify associations with linked alleles,
which may be for some reason difficult to genotype directly, it can make identification of
the primary risk allele or locus very difficult, as we shall see in Chapter 6.

The ability to detect a significant association through linkage

disequilibrium can increase the power of an association study
Several different measures of linkage disequilibrium have been proposed in the literature.
The simplest one is the covariance D, but the two most commonly used are D′ and r2. In
order to understand this we can consider two biallelic loci A and B, bearing alleles A and a
at the A locus, and B and b at the B locus, which result in four possible haplotypes: AB, Ab,
aB, and ab. The alleles have observed frequencies fA, fa, fB, and fb, as shown in the Punnett
Square below. The Punnett Square also shows all possible haplotypes with the frequencies
fAB, fAb, faB, and fab.

2967_Ch05.indd 167 07/07/2015 10:38

 168 CHAPTER 5 Statistical Analysis in Complex Disease

Locus 2
Locus 1 B b Totals for rows
A fAB fAb fA
a faB fab fa
Totals for columns fB fb 1

Calculating D
D is a measure of linkage disequilibrium, which is determined as the difference between
the observed frequency of a two-locus haplotype and the frequency that we would expect
if the alleles are randomly segregated. Suppose that fAB is the observed frequency of the
haplotype AB formed by the two alleles A and B. The expected haplotype frequency for AB
is calculated by the product of the allele frequency of each of the two alleles:
fAB = fA × fB
If the alleles are in linkage equilibrium, the observed frequency of the AB in the test popu-
lation will be the same as that which is expected. However, if they are in linkage disequi-
librium, we would expect a departure from the expected frequency.
fAB = fA × fB + D
faB = fa × fB – D
fAb = fA × fb – D
fab = fa × fb + D
The value of D can be obtained by anyone of the above equations as all will result in the
same value of D. D can also be calculated by summarizing all of the above equations:
D = fAB × fab – fAb × faB
When D = 0 there is no linkage disequilibrium. The method can be further illustrated
by considering a population where the two loci above show the observed genotyping
frequencies as:
fAB = 0.85
fAb = 0.05
faB = 0.05
fab = 0.05
The frequencies of each allele are first calculated from the observed data as:
fA = fAB + fAb = 0.85 + 0.05 = 0.9
fa = faB + fab = 0.05 + 0.05 = 0.1
fB = fAB + faB = 0.85 + 0.05 = 0.9
fb = fAb + fab = 0.05 + 0.05 = 0.1

2967_Ch05.indd 168 07/07/2015 10:38

 How To Interpret a GWAS 169

Then D can be calculated using the below equations:

DAB = fAB – fA × fB = 0.85 – (0.9 × 0.9) = 0.04
DaB = fa × fB – faB = (0.1 × 0.9) – 0.05 = 0.04
DAb = fa × fB – fAb = (0.1 × 0.9) – 0.05 = 0.04
Dab = fab – fa × fb = 0.05 – (0.1 × 0.1) = 0.04
D = fAB × fab – fAb × faB = 0.85 × 0.05 – (0.05 × 0.05) = 0.04
Note that all the equations give the same value of D (0.04). Though D intuitively explains
the concept of linkage disequilibrium, the numerical value is of little use as it does not
really quantify the strength of linkage disequilibrium as D varies as allele frequencies
change. Consequently, D is no longer used to estimate degrees of linkage disequilibrium,
but it remains important as both D′ and r2 discussed below are derived from D.
Calculating D′
D′ is a widely used measure of linkage disequilibrium and is the ratio of D compared with
its maximum value. The absolute value of D′ is determined by dividing D by its maximum
value, given the allele frequencies at the two loci:

D
D′ =
D max

where |D|max = min( fA × fb, fa × fB). This value is calculated from D as:

If D > 0, |D|max is equal to the smaller value between fA × fb and fa × fB
If D < 0, |D|max is equal to the smaller value between fA × fB and fa × fb
A value of D′ = 1 denotes complete linkage disequilibrium. Over time, repeated recom-
bination results in the decay of D′ towards 0. However, when D′ < 1 there is no clear
indication of the strength of linkage disequilibrium and it may be difficult to determine
whether alleles are in linkage disequilibrium or not. As a rule of thumb, when D′ = 1 there
is linkage disequilibrium between the two loci, when D′ < 1, the likelihood of linkage
disequilibrium increases with the increasing proximity of D′ to 1. Haploview software, for
example, visualizes a strong linkage disequilibrium when D > 0.95. One disadvantage of
using D′ is that it is highly inflated in small samples and when alleles are rare.
Calculating r 2
r2 is the statistical coefficient of determination, a measurement of correlation between a
pair of variables. The measure r2 is in some ways complementary to D′ as r2 is equal to D2
divided by the product of the allele frequencies at the two loci:

D2
r2 =
f A f a f B fb

where fA, fa, fB, and fb are the frequencies of each allele, and D is the linkage disequilibrium
measure introduced above. r2 ranges between 0 (loci are in complete linkage equilibrium)
and 1 (loci are in complete linkage disequilibrium).

2967_Ch05.indd 169 07/07/2015 10:38

 170 CHAPTER 5 Statistical Analysis in Complex Disease

If the value of r2 between two loci equals 1 this implies the markers at the two loci are
providing the same information. Consequently, the genotypes of alleles of one SNP are
directly predictive of the genotypes of the other and genotyping either SNP or allele will
result in the same P value. In the light of this it is sometimes reasonable to genotype only
one of the two selected SNPs. This can be helpful if specific SNP sequences are difficult to
genotype. A surrogate SNP used in this manner is often referred to as tagged SNP.
The relationship between D′ and r 2
D′ and r2 are both measures that can be written in terms of allele frequencies. To illustrate
this, consider D′ such that D ≥ 0 and fA ≥ fB. In this case, Dmax = fa fB (being fA × fb > fa × fB).
In this case:

D
D′ =
fa fB

We can extrapolate D from the above equation as D = D′ × fa  fB and substitute r2 in the
equation, thus:

fa fB
r 2 = ( D ′ )2 ×
f A fb

The above equation describes the relationship between D′, r2, and allele frequencies
when D ≥ 0 and fA ≥ fB. The upper boundary of r2 is D′ and is reached only when fA = fB.
The implication of this is that D′ provides the upper limits of r2. D′ is a commonly used
measure of recombination and gives information on the physical extent of useful linkage
disequilibrium. If we know the value of D′ we can calculate r2. For example, if a recom-
bination point results in a D′ = 0.7, the maximum possible r2 for these alleles would be
0.49.

Most association analyses identify multiple SNPs, other genetic

variants, and haplotypes
Analyses based on single SNPs have limited statistical power to detect true genetic asso-
ciations. Most genes encode multiple SNPs (and other genetic variants) and it is more
appropriate to consider haplotypes composed of multiple genetic variants (including
SNPs). A haplotype is a set of genetic polymorphisms (mostly SNPs) found on the same
chromosome. Some haplotypes exhibit extreme linkage disequilibrium (for example
HLA 7.2 and HLA 8.1 in Chapter 6). Haplotypes can be inferred throughout a process
called phasing. This term is important in linkage analysis as described in the example of
nail–patella syndrome at the beginning of this chapter. This method assigns each allele
of the genotype to one or other chromosome and is based on the observation that cer-
tain haplotypes are common in certain regions. Several phasing algorithms have been
developed, among which a popular algorithm is the PHASE program that was used in
the HapMap project (http://hapmap.ncbi.nlm.nih.gov/). This program uses posterior
probabilities implemented in a Bayesian approach in order to capture the tendency of
haplotypes to cluster together over regions of the chromosome. A Bayesian approach
incorporates the prior probability of association. As clustering can change as we move
along the chromosome due to recombination, a flexible model for the decay of linkage
disequilibrium is then used.

2967_Ch05.indd 170 07/07/2015 10:38

 Conclusions 171

Once phased, the inferred haplotype can be assessed for associations using the same statis-
tical methods as described previously. However, there are problems with haplotype-based
analyses. For example, inclusion of rare haplotypes in studies increases the number of tests
performed and reduces the statistical power of the testing. A common solution to this
problem is to combine all of the rare haplotypes into a single category. By doing this we
preserve the data and maintain the statistical power of the study. However, this could also
introduce problems because individual rare haplotypes may be important in disease and
these associations may be overlooked when clustered in this way.

Conclusions
In complex disease it is important to include statistics into study design and to do so at
the earliest moment. Deciding what analysis packages and procedures to use after a study
has been performed is not good practice and may prejudice the final conclusions. Data
handling in studies of the genetics of complex disease has become complicated. This is the
era of computerized science. Without computers it would not be possible to handle the
data in the way we do. The Human Genome Project would have been impossible and
the future would simply not exist for this type of science. Fortunately, it does exist and
statistical analysis on various computer platforms helps us to achieve our goals. This is also
the era of collaboration when competing centers work together for the greater good, by
pooling samples so that statistical thresholds can be met and statistical power enhanced.
Meta-analysis is the natural extension of this collaboration.
When looking at data analysis in complex disease we can start with calculating LOD
scores in linkage studies and calculating ORs in association studies. Of course we need
to consider the problems with false-positive and false-negative findings and the impact of
sample size (large-scale studies) on statistical power. We need to be aware of sample relat-
edness, case call rate, and SNP call rate before we begin any study.
Whether we are testing allelic polymorphisms in a single gene or the whole genome we can
use Pearson’s χ2 test, Fisher’s exact probability test, and the Cochran–Armitage test. There
are different models in the Cochran–Armitage test. The next step in complexity is logistic
regression or linear regression analysis; however, these are complex tests and researchers
are advised to use appropriate computer software and refer to genuine statistical genetics
experts for guidance and help with their studies, but a little basic knowledge of the meth-
ods is also advised. It is also important to understand the limitations and potential pitfalls
for different types of study, and consider how those pitfalls may be overcome by good
study design and use of advanced analytical techniques. Other things to consider are how
we present our data, e.g. should we use Q–Q plots or Manhattan plots.

2967_Ch05.indd 171 07/07/2015 10:38

 172 CHAPTER 5 Statistical Analysis in Complex Disease

Further Reading
Books geographical distribution and relevance to dis-
Strachan T & Read AP (2011) Human ease. Immunobiology 217:265–271.
Molecular Genetics, 4th ed. Garland Science. Ewens WJ & Spielman RS (1995) The transmis-
Chapter 16 deals with identifying human genes sion/disequilibrium test: history, subdivision,
and susceptibility factors, and provides excellent and admixture. Am J Hum Genet 57:455–464.
additional data for students wishing to delve Hill WG & Robertson A (1968) The effects of
deeper into medical genetics. There is also addi- inbreeding at loci with heterozygote advantage.
tional relevant information in the other chapters. Genetics 60:615–628.
Hirschhorn JN, Lohmueller K, Byrne E &
Articles Hirschhorn K (2002) A comprehensive review of
Armitage P (1955) Tests for linear trends genetic association studies. Genet Med 4:45–61.
in proportions and frequencies. Biometrics Ke X, Hunt S, Tapper W et  al. (2004) The
11:375–386. impact of SNP density on fine-scale patterns
Balding DJ (2006) A tutorial on statistical of linkage disequilibrium. Hum Mol Genet
methods for population association studies. 13:577–588.
Nature Rev Genet 7:781–791. Lewis CM & Knight J (2012) Introduction to
Barrett JC, Fry B, Maller J & Daly MJ (2005) genetic association studies. Cold Spring Harbor
Haploview: analysis and visualization of LD and Protoc 2012:297–306.
haplotype maps. Bioinformatics 21:263–265. Lohmueller KE, Pearce CL, Pike M et  al.
Chapman JM, Cooper JD, Todd JA & Clayton (2003) Meta-analysis of genetic association
DG (2003) Detecting disease associations due studies supports a contribution of common
to linkage disequilibrium using haplotype tags: variants to susceptibility to common disease.
a class of tests and the determinants of statisti- Nat Genet 33:177–182. This is a useful text on
cal power. Hum Hered 56:18–31. meta-analysis in common complex disease.
Clarke GM, Anderson CA, Pettersson FH et al. McIntosh I, Dunston JA, Liu L et  al. (2005)
(2011) Basic statistical analysis in genetic case- Nail patella syndrome revisited: 50 years after
control studies. Nat Protoc 6:121–133. linkage. Ann Hum Genet 69:349–363. This is a
Clayton DG, Walker NM, Smyth DJ et  al. useful update on nail–patella syndrome genetics.
(2005) Population structure, differential bias Morton NE (1956) The detection and estima-
and genomic control in a large-scale, case-con- tion of linkage between the genes for elliptocy-
trol association study. Nat Genet 37:1243–1246. tosis and the Rh blood type1. Am J Hum Genet
Collins FS & McKusick VA (2001) 8:80–96.
Implications of the human genome project for Price AL, Patterson NJ, Plenge RM et  al.
medical science. JAMA 285:540–544. This is a (2006) Principal components analysis corrects
very important review that provides the reader for stratification in genome–wide association
with a guide to and an illustration of the poten- studies. Nat Genet 38:904–909.
tial for understanding complex disease in the Pritchard JK, Stephens M & Donnelly P (2000)
immediate post-genome period. Inference of population structure using multi-
de Bakker PI, Yelensky R, Pe’er I et al. (2005) locus genotype data. Genetics 155:945–959.
Efficiency and power in genetic association Purcell S, Neale B, Todd-Brown K et al. (2007)
studies. Nat Genet 37:1217–1223. PLINK: a tool set for whole-genome associa-
Dudbridge F & Gusnanto A (2008) Estimation tion and population-based linkage analyses. Am
of significance thresholds for genome-wide J Hum Genet 81:559–575.
association scans. Genet Epidemiol 32:227–234. Sasieni PD (1997) From genotypes to
Epstein MP, Allen AS & Satten GA (2007) A genes: doubling the sample size. Biometrics
simple and improved correction for population 53:1253–1261.
stratification in case control studies. Am J Hum Schaid DJ & Jacobsen SJ (1999) Biased tests
Genet 80:921–930. of association: comparisons of allele frequencies
Ermini L, Wilson IJ, Goodship TH & Sheerin when departing from Hardy–Weinberg pro-
NS (2012) Complement polymorphisms: portions. Am J Epidemiol 149:706–711.

2967_Ch05.indd 172 07/07/2015 10:38

 Further Reading 173

Stephens M & Scheet P (2005) Accounting for Wang WY, Barratt BJ, Clayton DG & Todd JA
decay of linkage disequilibrium in haplotype (2005). Genome-wide association studies: the-
inference and missing-data imputation. Am J oretical and practical concerns. Nat Rev Genet
Hum Genet 76:449–462. 6:109–118.
Snyder LH (1932) Studies in human inheritance Wittke-Thompson JK, Pluzhnikov A & Cox
IX. The inheritance of taste deficiency in man. NJ (2005) Rational inferences about departures
Ohio J Sci 32:436–468. This paper is considered from Hardy–Weinberg equilibrium. Am J Hum
to be the first direct pharmacogenetics study. Genet 76:967–986.
The 1000 Genomes Project Consortium (2010) Zhang K, Calabrese P, Nordborg M & Sun
A map of human genome variation from pop- F (2002). Haplotype block structure and its
ulation-scale sequencing. Nature 467:1061– applications to association studies: power and
1073. This is a very exciting project with the study designs. Am J Hum Genet 71:1386–1394.
promise of more yet to come. This is a good paper to read to understand how
The ENCODE Project Consortium (2012) An haplotypes can be used in GWAS.
integrated encyclopedia of DNA elements in Zhang Y, Guan W & Pan W (2013). Adjustment
the human genome. Nature 489:57–74. This for population stratification via principal com-
is an invaluable research paper for those start- ponents in association analysis of rare variants.
ing out on the investigation of the genetics of Genet Epidemiol 37:99–109.
human disease.
The Human Genome (2001) Nature 409:813–
958. The complete issue of Nature mostly Online sources
dedicated to the Human Genome Map with
multiple papers, letters, and editorial commen- http://hapmap.ncbi.nlm.nih.gov
tary from multiple authors. It is full of useful This is the site of the results of the International
insight and critical discussion. Any student of HapMap project.
human genetics should read this issue. http://pngu.mgh.harvard.edu/~purcell/plink
The International HapMap Consortium (2005) PLINK is a freely available, open-source, whole-
A haplotype map of the human genome. Nature genome association analysis toolset designed
437:1299–1320. for quality and analysis of GWAS data.
The International HapMap Consortium (2007) http://pritchardlab.stanford.edu/structure.html
A second generation human haplotype map of This site includes STRUCTURE software
over 3.1 million SNPs. Nature 449:851–861. referred to in the text.
The International HapMap 3 Consortium This is an essential site for updates and informa-
(2010) Integrating common and rare genetic tion on genetic disease of all types.
variation in diverse human populations. Nature http://www.ncbi.nlm.nih.gov/projects/SNP/
467:52–58. This is essential reading for those snp_summary.cgi
involved in complex disease genetics. This site provides information about millions of
The Wellcome Trust Case Control Consortium validated reference SNP clusters in the genome.
(2007) Genome-wide association study of
14,000 cases of seven common diseases and http://www.ncbi.nlm.nih.gov/SNP
3,000 shared controls. Nature 447:661–678. This an essential site for updates on SNPs.
This is another landmark paper highlighting www.r-project.org
the development and application of new tech- This is another freely available and excellent
nologies (GWAS) in complex disease research, package for statistical computing that can be
and it provides a mass of useful information for used with or without existing commercial pack-
those studying complex disease. ages such as SPSS or SAS.

2967_Ch05.indd 173 07/07/2015 10:38

2697_Prelims.indd 2 09/07/2015 09:35
 CHAPTER
6
The Major Histocompatibility
Complex

The extended major histocompatibility complex (xMHC) located on the short arm of
chromosome 6 (6p21.3) is perhaps the most complex genetic system in the human genome
with 252 expressed genes in a region of approximately 7.6 Mb of DNA. The MHC is so
named because it encodes genes with vital functions in determining the compatibility of tis-
sues. In particular, it encodes the human leukocyte antigens (HLAs), matching of which is
vital in some, but not all, forms of clinical transplantation. JJ van Rood, one of the pioneers
of immunogenetics, suggested that all discoveries go through three phases: the discovery
itself, the development of tools with which to investigate the discovery, and the final expo-
sure of the discovery (the latter of which he likened to the charting of a new continent). The
scientific process is an iterative cycle of discovery, hypothesis generation, and testing, con-
tinually looking back. This is particularly appropriate when we consider the human MHC.
This chapter will briefly examine the discovery of the MHC, the complex naming of HLA
antigens and alleles, the structure and normal function of the products of the HLA alleles,
and how genetic variation in HLA may determine susceptibility and resistance to disease
using a number of key examples. We will also consider a selection of other non-HLA
MHC-encoded immunoregulatory genes and discuss how they may also play a role in
susceptibility to complex disease.
Selecting disease examples for this section is a very difficult task because there are many
diseases with genetic associations with the MHC. Genetic associations with HLA were
first reported in 1967 by Amiel, who reported an association between HLA and non-
Hodgkin’s lymphoma. Later, in the 1970s, Ceppellini et  al. reported an association

2967_Ch06.indd 175 07/07/2015 11:57

 176 CHAPTER 6 The Major Histocompatibility Complex

between endemic malaria and HLA in four Sardinian villages. Reports of other associa-
tions quickly followed: HLA-B8 and celiac disease by Stokes et al. in 1972, HLA-A3 and
multiple sclerosis by Terasaki et al. in 1972, B13 and psoriasis by Russell et al. in 1972,
and B27 and ankylosing spondylitis by Brewerton et al. in 1973. There are currently so
many genetic associations with HLA that there are complete textbooks devoted to the
subject. Therefore, we will focus here on a small number of illustrative examples, some
of which are selected because they are rarely discussed, and some because they are fre-
quently discussed and have become exemplary models. The selected examples in this chap-
ter include hemochromatosis, psoriasis (type I and type II), severe cataplectic narcolepsy,
three autoimmune liver diseases [primary sclerosing cholangitis (PSC), primary biliary
cirrhosis (PBC) and autoimmune hepatitis (AIH)], multiple sclerosis, and systemic lupus
erythematosus (SLE). The selection illustrates the impact of the MHC immunoregulatory
genes in clinical disease, and informs the debate on disease diagnosis, treatment and care,
and disease pathogenesis for a large variety of common and some less common diseases.
Type 1 and type 2 diabetes (T1D and T2D), rheumatoid arthritis, ankylosing spondylitis,
cancer, and infectious disease are all discussed elsewhere in this book as part of other chap-
ters or as specific chapters, and therefore the role of the MHC and MHC encoded genes
in these diseases are not included here.

6.1 Histocompatibility
Many geneticists avoid discussing the MHC because of its complexity and as a conse-
quence it is seen by many as an area for specialists only. These specialists are mostly tissue-
typers who provide data for transplant programs in particular. A summary of the history
of the major developments in the MHC is shown in Figure 6.1.

The idea of histocompatibility first started with blood groups

ABO blood groups can technically be considered as the first detected histocompatibil-
ity antigens in man. Histocompatibility refers to the acceptance (compatibility) of tissue
and is normally applied to the acceptance of tissue transferred between individuals via
either blood transfusion or solid-organ transplantation. Histocompatibility can involve
exchange between species (xeno-transplantation) or, more frequently, between individuals
of the same species (allo-transplantation).
Interest in compatibility of tissues arises from the clinical problems associated with incom-
patibility, i.e. the failure of the host or donor tissue to tolerate this transfer. The ABO
system is not encoded within the MHC, or even on the same chromosome, but it is impor-
tant in clinical transplantation, especially in skin, kidney, heart, and liver transplantation.
When blood transfusions are performed between A or B incompatible donors naturally
occurring antibodies in the recipient attack the transferred blood from the donor and this
causes agglutination and hemolysis (Figure 6.2). The problem has been known since James
Blundell published his report of a blood transfusion in 1829, but the reasons for the trans-
fusion reaction were not fully understood until Karl Landsteiner published his work on the
ABO blood groups in 1900.

The MHC-encoded HLA antigens are the second major

histocompatibility group
The first clues to the existence of the MHC came from mouse models of tumor immunol-
ogy. It was clear from 1900 onwards that factors other than the blood groups influenced

2967_Ch06.indd 176 07/07/2015 11:57

 Histocompatibility 177

GWAS
HGMP
1991 PCR
1988 RFLP
1987 HLA and T1D
1987 A2 crystal
1984 Classification class II
1970 HLA-Cw (now C)
1967 HLA-B (and haplotypes)
1967 First IHTW
1964 Microlymphocytoxicity test
1963–1965 HL-A (and MAC now HLA-A0*2)
1958 Antibodies from pregnancy for HLA
1944–1946 Peter Medawar – LA
1937 Peter Gorer – H2 mouse MHC
Little, Snell, et al., – Mouse tumors and H2
1900 Karl Landsteiner – ABO blood groups
1829 James Blundell – Blood transfusion

Figure 6.1: Time pyramid for major discoveries and developments in the world of HLA. The figure is a time
pyramid illustrating the timing of some of the major discoveries and developments in the world of HLA. It shows
selected developments over the last millennium from recent major technological advances to early discoveries
that underpin our knowledge today. As students of science we should not forget the principle that today’s work
is based on the work of those who came before us. Many of the above achieved fame in their own lifetimes,
but not all are as well remembered as they deserve. The list includes several Noble Prize winners. The top of the
pyramid is left unfilled for the next great leap ahead.

the outcome of tissue grafts in both animals and in humans. These factors turned out to
be encoded in the MHC. In the early days, at least, success or failure in clinical transplan-
tation could depend on matching donor and recipient HLA types. The HLA antigens
encoded in the MHC are important in bone marrow and kidney transplantation, and less
important in heart and liver transplantation. The history of the discovery of the MHC is
discussed in brief in Box 6.1.

Naming the HLA antigens and alleles up to and including the early
molecular genotyping era
In addition to being one of the most complex systems in the human genome, some of the
genes in the MHC are the most polymorphic. To understand the nomenclature applied to
HLA we need to look back at the history; however, because this is neither an immunology
book nor a history book, a prolonged discussion of early nomenclature is not appropriate
here. Table 6.1 indicates some of the chaos that was present in the early world of HLA
typing when it came to naming new alleles and genes, and how this was adjusted to cre-
ate current standard nomenclature. Therefore, this will only be dealt with in brief. The
International Histocompatibility Testing Workshops (IHTWs) that were first set up in

2967_Ch06.indd 177 07/07/2015 11:57

 178 CHAPTER 6 The Major Histocompatibility Complex

Donor blood group

Recipient
blood O A B AB
group

O
Anti-A
Anti-B
antibodies

A
Anti-B
antibodies

B
Anti-A
antibodies

AB
No ABO
antibodies

Figure 6.2: Donor and recipient interaction for ABO blood groups. The four main ABO blood groups are
encoded on chromosome 9. Those with blood group O can donate to all other groups, but because they can
produce antibodies to A and B antigens they can only receive blood from O donors. In contrast, those with AB
can only donate to other AB carriers because O, A, and B carriers can all produce antibodies to either A and B
(O blood group), or A (B blood group) or B (A blood group). However, AB-positive individuals do not produce
antibodies against the other blood groups and can receive blood from any donor.

BOX 6.1 OF MICE AND MEN: HOW EARLY WORK WITH MICE REVEALED
THE POSSIBILITY OF THE MHC AND THE HLA ANTIGENS

The first clues to the existence of the MHC came from mouse models of tumor immunol-
ogy. As early as 1903, Jensen noted that different strains of mice were not equally suscep-
tible to the growth of grafted tumor cells. Little and colleagues, who were also working
on mice, proposed that acceptance of tumor grafts depended on the graft and host tis-
sues possessing a large number of susceptibility factors in common. Bauer, working with
human subjects in 1927, had also noted the acceptance of skin grafts in identical twins
was consistent with the theory that compatibility is genetically determined. In 1937,
Peter Gorer proposed a genetic theory for transplant rejection: “Normal and neoplastic
tissues contain isoantigenic factors (which are) genetically determined. Isoantigenic fac-
tors present in the grafted tissue and absent (from the) host are capable of eliciting a
response which results in the destruction of the graft.”
These ideas may not seem revolutionary in 2015, but in 1937 they represented a great
leap forward in both genetics and immunology, and Peter Gorer is considered one of
the founding fathers of immunogenetics. Peter Medawar followed up Gorer’s idea work-
ing on skin grafts in rabbits from which he was able to suggest that leukocytes and skin
share important transplantation antigens. These were the first ideas pointing at HLA.
Leukocytes are white blood cells—specifically, the mononuclear cells and these antigens
were later called leukocyte antigens. After much work by many different groups, Jean
Dausset published his hypothesis in 1952 suggesting that a similar antigenic system to
that seen in mouse erythrocytes (red cells) by Gorer existed on the surface of human

2967_Ch06.indd 178 07/07/2015 11:57

 Histocompatibility 179

BOX 6.1 OF MICE AND MEN: HOW EARLY WORK WITH MICE REVEALED
THE POSSIBILITY OF THE MHC AND THE HLA ANTIGENS (Continued)

leukocytes. Subsequently the “antigens” were called “human leukocyte antigens” (HL-A
or HL-antigen and finally HLA). One distinct feature of the HL-antigen is that, these
antigens are not expressed on red blood cells, unlike the ABO blood groups. This system
is the HLA system we know today and Dausset and several others were later awarded the
Noble Prize for this work and related work.

Table 6.1 Early nomenclature for HLA.

Example early name(s)

assigned Locus or allele Adjusted (current) name
Class I genes Each antigen is constructed Each gene is highly
from a single α-chain supported polymorphic
by a β-2-microglobulin
molecule encoded on
chromosome 15
LA1, LA2, LA3 HLA-A HLA-A
HU-1 HLA-A HLA-A
4a, 4b HLA Bw4 and Bw6 Bw4 and Bw6
(see Table 6.3)
W5 HLA B35 B*35:
4c HLA B5 Describes a series of families
(superfamily)
Class II genes Each antigen has two genes: Each gene is polymorphic; the
A encoding the α-chain and B level of polymorphism varies
encoding the β-chain (see Figure 6.8)
D-related HLA-DR There are nine different DRB
genes, four are expressed. DRB1
is the predominant DRB gene
DC1, MB1, (LB12), MT1 DQ DQ1
DC3, MB2, LB-E17, Te24 DQ DQ2
DC4, MB3, LB13, TA10, DQ DQ3
MT4, TB21
LB13 DQ DQ3 (split)
TA10 DQ DQ3 (split)
MT2 DRw52 DRB3
MT3, BR3, BR4 DRw53 DRB4
SB DP DP

The table illustrates the complexity of the original naming systems for new antigens in the left-hand column.
Through a series of committees and workshops, these differences have been clarified so that a single naming
system works. A substantial restructuring of HLA naming has taken place in recent times.

2967_Ch06.indd 179 07/07/2015 11:57

 180 CHAPTER 6 The Major Histocompatibility Complex

1965 have ensured a consistent procedure for identification, reporting, and naming of all
HLA alleles. The reason for this is so that different laboratories can perform HLA typing
on the same sample and produce the same result. The history of how this came together is
indicated in Box 6.2. Standardization of the naming system for HLA was essentially for
clinical transplantation, where matching of donor and recipient is required.

BOX 6.2 EARLY HLA NOMENCLATURE: CREATING

INTERNATIONAL STANDARDS

Early studies suggested a single locus for the histocompatibility antigen. Bodmer & Payne
proposed a prefix of LA for leukocyte antigen. Dausset proposed the prefix HU for human
system. Using the latter system, HU-1 meaning human system-1 was the first antigen.
This was a time of considerable confusion. Each group began naming identified antigens
differently: 4a and 4b (Van Rood and Leeuwen, 1963), LA1, LA2, and LA3 (Payne and
Bodmer, 1964), Ao (Amos), Bt (Batchelor), To (Ceppellini, who is credited with coining
the term “haplotype”), Da (Dausset) and Te (Terasaki, who was instrumental in develop-
ing the microcytotoxicity test). These early pioneers soon realized that it would be essential
to standardize the systems for naming and testing these HLA antigens. In order to do this
they assembled and decided on steps to unify and clarify the naming and identity of the
HLA antigens. The first step was to set up regular International Histocompatibility Testing
Workshops (IHTWs). The first of these was held in 1965. In 1968, the term HL-A stand-
ing for “Human Leukocyte A” was coined; the hyphen was dropped from the name follow-
ing the recognition of other class I loci. The different antigen families (genes today) were
called HLA A, HLA B, and HLA Cw. As part of standard procedure all newly identified
antigens were given a “w” until confirmed at the next IHTW. HLA C antigens (and later
alleles) retained the “w” in their name because complement components are also named
using an alpha-numeric code with C as the letter and a single number. However, C alleles
are no longer indicated by Cw, but by C alone.
Throughout the 1970s and early 1980s, a number of authors proposed the existence of
novel class II loci and alleles; these included DC and MB (some of which are the same as
LB), MT, SB, Te, and BR. This was very much like a repeat of the late 1960s and early
1970s. At the 1984 IHTW these novel class II antigens were reclassified and assorted,
giving rise to the present DR, DQ, and DP picture for class II. Thus, DC, LB, and MB
were all identified as DQ antigens; MT1, MB1, and DC1 were recognized as equivalent
to DQ1; MT2 and MT3 were reassigned DRw52. BR3, BR4, Te24, and MB2 were
reassigned DRw53. DC3 and LB-E17 were assigned as DQ2. MB3, MT4, DC4, and
TB21 were reassigned DQ3. All SB antigens were identified as belonging to a single
locus renamed DP to fit the sequence DR, DQ, DP. While some HLA DQ antigens were
detectable by serology, HLA DP antigens (previously SB) were not.
Current nomenclature is focused on the genotype rather than the phenotype as before.
There is an enormous amount of polymorphism in the HLA genes as seen in this chapter
and consequently in 2010 the nomenclature was reformed. The group of genes is referred
to in italics with the suffix HLA then each locus joined by a hyphen (e.g. HLA-A) next the
allele or allele family is identified (e.g. HLA-A*01) with A*01 referring to the first group
or family of alleles of the HLA-A gene family. These first two digits refer to the antigen
and are essentially the same as the serotype or phenotype. In early studies the example

2967_Ch06.indd 180 07/07/2015 11:57

 Histocompatibility 181

BOX 6.2 EARLY HLA NOMENCLATURE: CREATING

INTERNATIONAL STANDARDS (Continued)

above would be the A1 antigen. In the new era sub-typing this group means there has
to be a greater number of possible alleles. This means more codes are needed to distin-
guish them all from each other. Therefore, the use of a colon and second set of digits is
applied to identify variants of the HLA-A*01 family thus HLA-A*01:01. These alleles
must encode an amino acid change. If the variant does not encode an amino acid change
(i.e. it is a synonymous polymorphism or noncoding polymorphism) it is listed using a
third set of digits after the insertion of a second colon for example HLA-A*01:01:01.
Alleles that encode changes in the intronic sequences are listed by use of a fourth set of
digits after a third colon thus HLA-A*01:01:01:01. The new naming system is designed
to cope with the most polymorphic genetic system in the human genome.

Antigens are proteins; in the case of HLA, these are usually expressed on the cell sur-
face. Alleles are genetic variants, which in the case of HLA usually refers to the region
of the gene that encodes these proteins. The two terms are often used interchangeably.
Early work with serum and cells was based on detecting antigens. Later work based on
restriction fragment length polymorphism (RFLP) or polymerase chain reaction (PCR)
analysis detected genetic variants and therefore uses the term alleles. Occasionally, depend-
ing on the level of specificity applied, the term family may also be used, i.e. the HLA A2
family, where HLA A2 is tested rather than a range of different HLA A2 variants. The term
family will be applied to both genes and antigens when appropriate.
HLA class I
Understanding nomenclature is difficult in any subject and no more so than with regard to
HLA class I. Although there are many HLA class I genes, there are only three major HLA
class I genes that have been studied in detail: HLA-A, HLA-B, and HLA-C.
HLA class II
The major HLA class II antigens are DR, DQ, and DP encoded by the gene pairs DRA
and DRB, DQA and DQB, and DPA and DPB. The changes heralded by molecular geno-
typing, which allowed more accurate HLA typing (genotyping), also heralded a new
era. They confirmed what serology had suggested, i.e. that there may be more than one
expressed DRB gene on some, but not all, MHC haplotypes, and identified the DQ and
DP families of genes with a single pair of expressed DQA and DQB genes encoding the
DQ molecule and a single pair of expressed DPA and DPB genes encoding the DP mol-
ecule for all MHC haplotypes.
The second expressed DRB gene
There are nine DRB genes numbered DRB1 to DRB9. The first of these, DRB1, is expressed
on all haplotypes. Some haplotypes express only the DRB1 gene (DRB1) and others
express a second DRB gene. Which second DRB gene is encoded on each haplotype can
be determined from the DRB1 family encoded on each haplotype. The second expressed
DRB gene on haplotypes encoding alleles of the DRB1*15 and DRB1*16 families is
DRB5. DRB3 is the second expressed DRB gene expressed on haplotypes encoding alleles
of the DRB1*03, DRB1*11, DRB1*12, DRB1*13, and DRB1*14 families. DRB4 is the
second expressed DRB gene on haplotypes encoding alleles of the DRB1*04, DRB1*07,

2967_Ch06.indd 181 07/07/2015 11:57

 182 CHAPTER 6 The Major Histocompatibility Complex

and DRB1*09 families. The DRB1 families that express only a single DRB1 gene are the
DRB1*01, DRB1*08, and DRB1*10 families.
The second expressed DRB genes are all thought to be only weakly expressed in each case.
However, even though these genes are thought to be weakly expressed in comparison
with DRB1, expression levels may alter under immune stress. The significance of having
two, as opposed to one, expressed DRB gene, on some haplotypes, both of which may be
polymorphic, is not properly understood in the context of complex disease, but the second
DRB locus should not be overlooked.
The DRB pseudogenes
The final set of DRB genes to consider are DRB2, DRB6, DRB7, DRB8, and DRB9, all of
which are pseudogenes and are not expressed. All haplotypes carry the DRB9 pseudogene,
but the number of other pseudogenes carried on each varies depending on the DRB1
family. The different haplotypes (including pseudogenes) are illustrated in Figure 6.3 and
details of some DRB gene family nomenclature is given in Table 6.2.
Splits
These developments also helped to identify what were previously referred to as “splits”
(more correctly “isoforms”). This is demonstrated in Figure 6.4. For example, the HLA
B5 antigen was found to have three variants; these were first called Bw51, Bw52, and
Bw53. The “w” in the name was dropped when their status was confirmed, leaving the
labels B51, B52, and B53. These are now more correctly identified as B*51, B*52, and
B*53. When they were discovered these three variants were named according to the next
available unused number in the catalogue for HLA A and B antigens. Each of these vari-
ants has multiple subdivisions. However, not all new allele assignments were bone fide,
some turned out to be incorrect and some alleles are null alleles. When an allele assign-
ment is incorrect, the number is deleted from the official sequence record and the number
is not reused. Consequently there are gaps in the allele sequences numbers, e.g. there
is no B*51:25. In addition, the level of variation does not stop after four figures. Some
alleles carry non-synonymous polymorphisms, which can lead to further splits and names
like B*51:01:01. In fact, there are seven members in the B*51:01 family (B*51:01:01 to
B*51:01:07). Though these are all members of the B5 family, not all of the alleles were
detectable by serological typing and quite a few B5s identified by serology were incor-
rectly assigned as members of cross-reactive groups. DNA genotyping revolutionized HLA

DRB1 DRB6 DRB9

DRB1 DRB6 DRB5 DRB9

DRB1 DRB2 DRB3 DRB9

DRB1 DRB7 DRB8 DRB4 DRB9

DRB1 DRB9

Figure 6.3: Organization of the HLA DRB genes. Each haplotype may carry two or more DRB genes. However,
a significant number of the DRB genes are pseudogenes (DRB2, DRB6, DRB7, DRB8, and DRB9; shown in gray)
and are not expressed.

2967_Ch06.indd 182 07/07/2015 11:57

 Histocompatibility 183

Table 6.2 Commonly used DRB allele family names past and present.

Current equivalent allele

Old name families or genes Gene or haplotype
DR2 DRB1*15 or DRB1*16 DRB1
DR3 DRB1*03 DRB1
DR4 DRB1*04 DRB1
DR5 DRB1*11 or DRB1*12 DRB1
DR6 DRB1*13 or DRB1*14 DRB1
DRw52 DRB3 DRB3
DRw53 DRB4 DRB4
Ancestral haplotypes
A1–Cw7–B8–DR3–DQ2 8.1
A3–Cw7–B7–DR2 (DRB1*15)–DQ1 (DQB1*06:02) 7.2

Names are used in this chapter as they have been historically and as they are currently; the evolution of HLA
and the change in nomenclature are explained in some detail. To make reading easier, the simplest names have
been used where possible, i.e. DR3 in place of DRB1*03:01; however, it is not always appropriate to make this
exchange, especially when referring to molecular structures. For the ancestral haplotypes, a mixture of old and
new nomenclature is used depending on published history and the nature of the studies performed. For example,
B8, which is old nomenclature, is often used. This has been applied because although this example should be
correctly labeled HLA-B*08, old studies did not use this naming system. In contrast most studies of DQ used the
current nomenclature, e.g. DQB1*06:02.

typing to such an extent that the number of detectable alleles and variants (splits) cata-
pulted from around 100 A and B antigens to several thousand alleles at each locus. The
same process had the same effect on the HLA class II allele groups.
Taken altogether, i.e. identifying alleles on the new genes and the improved quality of
typing (illustrated in Tables 6.3 and 6.4), which also enabled a greater number of v­ ariants
to be accurately identified, these developments have been astounding. HLA typing is now
well and truly in the molecular genetics era and nomenclature has been updated too.

The current naming system for HLA alleles and genes allows for a
greater level of resolution to be reported
As HLA was at first thought to be a single locus system, antigens were first numbered
sequentially A1, A2, etc., as they were discovered and approved. However, it soon became
clear that there were at least two HLA loci: A and B. On close inspection when the antigen
sets for the two loci (A and B) were pulled apart, it was discovered that there were quite
a few B antigens that had A antigen labels. It was decided that for numbering purposes,
A and B would share the same number set. Thus, no A antigen has the same number
sequence as a B antigen and vice versa; the numbers assigned for A and B antigens are
exclusive. Therefore, there are HLA A1, A2 and A3 antigens, but no B1, B2, and B3 anti-
gens. This tradition has persisted. This way of naming new antigens and alleles has been
subjected to a standard practice since the 1965 IHTW. It is also important to know that all
B, and some A, antigens have a sequence variation that labels them as belonging to either
the Bw4 or Bw6 family. This is illustrated in Table 6.5.

2967_Ch06.indd 183 07/07/2015 11:57

 184 CHAPTER 6 The Major Histocompatibility Complex

Antigen
family
HLA-B5

Split Split Split

B51 B52 B53

B*51:01 B*52:01 B*53:01

to to to
B*51:42 B*52:09 B*53:11

Figure 6.4: HLA nomenclature tree: splits. The figure illustrates the development of HLA nomenclature based
on one member of the HLA-B antigen family. B5 was identified early on in studies of HLA (as indicated by the low
number B5). Three variants of B5 were later identified. These variants were often called “splits” in tissue-typing
laboratories and because at that time the numbering sequence for the combination of HLA A and B antigens had
reached 50, the next numbers to be assigned were given to these three variants, i.e. B51, B52, and B53. As these
antigens were first identified by serotyping, the original names reflect this naming, i.e. the phenotype (e.g. B51)
not the genotype (e.g. B*51). In the molecular era further variation has been identified for each of these antigens
with allele labels suggesting more than 40 B51 (B*51:01 to B*51:42) variants and approximately 10 for both B52
(B*52:01 to B*52:09) and B53 (B*53*:01 to B*53:11).

In the molecular era we are dealing with alleles and genotypes. The alleles identified are named
and numbered in a similar way to the antigens, but more information can be provided. For
example, DR alleles of the DRB1 family are referred to as DRB1*01 and a further :01 can
be added to indicate a higher level of specificity (or resolution). The allele thus labeled is
DRB1*01:01. The use of DRB1 indicates that this is the product of the DRB1 locus and not
of the other expressed DRB loci (DRB3, DRB4, and DRB5). Considering the large number of
HLA alleles so far identified it is important to be as precise as possible when naming alleles or
allele families. Table 6.6 illustrates the correct use of the current naming techniques. When
looking at this we need to consider whether studies are being performed at low or high resolu-
tion, or somewhere in the middle. The term family can be useful when genotyping is being
performed at low resolution. Current nomenclature is frequently updated and can be found at
http://hla.alleles.org/antigens/index.html or http://www.ebi.ac.uk/imgt/hla/stats.html.

The MHC encodes a cornucopia of genetic diversity within the

HLA genes
The importance of having a consistent method for assigning allele names is clearly illus-
trated when we consider the degree of genetic diversity at each of the HLA loci.
For example, in April 2015 there were 3107 HLA-A alleles, 3887 HLA-B alleles, 2623
HLA-C alleles, and 1726 HLA-DRB1 alleles identified. Regular updates on all of these
alleles (numbers and nomenclature, plus additional information) can be found at
http://www.ebi.ac.uk/imgt/hla/stats.html.

2967_Ch06.indd 184 07/07/2015 11:57

 Histocompatibility 185

Table 6.3 Analysis of the correlation between serological and RFLP DR typing in 1000 individuals.

True True False False Concordance

Allotype N positive negative positive negative Equivocal (%)
DR1 155 132 3 11 4 5 90
DR2 295 244 15 13 10 13 92
DR3 444 358 71 6 2 7 97
DR4 326 278 17 15 13 3 91
DR5 122 109 1 4 5 3 93
DR6 208 146 5 4 41 12 78
DR7 357 311 31 9 4 2 96
DR8 37 23 0 5 8 1 69
DR9 15 9 1 4 1 0 66
DR10 6 2 0 1 3 0 33
DRBr 35 – 35 – – – –
Total count 2000 1612 179 72 91 41 91.8

This table shows the correlation between two different methods of HLA DR typing: serological phenotyping versus
RFLP genotyping. In this model, the analysis shown in the table assumes the RFLP technique always correctly
identifies the antigen/genotype. The earlier method is of lower quality due to the poor quality of antibodies for
testing the less common DR antigens DR8, DR9, and DR10, and also the inability to accurately detect DR6, which
was often unassigned especially in heterozygotes. True negative indicates true homozygotes.

The level of genetic polymorphism in the HLA genes is illustrated in Figure 6.5. However,
the level of variation in the expressed gene is not always equal to the number of protein
polymorphisms in the coding regions of the gene. For example, the DNA sequence may
change, but the polypeptide composition may remain unchanged. This is because there is
redundancy within the genetic code so that a single amino acid may be encoded by more
than one DNA triplet (e.g. CCC, CCA, and CCG each encode the amino acid proline).
Thus, the numbers of proteins encoded by the three HLA class I loci (A, B, and C ) are
2185, 2870, and 1850, respectively. Each of these values is lower than the figures given
for the number of alleles. Some of the alleles are null alleles with no expressed product; at
present there are 147 null HLA-As listed, 124 HLA-Bs, and 86 HLA-Cs.

Comparing the levels of genetic diversity at DR with those at DQ can

make DQ look like a poor relation
Based on the level of variation at the DRB locus DRB1 (1726 alleles) compared with
DQB1 (780 alleles), DQB1 looks like a poor relation. This is misleading, however, because
the DRA gene has only seven alleles and encodes only two protein variants. Generally,
DRA is considered to be essentially invariant (though this is not actually correct). This lack
of variability in the α-chain of the DR molecule means that almost all of the polymor-
phism is restricted to the DRβ-chain. This means that the 1726 alleles that are involved
in producing 1262 protein variants in the DRB1 gene encode almost all of the molecular
variation in the complete expressed DR molecule. However, both DQA and DQB are
widely polymorphic (54 and 780 alleles, respectively) and so there is a greater possibility

2967_Ch06.indd 185 07/07/2015 11:57

 186 CHAPTER 6 The Major Histocompatibility Complex

Table 6.4 Analysis of the correlation between RFLP and PCR oligonucleotide probing for HLA-DRB1
genotyping in 395 individuals.

True True False False Concordance

Allotype N positive negative positive negative Equivocal (%)
DR1 72 67 2 2 1 0 93
DR15 90 84 3 3 0 0 97
DR16 7 7 0 0 0 0 100
DR3-a 120 89 23 8 0 0 93
DR3-b 10 7 3 0 0 0 100
DR4 188 175 10 2 1 0 98
DR11 48 43 1 2 1 1 92
DR12 7 4 0 3 0 0 57
DR6-a 53 45 2 4 2 0 89
DR6-b 44 37 3 4 0 0 91
DR7 115 104 7 3 0 0 97
DR8 16 16 0 0 0 0 100
DR9 7 5 0 1 1 0 71
DR10 6 4 0 0 2 0 66
DRBr 10 10 0 0 0 0 100
Total 790 694 51 29 11 2 95
count

This table shows the correlation between two different methods of HLA-DRB1 genotyping: RFLP genotyping
versus PCR-oligoprobing using 24 oligoprobes. PCR clearly identifies 18 DRB1 allele families and this example
includes the rarely discussed DRBr allotype (which is in fact one of the DRB1*01 family members). In this model,
the analysis shown in the table assumes the PCR technique always correctly identifies the antigen/genotype.

Table 6.5 Sequence differences between HLA Bw4 and Bw6.

Group 77 78 79 80 81 82 83
Bw6 S L R N L R G
Bw4 D L R T L L R
Bw4 S L R T L L R
Bw4 S L R I A L R
Bw4 N L R I A L R
Bw4 N L R T A L R

The sequences of the Bw4 and Bw6 epitopes vary at positions 77 to 83 of the second exon of the α1 domain on
the HLA class I molecule. The critical amino acid difference between Bw4 and Bw6 appears to be a glycine (G) for
arginine (R) exchange at position 83. Variations at the other positions in this amino acid sequence define the five
Bw4 genotypes. Note that there is no variation at positions 78 and 79, which all carry leucine (L) and arginine (R),
respectively.

2967_Ch06.indd 186 07/07/2015 11:57

 Histocompatibility 187

Table 6.6 The current system for naming HLA genes and alleles.

Name Explanation
HLA-DRB1 This indicates a specific HLA locus (DRB1)
HLA-DRB1*01 This indicates a group or family of alleles that encode the DR1 antigen
HLA-DRB1*01:01 This indicates a specific allele from the DRB1*01 family (many scripts
use DRB1*0101 for this and omit the “:” sign)
HLA-DRB1*01:01:02 This indicates an allele that encodes a synonymous mutation of
DRB1*01:01
HLA- This indicates an allele that encodes a mutation outside the coding
DRB1*01:01:01:02 region of DRB1*01:01
HLA-DRB1*07:10N This indicates a “null” allele (i.e. an allele that is not expressed)
HLA-DRB1*01:01L This indicates an allele encoding a protein with significantly reduced
cell surface expression

for formation of a wider range of variation in the expressed DQA molecules than initially
indicated from the simple allele frequencies. In addition because every individual inherits
two haplotypes—one from each parent—DQA1 alleles of paternal origin can form func-
tional DQ molecules with DQB1 alleles of maternal origin (and vice versa), thus adding a
further level of diversity for DQ molecules. The same is also true for HLA DP.

HLA class II molecules can be expressed in trans or in cis

For the most part HLA alleles are co-dominantly expressed, meaning that most individu-
als have the opportunity to express both paternal and maternal antigens. The possibility

4000

3500

A
3000
C
2500 B
Polymorphisms

MICA
2000 TNF
C2/C4
1500
DRB1
DQA
1000
DQB
500 DPA
DPB
0
A MICA DRB1 DPA

Figure 6.5: Frequency of HLA polymorphisms. The figure shows the extent of polymorphism in HLA as of
April 2015. The very high frequency of HLA-B polymorphism gives a false impression regarding polymorphism for
some of the other genes listed. There are 101 MICA alleles, for example. One hundred and one is a considerable
level of polymorphism, it just looks small compared to the 3887 B alleles represented. The order of the genes is
indicated on the right-hand descending column.

2967_Ch06.indd 187 07/07/2015 11:57

 188 CHAPTER 6 The Major Histocompatibility Complex

Case A
DQB1*02:01 DQA1*05:01 DRB1*03

Encodes the DQ2

molecule in cis

Case B
DQB1*02:02 DQA1*02 DRB1*07:01

DQB1*03:01 DQA1*05:01 DRB1*11

Encodes the DQ2

molecule in trans

Figure 6.6: The HLA DQ2 molecule constructed in cis and trans. The figure shows the HLA molecule being
constructed in cis and trans. The cis construction (case A) is illustrated by full gray arrows with each allele coming
from a single haplotype from the same parent where both the α-chain and the β-chain of the final molecule are
encoded on the DRB1*03 haplotype by the alleles DQA1*05:01 and DQB1*02. The trans construction (case B) is
illustrated by the dotted arrows where more than one parental haplotype contributes a polypeptide chain to form
the final molecule. In each case, the DQA1 locus contributes the DQA1*05:01 allele that encodes the necessary
α-chain and the DQB1 locus contributes the DQB1*02 allele that encodes the necessary β-chain. The ability of the
DQ genes to construct molecules in trans is important as it increases the number of potential genetic variations
there are at the molecular level. HLA DR is less able to do this as the DRA1 gene is essentially monomorphic with
very little variation.

to increase the genetic variation through trans expression as opposed to the normal cis
pattern can have important personal consequences (Figure 6.6). Sollid et al. found that
patients with celiac disease have an increased frequency of HLA DR3 and DR7. The
genetic association with DR3 was well known, but the association with DR7 was less
well reported. Further analysis revealed that DR7 is only associated with celiac disease
in the presence of either DR3 or DR5. DR3 is always found with DQ2 (DQA1*05:01–
DQB1*02:01; sometimes called DQ2.5). DR5 is almost always found with DQ7
(DQA1*05:–DQB1*03) and DR7 occurs with either DQ9 (DQA1*01:02–DQB1*03)
or DQ2 (DQA1*01:02–DQB1*02:02). The DQA1 alleles of the DR3–DQ2 and DR5–
DQ7 haplotypes are almost identical except for codon 135 in the membrane proximal
domain of the DQ chain. DR5/DR7 heterozygotes with the DR7–DQ2 haplotype carry
similar DQA1 and DQB1 genes encoding alleles with essentially the same sequences in
the trans position as those found on the DR3– DQ2 haplotype in cis. The ability to form
DQA/DQB in trans may explain the complex HLA association with celiac disease.

The final groups of genes that need to be considered are those called
pseudogenes, gene fragments, and null alleles
Pseudogenes are genes that are not expressed. DRB pseudogenes have been discussed above,
but there are also HLA class I pseudogenes (see Table 6.7). Gene fragments are exactly as
described and are mostly fragments of once expressed genes, possibly the remains of our
evolutionary past. Null alleles are generally considered to be alleles that are not expressed.
In some cases there is doubt about expression: alleles may be found to be expressed in the

2967_Ch06.indd 188 07/07/2015 11:57

 The Extended Human MHC MAP 189

Table 6.7 A list of HLA other class I genes not discussed in the text.

Current names Characteristic

HLA-E; HLA-F; HLA-G Associated with class I HindIII fragments
HLA-J; HLA-K; HLA-L Class I pseudogenes associated with HindIII fragments
HLA-N; HLA-P; HLA-S; HLA-T; Class gene fragments associated with HindIII
HLA-U; HLA-V; HLA-W
HLA-X Class I gene fragment
HLA-Y Class I pseudogene
HLA-Z Class I gene fragment located in class II region
HLA-H Class I pseudogene

Pseudogenes are genes that are not expressed. Others include the HLA class I pseudogenes (genes that do not
encode functional products), gene fragments (fragments of once expressed genes) which are possibly part of our
evolutionary past, and null alleles (alleles that are not expressed). There is some overlap in the use of these terms.

cell only and not at the cell surface or may be soluble, but not found attached to the cell
surface. More information on this is given at http://hla.alleles.org/antigens/index.html.

6.2 The Extended Human MHC MAP

The MHC was extended in 2004 to include a region telomeric of HLA-A as far as the
hemochromatosis gene HFE. The details of the extended gene map (xMHC) can be found
in Horton et al. (2004). Though the map includes 252 expressed genes, only a select few
are appropriate for this book (Figure 6.7). It also includes 139 pseudogenes (i.e. genes
whose products are not expressed) and a number of gene fragments. A list of HLA genes
is given in Tables 6.7 and 6.8.
For historical reasons, the MHC can be divided into three subregions referred to as MHC
class I, MHC class II, and MHC class III. These regions were originally defined on the

DPB1/DPA1 DQB1/DQA1 DRB1 DRB3, 4, 5 DRA1

3300 2900 2700

C2/C4/Bf TNF MICB/MICA B C A

1500 300

Figure 6.7: Gene map of HLA in MHC (showing selected expressed genes only). The figure is not to scale.
The boxed figures and gray arrows mark the distances between loci. This figure covers the old 3.6-Mb pre-xMHC.
Whether a haplotype carries another expressed DRB gene (DRB3, 4, or 5) depends on the DRB1 family encoded
on the haplotype (see Figure 6.3).

2967_Ch06.indd 189 07/07/2015 11:57

 190 CHAPTER 6 The Major Histocompatibility Complex

Table 6.8 A list of HLA class II genes.

Current names Characteristic

HLA-DRA DRα-chain
HLA-DRB1 DRβ-chain encodes DR1–DR16 allele families
HLA-DRB2 Pseudogene
HLA-DRB3 Second expressed DRβ-chain found on haplotypes carrying DR3, DR11,
DR12, DR13, and DR14 allelic families in Northern European populations
HLA-DRB4 Second expressed DRβ-chain found on haplotypes carrying DR4, DR7, and
DR9 allelic families in Northern European populations
HLA-DRB5 Second expressed DRβ-chain found on haplotypes carrying DR15 and DR16
allelic families in Northern European populations
HLA-DRB6 DRB pseudogene found on DR1, DR2, and DR10 haplotypes
HLA-DRB7 DRB pseudogene found on DR4, DR7, and DR9 haplotypes
HLA-DRB8 DRB pseudogene found on DR4, DR7, and DR9 haplotypes
HLA-DRB9 DRB pseudogene isolated fragment
HLA-DQA1 DQα-chain
HLA-DQB1 DQβ-chain
HLA-DQA2 DQA-related chain not expressed
HLA-DQB2 DQB-related chain not expressed
HLA-DQB3 DQB-related chain not expressed
HLA-DOA DOα-chain
HLA-DOB DOβ-chain
HLA-DMA DMα-chain
HLA-DMB DMβ-chain
HLA-DPA1 DPα-chain
HLA-DPB1 DPβ-chain
HLA-DPA2 DQα-chain-related pseudogene
HLA-DPB2 DPβ-chain-related pseudogene
HLA-DPA3 DPα-chain-related pseudogene

basis that the genes within them encoded proteins with similar structures and functions.
This terminology is no longer fit for purpose; all three regions encode a variety of proteins
with variable structures and functions, including many none HLA antigens.
Of those genes listed in Figure 6.5, Tables 6.7 and 6.8, the most studied in this region have
been the classical HLA genes HLA-A, -B, -C, -DRB, -DQA, -DQB, -DPA, and -DPB. In
addition to these eight loci, some studies have also investigated a selection of other MHC
genes. These include a group of genes encoding proteins of the classical and alternative
complement pathways (C2, C4A, C4B, and Bf ), tumor necrosis factor (TNF)-α (TNFA),
and, more recently, the MHC class I chain (MIC)-like proteins (MICA and MICB).

2967_Ch06.indd 190 07/07/2015 11:57

 Molecular Structure of HLA Class I and Class II 191

6.3 Molecular Structure of HLA Class I

and Class II
A great deal can be learned about genetic susceptibility from looking at HLA antigen struc-
ture. As we will see in the section below, HLA plays a central role in the immune response.
Structurally, HLA class I and class II antigens are different. The class I antigens have a sin-
gle chain composed of three immunoglobulin domains supported by β-2-microglobulin
(encoded on chromosome 15). All of the variance in the molecule is encoded on the HLA
chain. The HLA class II molecules, however, are composed of two chains—an α-chain and
a β-chain—both encoded on chromosome 6p21.3 and both with some (though limited in
the case of DRA) polymorphism.

X-ray crystallography of HLA-A2 revealed the full structure

and much about the function of HLA class I
HLA class I molecules on the cell surface are composed of three immunoglobulin domains
(the HLA α-chain: α1, α2, and α3) and a fourth domain provided by β-2-microglobulin
(encoded on chromosome 15). In addition, there is a short transmembrane section that
anchors the expressed molecule to the cell surface. The immunoglobulin domains are
numbered from the furthest out to the nearest in (α1 to α3). The α3 domain and the β-2-
microglobulin form a supporting structure on which the two remaining immunoglobulin
domains are mounted (Figure 6.8). These two domains (α1 and α2) form a complex
structure with a series of β-pleated sheets and two α-helices (one each). Essentially, the
β-pleated sheet forms the floor and the two opposing α-helices form the walls of a groove.
X-ray crystallography shows that the groove is a closed structure of approximately 11 Å
(1.1 nm) in length. Within the groove there are nine pockets for the binding of anti-
genic side chains (Figure 6.9). The publication in 1987 of the HLA-A2 crystal structure
revealed the functional element of the HLA molecule to be the closed groove. This site is
known to be the site at which antigenic peptides are bound and presented to the T cell
receptor (TCR) of CD8+ T cells—a crucial process in adaptive immunity. Furthermore,
the antigens presented by HLA class I molecules to T cells are small peptides of no more
than eight or nine amino acids. Genetic comparison between A2 and other HLA mol-
ecules revealed that the majority of allelic variation encodes amino acid changes in and
around the antigen-binding groove. This provides a functional link to genetic diversity in
HLA class I, and may provide an explanation for the very large number of genetic associa-
tions between MHC and disease.

Antigen-
binding Figure 6.8: HLA class I molecule structure. The figure
α2 groove α1 represents a simple diagram of the HLA class I molecule
with three immunoglobulin domains (α1, α2 and α3)
encoded by the HLA class I allele and a β-2-microglobulin
molecule (encoded on chromosome 15). The α3 unit and
α3 β-2-micro- β-2-microglobulin unit (in gray shading) form a support
globulin structure for the α1 and α2 globulin domains, which create
a closed groove for the binding of antigenic peptides (the
antigen-binding groove). In the case of class I molecules
the antigenic peptides are usually eight to nine amino
acids in length. The majority of HLA polymorphisms
encode amino acid variations in and around the antigen-
binding grove.

2967_Ch06.indd 191 07/07/2015 11:57

 192 CHAPTER 6 The Major Histocompatibility Complex

Figure 6.9: Top-down view of the HLA class I binding groove showing some of the sites at which different
alleles encode amino acid variations. Some of the sites for amino acid variation are illustrated by dots. The
figure shows the β-pleated sheets (in gray) which form the floor and two opposing α-helices (in black) which form
the walls of the closed antigen-binding groove of the molecule. For HLA class I, short peptide antigens of around
eight to nine amino acids in length are bound in this groove and presented to the T cell receptor in the formation
of the immune synapse (Figure 6.11)—a key process in adaptive immunity. HLA class II molecules have a similar
structure but the groove ends are open, allowing for longer peptides to be bound. (Adapted from Parham P
[2009] The Immune System, 3rd ed. Garland Science.)

The X-ray crystallography structure of HLA class II structure revealed

the critical difference between class I and class II
The first X-ray crystallography studies of HLA class II antigens were published in 1991.
The critical difference between HLA class II and HLA class I is that the class I heterodimer
is formed by the interaction of the class I α-chain with a β-2-microglobulin molecule. The
result is a molecule with four domains as described above, but only the α-chain is polymor-
phic and encoded in the MHC. The class II heterodimer also has four domains all encoded
on chromosome 6p21.3, and both α- and β-chains are polymorphic (though DRA poly-
morphism is limited). Two of the class II domains are provided by the A gene (α1 and α2
in each gene DRA, DQA, or DPA) and two are provided by the B gene (β1 and β2 in each
gene DRB, DQB, or DPB). In each case, one of the two domains provides structural sup-
port for the other, and both the α-chain and β-chain have a short transmembrane tail that
anchors them to the cell surface (Figure 6.10). The four-domain structure of the complete
class II molecule is very like that of class I. Once again the lower domains (α2 and β2) sup-
port the higher domains (α1 and β1). The upper portion of the molecule presents as a series
of β-pleated sheets and opposing α-helices which form the floor and walls of the antigen-
binding groove. Unlike class I, the class II antigen-binding groove is an open-ended structure
that enables larger polypeptides of approximately 13–22 amino acids to be bound.
In agreement with findings for HLA class I structural analysis revealed the majority of
HLA class II polymorphisms are also encoded in and around the antigen-binding groove.
Once again, this provided a basis for linking genotype to phenotype.

2967_Ch06.indd 192 07/07/2015 11:57

 Immune Function of HLA Class I and Class II 193

Antigen-
binding
β1 groove α1

β2 α2 Figure 6.10: HLA class II molecule structure. The figure

represents a simple diagram of the HLA class II molecule
with four immunoglobulin domains. These are α1 and α2
encoded by the HLA class II A gene and β1 and β2 encoded
by the class II B gene. As with class I molecules, two of the
units form a support structure (α2 and β2 in gray), while the
other two (α1 and β1) form the antigen-binding groove.

6.4 Immune Function of HLA Class I and Class II

The X-ray crystallography studies on HLA class I and class II indicate significant differ-
ences between the molecules. These differences are likely to have significant functional
importance and may be important in helping us to understand the different HLA associa-
tions in complex disease.

Class I molecules have distinct features

Class I molecules are present on all nucleated cells and encode a site for binding the CD8
protein on T cells. Class I molecules present short antigenic peptides to the TCR, and on
presentation there is co-binding between the class I molecule and CD8. This HLA peptide–
TCR interaction is often referred to as the formation of the immune synapse (Figure 6.11).
This interaction activates CD8+ T cells, which are crucial in adaptive immunity.
Class I antigens are not only involved in antigen presentation and adaptive immunity.
They may also be involved in cell regulation in innate immunity. Natural killer (NK) cells
use a variety of inhibitory receptors to detect alterations in the expression of HLA class I
molecules. CD94 (or NKG2A) is an inhibitory receptor that monitors overall expression
of HLA-A, HLA-B, and HLA-C proteins on human cells. This receptor uses the HLA-E
system molecule, which expresses bound segments of HLA-A, -B, and -C, to consider the
MHC status of the cell.
Unlike NKG2A, which is not MHC class I specific, the other form of inhibitory receptor
expressed on NK cells, i.e. the killer cell immunoglobulin-like receptor (KIR) is highly
selective and each NK cell must express at least one KIR that interacts with self HLA class
I for normal functioning to occur. The normal function of the KIR HLA interaction is to
monitor HLA class I function. Viral pathogens will often disrupt normal HLA expression
as a method of evading the host’s immune response, especially that of the CD8+ T cells.
Interestingly, it is thought that HLA-C is primarily involved in this process.

HLA class II is different to class I

HLA class II antigens are expressed on a selected variety of cells often referred to as antigen-
presenting cells (APCs), which include macrophages, dendritic cells, and B cells. Class II
presents antigenic peptides to CD4+ T cells and has a site for CD4 binding. The class II
molecule has an open-ended binding groove that can accommodate longer peptide antigens.

2967_Ch06.indd 193 07/07/2015 11:57

 194 CHAPTER 6 The Major Histocompatibility Complex

T cell

Figure 6.11: The immune synapse. The immune synapse is formed

from the interaction of three key elements: the antigenic peptide (gray
circle), the HLA molecule (in white), and the T cell receptor (TCR in
APC dark gray). This figure could represent presentation by HLA class I or
class II to CD8+ or CD4+ T cells, respectively.

Class II does not appear to be involved in NK cell activity. Furthermore, the mechanisms
by which HLA class I and class II engage and present polypeptides are different. HLA class
I presents antigenic polypeptides from within the cell (i.e. intracellular), whereas HLA class
II presents antigenic peptides that are from outside the cell (i.e. intercellular). The mecha-
nisms by which this occurs are complex and are illustrated in Figure 6.12.

HLA class I and class II have important similarities

There are several important similarities between HLA class I and class II. Both class I
and class II antigens have binding grooves with pockets for the binding of antigenic side
chains. In both cases the majority of HLA polymorphism encodes variation in and around
the floor and walls of the binding groove, and in both cases this genetic variation can be
functionally significant. The functional consequences of genetic variation in HLA alleles
are attributable to changes in the amino acid sequence leading to variation in electrostatic
properties of the binding groove, especially in and around the pockets for the binding of
antigenic side chains.

HLA class I and antigen engagement in the cell is different from

HLA class II
HLA molecules are rarely (if ever) presented on the cell surface without an occupied
antigen-binding groove. Binding of intracellular peptides inside the cell is a competitive
process. Peptides may be derived from normal cellular proteins broken down in the pro-
teasome or from pathogenic proteins (antigenic peptides). These antigenic peptides are
also generated by the proteasome (a large barrel-shaped complex found in the cytosol
responsible for digesting, degrading, and recycling proteins). The peptides that result from
this digestion are transported through the cytosol to the endoplasmic reticulum (ER)
with the help of a specialized transporter molecule TAP. HLA class I antigens are con-
stantly being produced in the cell and newly synthesized molecules are translocated to the
ER where they are stabilized by binding to β-2-microglobulin. Finally, these newly syn-
thesized HLA class I molecules bind one of the peptides delivered to the ER by TAP. This
is a competitive process and is governed by the peptide-loading complex. If the loaded
peptide is too long at the N-terminus, the ER amino-peptidase will remove excess amino
acids from the amino-terminal end to ensure a better fit in the antigen-binding groove.
Once the peptide has been bound, and if necessary trimmed, the completed molecule is
exported from the ER in a membrane-enclosed vesicle through the Golgi stacks to the
plasma membrane. The processes described above are continuous. Most of the peptides
presented are self-antigens, in particular HLA class I that appears to be generated in large
quantities in almost all tissues.

2967_Ch06.indd 194 07/07/2015 11:57

 Immune Function of HLA Class I and Class II 195

HLA class II HLA class I

ER
Extracellular Cytosol Intracellular
antigen antigen
Endocytic
vesicle
Proteasome
Peptide production Antigen processing
in phagolysosome Nucleus to peptides in
proteasome

Peptide binding
by HLA class II Peptide transport
into endoplasmic
reticulum (ER)

Peptide binding
HLA class II by HLA class I
in vesicle

HLA class II presents

peptide at cell surface
HLA class I presents
Golgi
peptide at cell surface

Figure 6.12: Antigen processing by HLA class I and II antigens. The right-hand side illustrates antigen
processing by HLA class I. Intracellular antigens from the cytosol, including self-antigens and those of pathogens
such as bacteria and viruses, are processed through the proteasome and transported through the endoplasmic
reticulum (ER) to the Golgi apparatus. The cell is constantly producing new HLA molecules. Antigenic peptides
contact these HLA molecules in the ER and the resulting HLA–peptide complex is transported through the
Golgi apparatus to the cell surface. On the left-hand side peptides from outside the cell (extracellular peptides)
are taken up by endocytosis and phagocytosis within endocytic vesicles into the cell. The figure illustrates a
macrophage. In this situation proteases in the vesicles break down captured proteins and they are bound to
newly synthesized class II molecules in the ER and transported through the Golgi apparatus to the cell surface.
(From Parham P [2009] The Immune System, 3rd ed. Garland Science.)

HLA class II and antigen engagement in the cell is different

Unlike HLA class I, HLA class II presents peptides derived from proteins imported from
outside the cell. The proteins are captured by endocytosis or phagocytosis and brought
into the cytosol in a vesicle. Proteases break down the captured proteins within these
vesicles and produce peptides. At the same time newly synthesized HLA class II molecules
coupled with the invariant chain are exported into the cytosol in vesicles. The invariant
chain acts as a molecular chaperone preventing premature binding of newly synthesized
HLA class II to peptides in the ER. Only when the vesicle interacts with another vesicle
bearing peptides is the invariant chain released through protease cleavage. This permits
peptides to compete for binding with the HLA class II molecule. Once this process has
occurred, the HLA class II–peptide molecule is transported to the cell surface for antigen
presentation to the TCR.

2967_Ch06.indd 195 07/07/2015 11:57

 196 CHAPTER 6 The Major Histocompatibility Complex

HLA class I, on one hand, and HLA class II, on the other hand, provide two of the
essential mechanisms of immune surveillance and induce immune responses to infec-
tious agents within the cell and outside the cell. These processes frequently overlap, and
are one of the main forces in acquired immunity and immune surveillance, which is
essential for survival.

6.5 HLA class I and Disease

As stated in the Introduction to this chapter, it is not possible to consider all diseases
here and for that reason a limited few have been chosen. For HLA class I, one disease per
gene locus was selected with the exception of HLA-B, where there is adequate coverage
elsewhere in the book, especially with reference to ankylosing spondylitis and also in
Chapter 8 on pharmacogenetics.

Hemochromatosis is an example of a Mendelian disease which maps

within the xMHC
Hemochromatosis has already been discussed in earlier chapters. This disease is a
Mendelian autosomal recessive disease with linkage to the MHC. The precise location
of the hemochromatosis gene HFE was identified in 1997 by Feder et  al.. Early stud-
ies of Northern European populations found linkage with HLA-A and association studies
identified the HLA-A3 family of alleles as the strongest HLA association with hemo-
chromatosis, but also implicated the extended haplotype HLA 7.2 that carries A*03–
C*07:02–B*07–DRB1*15:01–DQA1*01:02–DQB1*06:02 in disease susceptibility. Fine
mapping around the MHC finally identified HFE at a distance of approximately 3Mb
telomeric to the HLA-A locus indicating the extreme linkage disequilibrium found in
some but not all MHC haplotypes. One explanation for this extreme linkage disequilib-
rium is the lower than average rate of recombination within the MHC, which has been
calculated as 0.49 cM/Mb compared with 0.92 cM/Mb for the whole genome. This obser-
vation has given rise to the suggestion the MHC contains a greater number of recombina-
tion cold spots and a low number of hot spots. Overall the genetic association with HFE
illustrates the idea that not all HLA-associated diseases are genetically complex diseases;
some may be Mendelian.
However, this is not the whole story of hemochromatosis. In populations outside of
Northern Europe, disease-causing mutations in other (non-MHC) genes have been pro-
posed and penetrance for HFE homozygotes, even in populations with high levels of
Northern European ancestry, has been shown to be very low, suggesting a more complex
model than originally described.

Psoriasis proves the point that HLA-C is an important locus to

consider in genetic studies of the MHC
HLA-C has often been overlooked by researchers. The reason for this is that in the early
days HLA C antigens were difficult to detect and type. Even when molecular genotyp-
ing was introduced HLA-C remained difficult to genotype. As a result many studies have
failed to include HLA-C, potentially giving the false impression that the C locus is not
associated with disease. However, the history of HLA and psoriasis illustrates the impor-
tance of considering HLA-C.

2967_Ch06.indd 196 07/07/2015 11:57

 HLA Class II and Disease 197

Type I versus type II psoriasis

Clinically psoriasis occurs in two forms. Type I psoriasis is an early-onset disease (age 15–25
years) and type II psoriasis is an older-onset disease (approximately age 60 years). Type I psori-
asis is often familial, whereas type II psoriasis is not. Early studies of type I psoriasis identified
associations with HLA-B*13 and HLA-B*17 (later HLA-B*57). Studies in the class II region
found associations with the DRB1*07–DQA1*02–DQB1*03:03 haplotype. Altogether this
implicates one of two common DRB1*07 haplotypes. However, the association is not with
DRB1*07 itself; the strongest association is with HLA-C, specifically C*06:02 that is carried
on this haplotype. In contrast, studies on type II psoriasis have identified C*02 and B*27, and
have not reported associations with either DRB1*07 or C*06:02. This genotype difference
between early- and late-onset disease is important and frequently seen in complex disease.
Considering the pathogenesis of type I psoriasis from a functional perspective, it has been
proposed that HLA-C may have a direct role in this disease. A molecular model based on
alanine at position 73 has been proposed, but is not specific to HLA-C*06:02 and remains
controversial. It is, however, interesting that a member of the HLA-C family may be the
major MHC-encoded risk factor for psoriasis, a dermatological disease, considering the
role HLA-C plays in NK cell activation and innate immunity.

Before we leave HLA class I we need to consider Bw4 and Bw6

All HLA-B antigens and some HLA-A antigens carry either a Bw4 or Bw6 motif and his-
torically it was found that most HLA-B alleles carry the same Bw4 or Bw6 motif as that
carried by their family members, with a few exceptions. These exceptions were useful in
helping to confirm certain specificities where there were two or more alternative serotypes
(called splits). Which motif (Bw4 or Bw6) is expressed depends on the amino acid sequence
at positions 77, 80, 81, 82, and 83 of the molecule. Work by Arnett et al. suggests that
there may be more variation than previously recognized. For example, HLA-B*08:01 and
HLA-B*08:02 encode different motifs even though they are both members of the same
family (HLA-B*08). More importantly, further work also shows that the binding capacity
of these two alleles (HLA-B*08:01 and HLA-B*08:02) is quite different and that some of
these differences are due to the Bw motif. This difference would not be predicted from the
low-resolution genotyping of B*08 or from the observation that the two alleles are from
the same family but it does have potential functional consequence. This is important when
considering early studies and planning current studies because it may be necessary to con-
sider this when deciding on the level of resolution of genotyping to be applied in study.

6.6 HLA Class II and Disease

A number of examples have been selected for looking at HLA class II. Once again, we can-
not consider all diseases here. In this section we have decided to concentrate on two well-
known examples: cataplectic narcolepsy and multiple sclerosis, and three less well known
but interesting examples: AIH, PSC, and PBC. The latter three are all autoimmune liver
diseases that have different and overlapping features.

Severe or cataplectic narcolepsy has one of strongest HLA

associations ever reported
Among the neurological disorders, many have strong genetic associations with HLA,
including multiple sclerosis and myasthenia gravis. However, the strongest reported

2967_Ch06.indd 197 07/07/2015 11:57

 198 CHAPTER 6 The Major Histocompatibility Complex

association is that between severe cataplectic narcolepsy and the DRB1*15–DQB1*06:02

haplotype. Narcolepsy is a relatively rare condition characterized by abnormal regulation of
sleep. Symptoms may include hallucinations and sudden loss of muscle tone, which leads
to a sleep-like or unconscious state (cataplexy). Patients with severe narcolepsy suffer from
cataplexy following sudden and unexpected emotional stimulation such as fear or laughter.
In 1984, Langdon et al. reported that 100% of their cases were HLA-DR2 (DRB1*15)-
positive. This astonishing finding was very soon confirmed in several different popula-
tions, including studies in France and Japan, while studies in Germany and the Czech
Republic reported figures of 98% plus. Smaller studies based on African-Americans
reported considerably fewer patients had HLA-DR2 (approximately 70%). Further analy-
sis of African-Americans suggested that the true marker for HLA-encoded susceptibility
to narcolepsy may be DQ2. The DQ2 molecule, in this case, is constructed from the
products of the DQA1*01:02 and DQB1*06:02 alleles which encode the DQ α-chain
and β-chain, respectively. DQA1*01:02–DQB1*06:02 is carried on almost all DRB1*15
haplotypes in Northern Europeans and was found in the majority (27 out of 28 cases) of
the African-American HLA-DR15-negative cases.

There are different functional interpretations of the HLA association

with narcolepsy
Narcolepsy in dogs is caused by a single autosomal recessive gene in the immunoglobu-
lin heavy chain region. In canines, studies suggest that pathology is associated with the
immunoglobulin switch-like sequence and enhanced microglial expression at age 1–3
months. This causes axonal pruning and cell necrosis. In humans, familial narcolepsy
may occur in the absence of HLA-DRB1*15, suggesting there may be two forms of the
disease: familial (not associated with HLA) and sporadic (associated with HLA). Whether
narcolepsy is a single entity or not is unknown, but the DQ association may indicate that
this is an immune-mediated disease and this may help us understand the pathology of the
condition.
The DQA1*02:01–DQB1*06:02 haplotype that predisposes to narcolepsy is protective
of type 1 diabetes (T1D). T1D is in most patients, the consequence of autoimmune-
mediated destruction of the insulin-secreting pancreatic β cells. Studies have revealed that
susceptibility to and protection from T1D is determined by the amino acid at position 57
of the DQβ-chain. Aspartate 57 is protective whereas alleles encoding alanine, valine, or
serine predispose to T1D. The reverse pattern appears to be true for narcolepsy. In T1D,
a model based on position β57 has been proposed, whereby susceptibility and protection
are based on the ability to form a salt bridge between β57 and α79 in the expressed DQ
molecule. However, this model does not fit for narcolepsy; while DQB1*06:02 confers
susceptibility to narcolepsy, DQB1*06:01 that carries the same DQA1 allele confers pro-
tection. Comparing the polypeptide sequence encoded by the DQB1*06:02 allele with
that of DQB1*06:01 reveals nine amino acid differences, one of which is outside the pep-
tide-binding groove, while all the others are within the peptide-binding groove (of which
five are listed in Table 6.9). The listed amino acids are all in close proximity to a putative
T cell recognition surface or to a peptide-binding pocket within the groove.
There is also an important similarity here with rheumatoid arthritis as discussed in
Chapter 3, in relation to HLA-DRB1 and rheumatoid arthritis. The majority of genetic
variation in HLA encodes amino acid differences in and around the antigen-binding groove
of the expressed HLA molecule. These binding grooves each have up to nine pockets for
the binding of side chains on the antigenic peptide. These pockets are labeled P1–P9 in

2967_Ch06.indd 198 07/07/2015 11:57

 HLA Class II and Disease 199

Table 6.9 Narcolepsy amino acid positions for DQB1*06:01 versus DQB1*06:02.

Amino acid position

HLA class II allele 9 13 26 37 38
DQB1*06:01 L G L D V
DQB1*06:02 F L Y Y A

sequence. Depending on the associated allele, possession of a specific amino acid at a spe-
cific site can have profound effects on which antigenic peptides are bound and presented
to the TCR. In narcolepsy, the P4 and P9 pockets appear to have the most pronounced
effects. At P9 the critical change is with β37. At this position there is a tyrosine to aspartic
acid exchange that introduces significantly more negative charge into the P9 pocket and
this may subtly alter the anchor specificity. The P4 binding pocket is the largest pocket
in the expressed DQB1*06:02 structure. In contrast, the DQB1*06:01 has a smaller P4
pocket. This is due to polymorphisms at positions β13 and β26, which result in exchange
of glycine for alanine and leucine for tyrosine. The impact of these changes is considered
to be substantial.
Considering the function of HLA class II molecules and these molecular models, it is
plausible that narcolepsy is an immune-mediated disorder dependent on T cell responses.
At a molecular level, changes in the chemical structure influence the functional dynamics
of the antigen-binding groove in such a way that there may be expansion or contraction
of the range of peptides that may be preferentially bound and presented to the TCR.
In addition, subtle changes in the molecular structure of HLA molecules may influence
the orientation of the bound peptide and the recognition of the HLA–peptide complex
by the TCR. It is therefore interesting that three of the amino acid differences between
DQB1*06:01 and DQB1*06:02 are in the putative TCR recognition sites, suggesting that
both antigen binding (variation in the binding groove) and the TCR recognition site may
be important in determining HLA-encoded susceptibility to narcolepsy.
Finally, more recent studies in narcolepsy aimed at identifying immune genetic markers
in the Han Chinese population found several novel associations, including an association
with DQB1*03:01 and disease onset. The study also found that selected HLA-associated
markers differed significantly in cases where onset occurred after the recent influenza
pH1N1 pandemic. This study indicates three major points:
• The association in the Han Chinese population may be different to that in Northern
Europeans.
• The association may be related to age of onset or disease severity.
• The association may be related to other environmental factors.

Multiple sclerosis is a disease with a strong genetic association with

HLA class II
Multiple sclerosis is one of the most common neurological disorders and as many as one in
every 1000 individuals may have multiple sclerosis in Northern European populations. It
has been suggested that there are 400,000 cases in the USA alone. The disease presents in
young adults and is more common in women than in men (approximately twice as com-
mon). There is chronic degeneration of the myelin sheath on the nerve cells that leads to
the buildup of sclerotic plaques. Demyelination of the central nervous system leads to

2967_Ch06.indd 199 07/07/2015 11:57

 200 CHAPTER 6 The Major Histocompatibility Complex

motor weakness, impaired vision, and lack of coordination. It is a highly variable disease
and some sufferers experience slow progressive degeneration, while others have peaks and
troughs with periods of recovery in between periods of degeneration.
It is generally accepted that multiple sclerosis is a genetically complex disease involv-
ing both genes and the environment with a considerable genetic load associated with
specific MHC haplotypes, especially the DRB1*15:01–DQA1*01:02–DB1*06:02 hap-
lotype. This latter haplotype is normally found with HLA-A*03: and HLA-B*07:, and
the extended haplotype is labeled with the abbreviation HLA 7.2 because of the linkage
disequilibrium between DRB1*15 (originally the major form of DR2) and HLA-B7. This
haplotype is very common in European populations. Recent studies suggest the concor-
dance rate for multiple sclerosis in monozygotic twins may be as high as 25%, while in
dizygotic twins concordance is five times lower than in monozygotic twins and in sib-
lings concordance is 10 times lower. However, concordance rates should be considered
with caution as it appears that concordance rates mirror prevalence rates; therefore the
lower the prevalence in the population, the lower the level of concordance. For example,
there are significant differences in concordance rates between Northern and Southern
European populations. This difference is reflected in the lower prevalence for multiple
sclerosis in Southern Europe.
Early studies of multiple sclerosis
Early studies of the genetic basis of multiple sclerosis identified strong reproducible associa-
tions with the MHC and one haplotype in particular, the HLA 7.2 ancestral haplotype. The
first studies to indicate an association with the MHC were published in 1972 and 1973 by
the same group. Many others followed, but it would be wrong to give the impression that
all early studies identified the same HLA associations or that all studies found any associa-
tion. The reasons for these inter-study variations include differences in techniques applied
and the availability and quality of anti-serum for the early HLA assays as well as sample size,
diagnostic differences, and population differences (Chapters 3 and 5). For example, two
studies based in the UK reported quite different results—one was based on the population
of the remote Orkney Islands and the other based on the general UK population.
However, the association with the 7.2 haplotype has stood the test of time and remains a
confirmed multiple sclerosis risk marker. The same group of HLA class II alleles, especially
DQB1*06:02, are very strongly associated with narcolepsy (as we have discussed), but
DQB1*06:02 is protective against T1D and may have mild protective effects against other
autoimmune conditions (though not all; see below).
The association with the MHC is suggestive of an autoimmune or at least immune-medi-
ated etiology in multiple sclerosis. In fact, studies confirm this link—multiple sclerosis is
an immune-mediated disease. Furthermore, it appears to be an autoimmune disease since
there are myelin antibodies in cerebrospinal fluid. Other evidence to support this comes
from studies of T cell function in multiple sclerosis. The activation of T helper 1 CD4+
cells and production of interferon (IFN)-γ are implicated in the pathogenesis of multiple
sclerosis. These cells are found to be enriched in both the blood and spinal fluid of patients,
and activated macrophages that are found in sclerotic plaques can release cytokines and
other enzymes that are able to cause demyelination. IFN-γ was trialed as a potential treat-
ment for multiple sclerosis, but was found to worsen rather than improve symptoms.
With so much already known, some people question the value of further genetic studies
in multiple sclerosis. There are two reasons why investigations should continue. The first
is illustrated by the disappointing results of the IFN-γ trials, that is our knowledge to
date is incomplete and not sufficient to aid in the formulation of effective therapies. The

2967_Ch06.indd 200 07/07/2015 11:57

 HLA Class II and Disease 201

second is illustrated in the HLA association. Though we have this information, we have
no idea what the auto-antigen or auto-antigens are, and how many immune-regulatory
processes have failed or broken down to permit this disease to occur. Genetic studies offer
one way to identify these.

Recent studies of multiple sclerosis including genome-wide association

studies (GWAS)
Multiple sclerosis is an immune-mediated disease and so we would expect the MHC to be
a dominant risk factor, but what other genetic risk factors may be hiding. Autoimmunity
involves disordered immune regulation. There is a breakdown in T cell tolerance due to
incomplete deletion of self-reactive T cells in the thymus and/or the periphery coupled
with insufficient control of T cell co-stimulators. There is a whole family of additional
immune regulatory genes that could be involved. Studies of other similar diseases (related
syndromes) will provide clues to other potential risk alleles. Overlap is not uncommon, as
illustrated in the example of rheumatoid arthritis (see Figure 4.10).
In 2007, the number of identified risk alleles for multiple sclerosis was 29. In 2011,
Sawcer et al., as part of the International Multiple Sclerosis Genetics Consortium (IMSGC)
together with the Wellcome Trust Case Control Consortium 2 (WTCCC2), published a
follow-up GWAS on 9772 cases and 17,376 unaffected controls from 15 countries, and
reported a total of 57 associations. This group included 23 previously suspected associa-
tions, 29 novel associations above the significance threshold and a further five probable
associations at a lower significance threshold of P < 5 × 10−5 to P > 5 × 10−7. The evidence
of these associations also supports the likelihood that the inflammatory processes seen in
multiple sclerosis pre-empts the diagnostic symptoms seen and not the other way around
as some have suggested. Candidates identified in these studies include various chemokines
and cytokines, members of the tumor necrosis superfamily, various co-stimulatory and
signaling genes, and one or two genes involved in neurological development.
This clustering of susceptibility loci from the same gene family suggests common path-
ways in the genesis of multiple sclerosis, perhaps indicating that the polymorphisms and
mutations all with low individual risk work in an additive model to create a risk portfolio
for multiple sclerosis, i.e. the sort of interactive network that is illustrated in Figure 6.13.
It is relatively easy to create mind-maps linking potential candidates from association
studies for multiple sclerosis and these genetic discoveries can be used to inform the debate
on disease pathogenesis.

HLA class II and autoimmune liver disease

Liver disease is often thought of as a self-induced disease resulting from excess alcohol
consumption. However, for the majority of cases nothing could be further from the truth.
Liver disease affects the old and the very young, and comes in all shapes and sizes. There
are Mendelian autosomal liver diseases such as hemochromatosis and Wilson’s disease,
there are drug induced liver diseases such as those that occur following an adverse drug
reaction, there are autoimmune liver diseases, and there are viral liver diseases to consider.
Altogether, liver disease accounts for a very significant proportion of mortality and mor-
bidity within the population, and most forms of liver disease have a genetic association
with HLA (Figures 6.14 and 6.15).
Essentially there are three different autoimmune liver diseases: autoimmune hepatitis
(AIH), primary sclerosing cholangitis (PSC), and primary biliary cirrhosis (PBC). The
disease names are to some extent descriptive of the signs of the disease.

2967_Ch06.indd 201 07/07/2015 11:57

 202 CHAPTER 6 The Major Histocompatibility Complex

MHC
CD80/
CD86

Cytokines T cell
IL12, selection
TNFRSF THEMIS

NK cells T cells B cells

MS

Figure 6.13: Illustrating risk genes and cell targets for multiple sclerosis. This figure illustrates the complexity
of interaction between a few selected immune response genes associated with an increased risk of multiple
sclerosis (MS). There are several different explanations for this. Risk alleles may work independently or in an
additive model. The additive model may include a number of different groups. In an additive model, it may
be necessary to inherit several common polymorphisms before there is significant risk of multiple sclerosis.
Alternatively, under the correct set of environmental conditions a solitary polymorphism may be sufficient to
significantly promote the disease or protect from it.

AIH is a relatively rare classical autoimmune disease of the liver

AIH presents in children and in young and mature adults. The disease has all the classical
features of an autoimmune disease. AIH has a strong female preponderance. It can be con-
trolled in the majority of cases by use of immunosuppressive drugs; however, the disease is
progressive and not all cases respond to immune suppression. Antibodies against a variety
of antigens have been identified, though there is some controversy over which antibody is
most helpful in diagnosis and which is most informative in determining the pathogenesis
of the disease. The disease can be subclassified into different types, but we are concerned
only with type 1 AIH and will use the term AIH throughout this section.
A genetic association between AIH and HLA was first reported in 1972 (HLA-A1 and
HLA-B8). Subsequent studies confirmed this association, and added Dw3 (HLA-DR3)
and HLA-DR4 (in late onset cases) to the list of susceptibility alleles. A series of stud-
ies published by the same group from 1991 onwards, confirmed the association with
HLA-DR4 in late-onset cases, and extended the susceptibility haplotypes to include
DQA1-DQB1 as well as HLA-C and the C4A* null sequence (C4A*Q0). A summary of
this work is presented in Table 6.10.
A shared epitope hypothesis in AIH
The studies on populations from the UK and Northern USA identified HLA DRB1 as
the primary susceptibility locus and DRB1*03:01 and DRB1*04:01 as the key alleles.
As these studies had used molecular genotyping, it was possible to compare the amino
acid sequences encoded by the susceptibility alleles with those encoded by the protective
alleles. After a thorough statistical analysis, one particular amino acid sequence stood

2967_Ch06.indd 202 07/07/2015 11:57

 HLA Class II and Disease 203

AIH
HLA 8.1
DRB1*04
PSC
HAV HLA 8.1
DRB1*13:01 DRB1*13:01

Positive
HLA
Chronic associations PBC
HCV DRB1*08-
DQB1*03:01 DQA1*04:01

HBV DILI-FLU
DRB1*13:01 B*57:01

Figure 6.14: HLA and liver disease. The figure illustrates the major positive HLA associations with seven
different forms of liver disease. Clockwise these are: autoimmune hepatitis (AIH), primary sclerosing cholangitis
(PSC), primary biliary cirrhosis (PBC), drug-induced liver injury specifically flucloxacillin (DILI-FLU), hepatitis B
virus (HBV), hepatitis C virus (HCV), and hepatitis A virus (HAV). The first three are all autoimmune diseases; the
last three are viral liver diseases. For AIH, DRB1*04 represents DRB1*04:01 in European and North Americans of
European origin and DRB1*04:05 in Japanese, Korean, and South American adult cases. The association between
PSC and the HLA 7.2 haplotype is weak and controversial, and is not reported in the figure. The association with
PBC in Japan and China may involve a different DQA1-DQB1 haplotype: DQA1*01:03–DQB1*06:01. DILI-FLU
refers to the extraordinarily strong association with HLA-B*57:01 seen in a minority of DILI patients treated with
the commonly prescribed antibiotic flucloxacillin.

out: leucine–leucine–glutamic acid–glutamine–lysine–arginine (LLEQKR) at positions

67–72 of the DRβ-chain. This sequence, which is referred to as a shared epitope for
AIH, is not unique to DRB1*03:01 and DRB1*04:01, but is present on a number of
different DRB1 allele sequences. Alleles associated with resistance to AIH encode isoleu-
cine–leucine–glutamic acid–glutamine–alanine–arginine (ILEQAR) at positions 67–72,
respectively. There are only two major differences between these two sequences: the first
at position 67 and the second at position 71. The change of leucine for isoleucine at posi-
tion 67 is unlikely to be of major significance, but the change of lysine (a highly charged
polar amino acid) for alanine (a non-polar uncharged amino acid) is a major change.
Therefore, it is reasonable to hypothesize that this change in the amino acid sequence in
the DRβ-chain at position 71 may affect antigen presentation. The residue at position
DRβ71 interacts with the P4 binding pocket of the HLA DR molecule. This difference
may be critical in determining which antigens are preferentially bound and the orienta-
tion of the antigen for presentation to the TCR. The same epitope is also found to be
associated with an increased risk of rheumatoid arthritis. Genetic overlap is common in
complex diseases. However, it is likely that there are both disease specific and non-specific
risk alleles that determine which diseases occur in different individuals.
Studies of other populations in AIH
Looking at other populations can be very helpful in deciding where to place weight on these
findings. Studies in Japan and Korea identified DRB1*04:05 as the primary susceptibility

2967_Ch06.indd 203 07/07/2015 11:57

 204 CHAPTER 6 The Major Histocompatibility Complex

AIH
DRB1*15:01

DILI- Negative PSC

Augmentin HLA DRB1*04
DRB1*07:01 associations MICA*002

PBC
DRB1*11
DRB1*13

Figure 6.15: Protective HLA associations with three major autoimmune liver diseases and one example
of drug-induced liver injury. The figure illustrates four major negative (protective) HLA associations with
three autoimmune liver diseases and a selected example of drug-induced liver injury (DILI). Clockwise these
are autoimmune hepatitis (AIH), primary sclerosing cholangitis (PSC), primary biliary cirrhosis (PBC), and drug-
induced liver injury, specifically augmentin (DILI-Augmentin). Comparing Figure 6.14 with figure 6.15, DRB1*04
which is associated with an increased risk in AIH protects from PSC and DRB1*13 which is associated with an
increased risk of PSC, chronic HAV infection in South America, and AIH in children from Argentina is associated
with a reduced risk of PBC. The DRB1*07:01 haplotype that protects from augmentin DILI is the one which carries
HLA-B*57:01 – the major risk allele for flucloxacillin DILI.

allele in AIH. The Korean and Japanese populations have a shared ancestry and therefore
this was not surprising. However, compared with the European and North American find-
ings, initial analysis suggested the results were discordant and do not agree with the shared
epitope model for AIH. DRB1*04:05 does not encode LLEQKR at positions 67–72, it
encodes LLEQRR. However, under further scrutiny it was found that the only difference
is at position 71 where DRB1*04:05 does not encode lysine but it encodes arginine and
this amino acid is very similar to lysine, being a highly charged polar amino acid of similar
structure. Therefore, at a functional level these two alleles may encode molecules with very
similar preferences for antigen binding.
Extending the shared epitope model for HLA-encoded susceptibility to AIH
At present the LLEQKR model with emphasis on DRβ-71 is the best model to explain
HLA-encoded susceptibility to AIH. However, there may be scope to extend the model
to include amino acids upstream of position 71 as far as position 74. This idea was first
promoted in a paper from Korea by Lim et al. in 2008. In the extended shared epitope
model, LLEQKRGR is encoded by the DRB1*03:01 allele and LLEQK-(or R)-RRA is
encoded by the DRB1*04:01 and DRB1*04:05 alleles. In the extended model there is a
critical difference between DRB1*03:01 (on one side) and DRB1*04:01 and DRB1*04:05
(on the other). DRB1*03:01 encodes arginine at position 74, while both DRB1*04:01
and DRB1*04:05 encode alanine. As we have already seen, these are two very different
amino acids with different polarity. Essentially, Lim et  al. argue that DRB1*04:01 and
DRB1*04:05, which both carry arginine 71 and are positively charged, can form a salt
bridge with the negatively charged P4 binding pocket. DRB1*03:01 is also able to form

2967_Ch06.indd 204 07/07/2015 11:57

 HLA Class II and Disease 205

Table 6.10 Autoimmune Hepatitis and HLA alleles and haplotypes.

HLA locus
Haplotype A C B DRB3/4/5 DRB1 DQA1 DQB1
Haplotypes that encode susceptibility to AIH
8.1 *01 *07 *08 3*01:01 *03:01 *05:01 *02:01
DR4–DQ7 – – – 4* *04:01 *03 *03:01
DR4–DQ4 – – *54 4* *04:05 – *04:01
A11–DR4 *11 – – 4* *04 – –
*04:05
DR13 – – – – *13:01 *01:03 *06:03
Haplotypes that encode resistance (protection) to AIH
7.2 – *07 – 5*01:01 *15:01 *01:02 *06:02
DR4–DQ7 – – – 4* *04:06 – *03:01
DR13 – – – – *13:02 – –

The first two haplotypes are associated with AIH in UK, Northern Europe and USA. The third haplotype is
associated with AIH in Japan and Korea. The fourth and fifth haplotypes are found in pediatric AIH in Argentina. In
Argentina, DRB1*13:01 is the major susceptibility determinant in pediatric AIH, but not in adult cases. In adult AIH,
from South America DRB1*04:05 is the major allele. Not all studies are listed, only enough to give a cross-sectional
view of the data on AIH and those which contribute most to the scientific discussion of this subject. In cases where
individual groups have produced multiple studies, the study with the largest number of patient cases is shown.
Alleles are not listed at all loci for all haplotypes, in each case a dash is entered to indicate this.

this salt bridge as it has lysine at position 71. This confirms the idea that the interaction
of the DRβ-71 residue with P4 is of primary importance in AIH. This implies that the
amino acid at position 74 is of secondary importance. Thus, the alanine carried by both
DRB1*04:01 and DRB1*04:05 at position DRβ-74 does not appear to influence the effect
of the amino acid at position DRβ-74 on the P4 binding pocket due to its small size and
charge, whereas DRB1*03:01 has arginine at position 74 and this can also interact with
the P4 binding pocket to form a second salt bridge. In this extended model those carry-
ing DRB1*03:01 receive a double-dose effect through structural variations encoded by a
single allele. This may also go some way to explaining the differences in HLA distribution
reported in early- and late-onset disease in AIH (Figure 6.16).
A slightly more difficult situation arose when studies from South America reported associ-
ations with DRB1*13:01 in children with AIH. However, other studies in South America
have suggested that chronic hepatitis A virus is associated with DRB1*13:01 and this
virus has been identified as a potential trigger for AIH. Interestingly, adult patients from
Argentina have an excess of DRB1*04:05, that is the same allele that is associated with sus-
ceptibility to AIH in Japan and Korea. Thus, at least the data for South American adults
fits into the shared epitope hypothesis. The different data in children from South America
remains unexplained.
Haplotypes and clinical severity in AIH
The two AIH major susceptibility haplotypes DRB1*03:01 and DRB1*04:01 or DRB1*04:05
(depending on nationality) each carry a second expressed DRB gene: DRB3 for DRB1*03:01

2967_Ch06.indd 205 07/07/2015 11:57

 206 CHAPTER 6 The Major Histocompatibility Complex

DRBI*03:01 P4 pocket salt

Double strike bridge formation

R74 K71

DRBI*04:01 P4 pocket salt

DRBI*04:05 bridge formation
Single strike

A74 R71

Figure 6.16: Salt bridge formation over the P4 binding pocket in AIH may explain genetic associations
with different HLA alleles. The figure is a simplified illustration of the peptide-binding groove of an HLA
molecule focusing on the interaction between peptides at positions 71 and 74 on the DRB1-encoded β-chain.
The figure compares the formation of the salt bridge at the P4 binding pocket (fourth binding pocket for antigen
peptide side chains) in two groups; those with DRB1*03:01 and those with either DRB1*04:01 or DRB1*04:05.
These alleles are the major susceptibility alleles for autoimmune hepatitis in the European, European American
(the first two), and Japanese populations (the last). All alleles are able to form a salt bridge; however, those
carrying DRB1*03:01 have a double strike because they have positively charged amino acid at positions 71
(lysine K) and 74 (arginine R), while those with DRB1*04:01 and DRB1*04:05 have only a single strike because
though they have arginine at position 71, they have neutral alanine (A) at position 74. This may account for
the observation that AIH patients with DRB1*03:01 tend to present earlier and may have more severe disease.
The model presented is similar to that seen for other autoimmune diseases. In those with DRB1*04:06, there is
glutamate at position 74 and even though they have arginine at position 71 they are unable to form a salt bridge
because the negatively charged glutamate repels the negatively charged P4 peptide.

and DRB4 for DRB1*04:01/04:05. While DRB3 encodes LLEQKRGR, DRB4 does not.
Therefore patients with the DRB1*03:01 haplotype have at least two DRB alleles, each with
the same critical sequence, whereas those with DRB1*04:01 or DRB1*04:05 do not. It is
therefore possible that the increased number of LLEQKRGR-expressing alleles restricts the
selection of antigenic peptides available for binding to the expressed DR molecule. This
multihit hypothesis may explain why patients with DRB1*03:01 present earlier in life and
those with DRB1*04:05 present later. In fact, early-onset AIH is rarely seen in Japan or
Korea and the association with DRB1*04:01 is with late-onset cases in European popula-
tions. Based on all of the information above, the possible gene dose effect of having a single
DRB1*03 gene will increase to two hits, i.e. lysine 71 (first hit) and arginine 74 (second
hit) for the DRB1*03 allele alone, and this will increase from two to four hits in most cases,

2967_Ch06.indd 206 07/07/2015 11:57

 HLA Class II and Disease 207

as they will also all have the DRB3 gene that encodes the LLEQKRGR sequence. Both
genes on the same haplotype have a potential two-hit locus (two times two equals four).
Homozygotes for DRB1*03-DRB3 haplotype may have eight hits.
This does not rule out the possibility that other genes within the same MHC haplotype
are having an impact on disease risk. It is worth noting that recent findings on rheumatoid
arthritis indicate the possibility of another three MHC genes encoding five amino acids
as risk factors. With respect to AIH, it is also important to understand when comparing
populations that the DRB1*03:01 allele and the ancestral HLA 8.1 haplotype to which
most DRB1*03:01 alleles are connected are rare in Japan and Korea as is the DRB1*04:01
allele. It is therefore not surprising when the associations reported in Europe and the USA
cannot be replicated in these quite distinct populations, and it is a major advance when
unifying hypotheses (such as that above) can be developed from what may appear to be
conflicting data.

PSC is not a classical autoimmune disease

PSC is a chronic cholestatic syndrome of unknown etiology characterized by diffuse
inflammation, biliary obstruction, and cholestasis, leading to fibrosis of the biliary tree.
Even though PSC is quite rare (around 1:10,000 adults, which allowing for adult onset,
equates to approximately 4000 cases living in the UK in 2014), because there is no effec-
tive therapy it remains a leading cause for liver transplantation. The intrahepatic and
extrahepatic bile ducts are destroyed in the disease process, and biliary cirrhosis, portal
hypertension, and liver failure follow. First described in 1924 by Delbet, it was until
recently considered a rare condition, but it is now being identified more frequently due
to advances in diagnostic medicine. It is for the most part a severe disease that progresses
quite rapidly to a clinical endpoint. The disease is frequently found to be associated with
mild ulcerative colitis. In fact, some centers report that 100% of cases have ulcerative coli-
tis. The relationship between ulcerative colitis and PSC is unclear; however, even though
a large proportion of patients with PSC have ulcerative colitis, only a small proportion of
ulcerative colitis patients develop PSC.

PSC and genes

PSC is not a classical autoimmune disease. It is found predominantly but not exclusively
in adult males and does not respond to normal immune suppressive therapy. However,
genetically PSC looks like an autoimmune disease.
In terms of genetics, PSC has often been referred to as an autoimmune disease and there-
fore early studies investigated possible relationships with the MHC. Studies of HLA based
on very small numbers of patients in 1982 and 1983, reported associations with HLA-A1
and HLA-B8. These studies were followed up by a series of studies based on a relatively
large collection of cases from a single center. The studies first used serological phenotyp-
ing, then applied RFLP-based genotyping, and finally used a variety of PCR-based geno-
typing methods to investigate HLA-A, -B, -C, -DRB1, -DRB3, -DRB4, -DRB5, -DQA1,
and -DQB1. These studies and studies of other populations reported three HLA haplo-
types associated with an increased risk and two associated with a decreased risk of PSC.
The data from these studies is summarized in Table 6.11.
Comparing studies, the associations with haplotypes 2 (7.2) and 5 (MICA*002) remain
controversial. However, even these large-series studies are quite small by modern standards
and much larger numbers may be needed to confirm or refute these weaker associations
with a realistic degree of confidence.

2967_Ch06.indd 207 07/07/2015 11:57

 208 CHAPTER 6 The Major Histocompatibility Complex

Table 6.11 Primary sclerosing cholangitis and HLA alleles and haplotypes.

HLA locus
Haplotype A C B DRB3/4/5 DRB1 DQA1 DQB1
Haplotypes that encode susceptibility to PSC
8.1 *01 *07 *08 3*01:01 *03:01 *05:01 *02:01
7.2 *03 *07 *07 5*01:01 *15:01 *01:02 *06:02
DRB1*13 – – – 3*01:01 *13:01 *01:03 *06:03
Haplotypes that encode resistance (protection) to PSC
DRB1*04 – – – 4*01:03 *04:01 *03 *03:02
DRB1*07 – – – 4*01:03 *07:01 *02:01 *03:03
MICA*002 – – – – – – –

Note that the haplotype 7.2 is only weakly associated with PSC as is the DRB1*07 haplotype. The MICA*002
association is discussed later in the chapter. Note that both the 8.1 and 7.2 haplotypes carry the MICA*008 allele.
Alleles are not listed at all loci for all haplotypes, in each case a dash is entered to indicate this.

Leucine DRβ-38 versus asparagine DRβ-37

Analysis of these associations through consideration of the amino acid sequences encoded
by the different alleles initially suggested that the key amino acid was leucine at position
38 of the DRβ-chain. Leucine DRβ-38 is encoded on both of the second expressed DRB
genes for the HLA 8.1 and the HLA 7.2 ancestral haplotypes (i.e. DRB3 and DRB5),
but is not encoded by the DRB1 alleles. The protective haplotypes carry the DRB4 alleles
that encode alanine at this position and all other alleles encode valine. Further exploration
of the amino acid sequences in the DQ region revealed a potential role for DQβ-55 and
DQβ-87, explaining the genetic susceptibility with haplotypes 2 and 3. In 2011, however,
Hov et al. reinvestigated the electrostatic modifications of the HLA DR molecules associ-
ated with inheritance of different HLA DRB alleles in 356 PSC patients and 366 matched
controls, and identified DRβ-37 in the DRB1 gene and not DRβ-38 on the DRB3 and
DRB5 genes as the primary susceptibility determinant. This study reconsidered previous
studies in detail, pointing out that the second expressed DRB gene is expressed at a low
frequency and that the DQB1 alleles were different for each haplotype, though there are
shared sequences at critical sites. As the association with the DRβ-38 model is based on
the DRB3 and DRB5 genes, and the DQβ model makes no allowance for the association
with the DRB1*03:01–DQB1*02:01 haplotype (which is the strongest of all associated
HLA haplotypes in PSC), the overall weight of evidence is in favor of the findings of Hov
et al. The key amino acids are asparagine (which increases the risk of PSC) at position
37 that induces a positive charge in the P9 binding pocket, and tyrosine (which protects
against PSC) at position DRβ-37 that induces a negative charge in the P9 binding pocket.
Variation at position DRβ-86 was also found to influence the electrostatic properties of
the P1 binding pocket, and appears to have a secondary role in susceptibility and resis-
tance to PSC (Figure 6.17).
Just like the previous examples (narcolepsy and T1D), amino acid changes that alter the
electrostatic charge of the antigen-binding pockets (in this case P9) appear to be critical in
determining HLA-encoded susceptibility to the disease. Though in narcolepsy and T1D it
is the DQB locus and not DRB that is the primary susceptibility locus, the basic concepts

2967_Ch06.indd 208 07/07/2015 11:57

 HLA Class II and Disease 209

P9
N37
P1
V86

Figure 6.17: HLA DR binding groove and HLA-encoded susceptibility to PSC. The figure shows a simplified
version of the HLA peptide-antigen binding groove from a top-down perspective. The amino acids asparagine
(N) at position 37 and valine (V) at position 86 are highlighted. These two amino acids interact at the P9 and
P1 binding pockets, respectively. Studies indicate that HLA DRB1 alleles carrying the amino acids asparagine 37
and valine 86 are most strongly associated with PSC in European populations and account for the majority of
HLA-encoded susceptibility to PSC. (Adapted from Hov JR, Kosmoliaptsis V, Traherne JA et al. [2011] Hepatology
53:1967-1976. With permission from John Wiley and Sons.)

remain the same, and it is interesting that both AIH and PSC favor associations with DRB
rather than DQ (as does rheumatoid arthritis).
MICA in PSC
The MICA genes are encoded in the so-called MHC class III region, and are stress induc-
ible and expressed on the gastric epithelium. MICA can activate NK cells and may also
activate γδ T cells. All of these factors make them ideal candidates to consider when look-
ing at the MHC in PSC. There is considerable polymorphism in MICA. Based on their
putative function, their location, and the level of polymorphism in the MICA and MICB
genes, two studies in 2001 investigated MICA in PSC: one from the UK and the other
from Norway. The UK study described a strong association with MICA*008 and a strong
dominant protective effect of MICA*002. In the UK study there were two patient groups
totaling 112 altogether and both patient groups reported statistically significant associa-
tions with these alleles. Overall, the MICA*008 association appeared to be due to a very
high frequency of homozygous patients.
The Norwegian study described associations with the MICA5.1 and MICB24 markers.
The method employed for MIC genotyping in this study was different from that above.
MICA5.1 is a marker for the ancestral 8.1 haplotype. This study did not agree with that
from UK. One reason for this is the different techniques; the other lies within a subtle
difference between the two populations. These two populations though similar are not
the same. At the time of this study the Norwegian PSC patients were all diagnosed to be
positive for ulcerative colitis, whereas in the UK only 70% of cases of PSC are diagnosed
to be positive for ulcerative colitis.

2967_Ch06.indd 209 07/07/2015 11:57

 210 CHAPTER 6 The Major Histocompatibility Complex

The UK set have a weak but confirmed association with the ancestral 7.2 haplotype.
This association is not seen in either Sweden or Norway. The MICA*008 allele is carried
on both the HLA 8.1 and 7.2 ancestral haplotypes (haplotypes 1 and 2 in Table 6.11).
The absence of the weak association with the 7.2 haplotype in the Scandinavian popu-
lation may account for this difference, and it may also explain the difference reported
at the molecular level regarding electrostatic charge and position DRβ-37 versus posi-
tion DRβ-38. This difference may also reflect genuine population differences and case
ascertainment criteria.
In the case of DRβ-38 versus DRβ-37, clearly the latest data in favor of DRβ-37 are very
convincing and clearly correct, but the potential role of MICA has yet to be determined
in a positive or negative manner. There are four reasons for this:
• The Scandinavian team did not identify MICA*008, but a marker for the HLA 8.1
haplotype MICA*5.1.
• The sample sizes in the two studies were comparable (112) versus (130).
• The location of MICA expression in relation to the disease.
• The function of MICA and the fact that the liver has very high numbers of NK and
γδ T cells.
These observations are likely to be only a piece of the final picture and other MHC genes
may have a role to play in PSC. Until very large-series studies of the whole MHC region
have been performed in PSC we will have to wait to find out what else is hiding within this
region. Finally, we do not yet know the antigenic peptides that are preferentially bound by
these HLA molecules. Identifying these is the essential theme for this research as this will
provide insight into the disease pathogenesis and will hopefully lead to novel therapies.

PBC is an autoimmune liver disease with a genetic component

PBC is inflammatory non-suppurative disease of the intra-hepatic bile ducts. PBC leads to
biliary fibrosis, cirrhosis, and liver failure. This disease is more common than PSC, occur-
ring in approximately 1:3000 people. It is more common in women than in men, with
between 90–95% patients being female and it usually presents over the age of 40. The
disease is considered to be an autoimmune disease because of the female preponderance
and the presence of high titers of auto-antibodies and the auto-reactive CD4+ and CD8+
T cells, specific for epitopes derived from the pyruvate dehydrogenase complex E2 antigen
(PDC-E2). However, the disease does not respond to classical therapy with corticosteroids
or other immunosuppressive therapy.
Evidence of a genetic component to this disease is based on:
• Gender bias (up to 95% patients are female).
• Geographic and familial clustering (it has been estimated that 6.4% of cases have at
least one other affected family member).
• The sibling relative risk (λ) is 10.5 (with this figure rising to 58.7 for daughters of
affected mothers).
• Concordance in monozygotic twins is 64% (though this is based on a small self-report-
ing sample).
• Overlap with other autoimmune diseases including autoimmune thyroid disease and
SLE.

2967_Ch06.indd 210 07/07/2015 11:57

 HLA Class II and Disease 211

Early studies of PBC found no reproducible HLA associations

Contrary to the findings in AIH and PSC, early studies of PBC failed to find an associ-
ation with any of the markers of the HLA 8.1 or 7.2 ancestral haplotypes. Even though
two small phenotyping studies reported associations with DR2, DR3, and DR4, these
were not confirmed. During the 1980s there were multiple studies of HLA-A and -B
in PBC, but none proved fruitful. Studies of DR using serological phenotyping were
of variable quality and most were negative, with one exception. In 1987, Gores et al.,
working at the Mayo Clinic in Rochester Minnesota, USA, struck gold. In their study,
they used high-quality serotyping and detected a previously unreported association
with DR8. Prior to this study, this association had been missed because most com-
monly used serotyping trays were poorly equipped to detect the DR8 antigen. This
missed association illustrates the importance of considering the limitations of meth-
odology, before drawing conclusions from genotyping and phenotyping studies from
the past. The differences in methods are illustrated in Table 6.3, which shows a 69%
concordance rate for serologically typed HLA-DRB1*08-positive samples versus RFLP,
which was found to be 100% concordant with PCR-based methods (Table 6.4) on
the same samples. These data are from the mid-1990s. Prior studies were for the most
part unable to detect DR8 with even this degree of accuracy. Following the study of
Gores  et  al., multiple studies were able to confirm the association with DR8 using
RFLP analysis, which had only recently been introduced and is an excellent method
for detection of HLA-DR8 as indicated in Table 6.4, where all 16 DR8 RFLP-positive
samples were confirmed by PCR. The association between PBC and DR8 is unusual
in two ways.
• Few diseases are associated with HLA-DR8.
• The association accounts for less than 25% of the total patient group.
Throughout the 1980s and 1990s studies continued to find DR8 associated with PBC. In
Northern and Southern European populations, and in their descendants in North America,
the key allele appears to be DRB1*08:01, but in Japan the key allele is DRB1*08:03. These
two alleles differ by a single amino acid at position 67, where DRB1*08:01 encodes phe-
nylalanine and DRB1*08:03 encodes isoleucine. Extending the studies to the other class
II genes identified DQA1*04:01 and DQB1*04:02 as the primary DQ elements of the
DRB1*08:01 haplotype in Europeans and DQ3 as the primary DQ element in Japanese
patients. The finding of different DQ alleles on the two DR8 haplotypes and the absence
of a second expressed DRB gene on the DR8 haplotypes suggested that DRB1 is the pri-
mary susceptibility locus.

Recent studies in UK and Italy

More recent studies (2006 onwards) on larger series in the UK and Italy identified previ-
ously overlooked protective associations with DRB1*11 and DRB1*13 family members.
This made it possible to review the data from a molecular point of view. Comparing the
alleles associated with increased susceptibility to those associated with reduced risk of PBC
shows there are four significant amino acid residues. Presented in the order DRB1*08 ver-
sus DRB1*11 or DRB1*13 below, these are:
• Glycine for serine at position 13.
• Tyrosine for histidine at position 16.
• Tyrosine for phenylalanine at position 47.
• Leucine for alanine at position 74.

2967_Ch06.indd 211 07/07/2015 11:57

 212 CHAPTER 6 The Major Histocompatibility Complex

As the associations with DRB1*11 and DRB1*13 are weak in some populations this model
is speculative, but it provides a basis for considering the molecular biology of this associa-
tion and the binding groove, and moves the discussion away from the simple process of
genotype collecting. Three of the four positions above are of potential functional impor-
tance. The amino acid at position 13 affects the binding of antigenic side chains associated
with both the P4 and P7 binding pockets, while the amino acids at positions 47 and 74
affect P4 and P6, respectively.

HLA-DQA1*04:01 may be the primary PBC risk allele in the MHC

Prior to 2012, the situation for PBC looked very like that for AIH and PSC with suscep-
tibility mapping to DRB1. However, work on PBC based on imputed HLA genotypes
from GWAS studies in 2012 revealed that the likely primary MHC encoded susceptibility
alleles are on the HLA-DQA1 locus, not DRB1. The allele in question is DQA1*04:01,
which is almost always found with DRB1*08:01, though a very small proportion of
DRB1*08:01 haplotypes do not carry DQA1*04:01. The reason for the difference between
the 2012 work and earlier studies is simple. Earlier studies were based on smaller numbers
(compared with this GWAS-based study) and given the extreme linkage disequilibrium
between DRB1*08:01 and DQA1*04:01, early studies will have been unable to distin-
guish between the influence of DRB1 and DQA1 because there were so few DRB1*08:01
cases without DQA1*04:01. In fact, few studies included DQA in their analysis, though
most recent studies have included DQB. Separating DQA1*04:01 from DRB1*08:01 has
required a very large cohort of cases. In addition, the DQA1*04:01 association currently
reported is based on imputed data, not HLA genotyping of samples, and this too may
have had an impact.
It is important to compare genetic associations in different populations before coming to a
final conclusion in PBC and other diseases, such as AIH and PSC above. If the association
with DQA1*04:01 is the primary association with PBC, then the role of DRB1 needs to
be reconsidered. However, we should not cast aside the work of earlier studies and indeed
there is some discord over this latter finding. In the Japanese and Han Chinese popula-
tions, the most common DRB1*08 family member is DRB1*08:03. In these populations
the DRB1*08:03 allele most frequently occurs with DQA1*01:03-DQB1*06:01 and it
rarely occurs with DQA1*04:01.
It is important to consider these findings and differences between studies with care.
Current studies are the fruits of earlier work and even though the focus of attention has
now moved from DR to DQ, there is no doubt about the initial studies being correct in
that there is a strong association with DRB1*08 in PBC. Whatever the final interpretation
of the data on HLA in PBC, future studies need to consider other populations, and also
leave room for recently discovered protective associations with DRB1*11 and DRB1*13.
In addition when considering HLA we must not forget that the MHC is complex and
there is always room within a haplotype for a second bite of the susceptibility cherry, as we
have seen in AIH. Identifying the key to this strong association with HLA will ultimately
help us to unpack the pathogenic mechanisms that lead to PBC.

Non-MHC-encoded regulatory genes in PSC and PBC and in PSC and UC

A number of markers have been associated with both diseases. Currently, there are 12
non-MHC risk loci for PSC (nine of which were identified in 2013) and 25 in PBC
(three of which were new in 2012). In the 2012 PBC study by Liu et  al., only 16 of
the 22 previously identified associations were significant at P < 5 × 10−7. In both PSC
and PBC there is overlap of genetic risk with other diseases, e.g. TYK2 is found in both

2967_Ch06.indd 212 07/07/2015 11:57

 Non-HLA MHC Genes and Disease 213

PBC and ankylosing spondylitis, but there does not appear to be any overlap between
PSC and PBC, though some members of both risk groups have roles in cytokine produc-
tion and regulation (IL12A in PBC; IL2 and IL2RA in PSC). Ulcerative colitis (usually
mild) occurs in 70% of all PSC cases. These two diseases share a number of risk alleles
and have some that are not shared. Two examples of PSC/ulcerative colitis risk alleles are
MST1 and IL21; others are found in PSC only, e.g. IL2RA and SIK.

6.7 Comparing the Hla Associations of the

Three Liver Diseases
There is a great deal of overlap between diseases. This can be clinical overlap where there
are two or more diseases that fall into the same group or syndrome, or it can be genetic
overlap. Diseases often share the same risk alleles but also have differences. Looking at
HLA in these three diseases we see a lot of genetic differences and some overlap (Figures
6.14 and 6.15). The primary susceptibility haplotype for AIH and PSC is the ancestral
8.1 haplotype. However, that is where the similarity between AIH and PSC stops. In
PSC, the 7.2 haplotype is associated (albeit weakly) with increased risk, whereas in AIH,
the 7.2 haplotype is associated with reduced risk. In complete contrast to both AIH
and PSC, neither of these two ancestral haplotypes has been associated with PBC. In
PBC, DRB1*13 is associated with a reduced risk of disease, whereas in PSC, one of the
DRB1*13 haplotypes is strongly associated with increased risk of disease. These differ-
ences are useful in evaluating the different molecular models proposed for these diseases
and provide further insight into the different mechanisms of disease pathogenesis.
However, even more interesting may be consideration of other diseases, such as the viral
liver diseases hepatitis A, hepatitis B, and hepatitis C. Chronic infection with hepatitis B
virus is associated with an increased frequency of some DR13 haplotypes and DR13 is also
associated with sustained hepatitis A virus infection in South American children. These
associations, especially when they are shared, may give us a clue to the potential of viral or
other infectious triggers for autoimmunity.

6.8 Non-HLA MHC Genes and Disease

While the MHC encodes 252 expressed genes, the majority are not HLA genes. Overall,
only a few genes within this region have been investigated in detail. One consequence
of this is that GWAS studies which often use tagged single nucleotide polymorphisms
(SNPs) across the MHC, seeking to identify novel associations in this area and to confirm
existing associations, are beginning to identify some non-HLA genes as strong contestants
for the title of primary susceptibility locus. These tagged SNPs often have very high odds
ratio and P values associated with them and in some cases these are higher than those
associated with specific HLA alleles. However, we should not over-interpret this. In the
majority of cases the HLA associations reported do stand the test of high-resolution analy-
sis with tagged SNPs. One problem with the MHC is the high level of linkage disequi-
librium seen across the region that makes dissecting the 6p21.3 region especially difficult.
Very large studies will be needed to do this, but the rewards could be great. The MHC
does indeed contain a cornucopia of interesting genes such as those in the MHC class III
region discussed below.

2967_Ch06.indd 213 07/07/2015 11:57

 214 CHAPTER 6 The Major Histocompatibility Complex

The MHC class III region complement, MICA, and TNFA genes in
complex disease
The gene products of the MHC class III region are an assortment of proteins, many with
important roles in immunity and immune regulation. We will consider six genes but only
one disease example. The genes are:
• Complement C2, C4A, C4B, and Bf.
• MICA (MHC class I chain-related A).
• TNFA (TNF-α).

MHC-encoded complement genes and disease

Complement is best described as a system of plasma proteins that mark pathogens for
destruction. It is a critical element in innate and adaptive immunity. Innate (or in-born)
immunity and adaptive (or acquired) immunity are the two central pillars of immunity.
The innate system is active at birth, whereas the adaptive system develops as we grow.
Though there are more than 30 proteins that make up the complement system, C3 is one
of the most important. There are three pathways for complement activation: the classical
pathway, the alternative pathway, and the lectin pathway. For C3 to be activated it needs
to be cleaved into its subcomponents C3a and C3b. Of these two subcomponents, C3b
tags bacteria for destruction, while C3a is involved in recruitment of phagocytes. Though
the 30 complement proteins are encoded by genes located throughout the genome, the
genes that encode the components that cleave C3 in both the alternative and classical
pathways are located in the MHC class III region (see Figure 6.18). The MHC-encoded

Classical pathway Alternative pathway

C4 C2 Bf

C4a C4b C2a C2b C3b

C4bC2a C3bB D

C3 BbC3b Bb Ba

C3a C3b

Recruits Tags bacteria Creates more Creates C5

phagocytes for destruction C3 convertase convertase

Figure 6.18: Complement system and C3 conversion. The two major complement pathways—the classical
pathway and the alternative pathway—converge at the point at which C3 is converted into subunits C3a and
C3b. C2a and C4b subunits are primarily responsible for this process in the classical pathway. Thus, converted
C3 has many different functions. C3a is active in macrophage recruitment. C3b interacts with the factor B of the
alternative pathway and along with factor D cleaves the protein into two subunits: Ba and Bb. The Bb subunit
interacts with C3b and forms a C3 convertase. Through this interaction C3 once converted creates a positive
feedback loop for further conversion of C3. C3 conversion producing C3b also leads to C5 conversion in the
classical pathway. In many ways C3 conversion is the critical stage in the complement pathways.

2967_Ch06.indd 214 07/07/2015 11:57

 Non-HLA MHC Genes and Disease 215

complement genes are C2, C4A, and C4B for the classical pathway and Bf (factor B) in
the alternative pathway.
Polymorphisms in the C2 and C4 genes have been associated with a number of autoimmune
diseases. One reason for this is the very strong linkage disequilibrium in the HLA 8.1 ances-
tral haplotype. Studies of complement genes in diseases where there are associations with
this haplotype will also report associations with C4A null alleles. These nulls are large-scale
deletions of the C4A–CYP21A region of 6p21.3 and they do not have a functional product.
Though these are called null alleles by some, they are in fact null genes as the whole gene
sequence is deleted. However, we each inherit two copies of our genome and in complement
there are two functional forms of C4, designated C4A and C4B. As these have similar func-
tions, the partial absence of an expressed C4A gene (heterozygote) or even complete absence
(homozygote) frequently has no phenotypic consequences. C4B simply makes up for C4A.
Interestingly, there are also two forms of C4B, designated long and short, which differ by as
much as 6 kb in size and there are also null alleles for C2. Copy number variation (CNV)
also occurs with these genes whereby some haplotypes carry multiple copies of the genes. As
C4A null is frequently carried on the HLA 8.1 haplotype it is difficult to select whether or
not this null allele has any effect in diseases with the HLA 8.1 association. However, there is
one outstanding exception: systemic lupus erythematosus (SLE).

Complement and SLE

SLE is an autoimmune disease where the immune response appears to target a whole
system rather than a particular cell type or organ. SLE is characterized by deposition of
immune complexes in the small blood vessels, kidneys, and joints, leading to intense
outbreaks of inflammation. It is thought to occur as a result of complement deficiency;
indeed, 75% of cases have familial disease due to C1q and C2 deficiencies. The remain-
ing cases have sporadic disease and this has been associated with inheritance of the HLA
8.1 ancestral haplotype. However, the association with HLA does not explain all of
the sporadic cases of SLE. Fielder et al., by retyping patients focusing on C4, found a
higher percentage of patients had C4A null alleles than for any of the previously associ-
ated HLA alleles, suggesting the true association was with C4A null rather than with
the HLA class I or class II genes. Further studies in the Japanese population confirmed
this finding. In Northern Europeans the 8.1 haplotype carries the following comple-
ment alleles C4A*Q0–C4B*1–C2*C–Bf*S.
Of course, the association above makes sense at the functional level and it is a good exam-
ple of both an MHC class III-associated disease and the importance of considering the
functional relationship between the identified susceptibility allele and disease. Almost all
individuals who carry the ancestral 8.1 haplotype carry a C4A null (C4A*Q0) but few of
these develop SLE. This is because most people are heterozygotes and also because most
people have a functional C4B allele on the haplotype (C4B*1). C4A and C4B are two
different isoforms of C4 with similar overlapping functions. As most HLA 8.1 carriers
are heterozygotes and because there is the potential for C4B to make up for the absence
of C4A to some degree, most C4A*Q0 carriers do not develop SLE. It is possible that in
SLE disease may only arise when the immune system is stressed and the level of produc-
tion of C4 cannot meet demand. This hypothesis may explain the low level of penetrance
for C4A*Q0.

MHC class I chain-related A

There are two expressed MHC class I chain (MIC  )-related genes (MICA and MICB)
located on the centromeric side of HLA-B and three MIC pseudogenes (MICC, MICD,

2967_Ch06.indd 215 07/07/2015 11:57

 216 CHAPTER 6 The Major Histocompatibility Complex

and MICE  ). Polymorphisms in the A and B genes have been related to a number of
autoimmune diseases including Behçet’s disease and Addison’s disease. MICA encodes an
MHC class I-like molecule with three extracellular immunoglobulin domains, a trans-
membrane segment, and a cytoplasmic tail to anchor the molecule to the cell. MICA is
stress inducible and expressed on gastric epithelium. MICA can activate NK cells and may
also activate γδ T cells. There is considerable polymorphism in MICA. Currently, there are
101 MICA and 41 MICB alleles (encoding 80 and 27 proteins, respectively). A potential
association between MICA polymorphism and PSC is reported above.

TNF-α
TNF-α is encoded by the TNFA gene in the MHC class III region. It is a pro-inflammatory
cytokine with powerful effects that can be localized to infected tissue or produced systemi-
cally through the body. TNF-α is released by macrophages following Toll-like receptor
stimulation. It can be beneficial or harmful depending on whether it is local or systemic.
The gene encodes a number of well-characterized SNPs that have been investigated in
a number of diseases. The most commonly investigated SNP is the A/G –308 SNP.
The TNF-308A allele has been identified as being associated with increased TNF-α
production. Early studies found it was increased in most diseases studied. However,
interest in TNFA soon faded and then stopped when it was realized that the TNFA-
308A allele was in linkage on the HLA 8.1 haplotype. Interest in TNFA did, however,
spark a general interest in cytokine gene polymorphisms as potential potent immune-
regulatory genes.

6.9 A Single Gene or a Risk Portfolio

Though the MHC encodes 252 expressed genes, most studies have considered less than 14
of these: HLA-A, -B, -C, -DRB1, DRB3, -DRB4, -DRB5, DQA1, -DQB1, -DPA1, -DPB1,
MICA, C4*A, and TNFA. The clustering together of so many important immune-regula-
tory genes into some haplotypes may be of great significance in terms of human survival
against infectious disease.
One key question therefore is whether MHC-associated genetic susceptibility to disease
is determined by single genes, or by the collective activity of several or all the members
carried on extended haplotypes as part of our individual risk portfolios. We can consider
two possibilities.

A single gene may explain MHC-encoded genetic susceptibility to

disease
The MHC almost always seems to have a large risk impact. In all studies, the size of effect
depends on many factors, including the size of the population studied, the specificity of
the disease studied, and a number of other factors related to sample collection and testing.
Even so, the MHC seems set aside or different. The question is why.
Part of the answer to this question lies with considering the central role that is played by
the HLA molecules in T cell activation and in innate immunity. Interestingly, many of the
HLA associations described above were initially described as associations with the same
common ancestral haplotype. However, it is difficult to believe that one or two haplotypes
are the major determinants of genetic risk in the majority of human diseases. It is possible

2967_Ch06.indd 216 07/07/2015 11:57

 A Single Gene or a Risk Portfolio 217

that these links to common haplotypes reflect the fact that each carries more than one
potential risk allele. However, this does not suggest that the disease results from the effect
of several genes; on the contrary, it suggests perhaps a range of diseases are associated
with the same haplotype because the haplotype encodes several risk loci. For example, an
allele that is associated with an increased risk of SLE may be associated with AIH through
carriage on the common 8.1 haplotype. In SLE, it is the C4A gene that is thought to
be important in disease pathogenesis, whereas in AIH it is thought to be DRβ-71 and
DRβ-74. In this hypothesis the association between AIH and C4A*Q0 simply reflects
linkage disequilibrium in the 8.1 haplotype, and the real disease risk is explained by the
association with DRB1 alleles, which carry lysine (and arginine) at DRβ-71. This suggests
that the association with C4A*Q0 may be false and due simply to linkage disequilibrium.
In SLE, the situation is reversed—the association with HLA genes on the 8.1 haplotype
may be simply due to linkage disequilibrium with the C4AQ0 allele.

Alternatively there is always room for a second bite of the cherry:

a multihit hypothesis
In contrast to the hypothesis above, the magnitude of the disease risk encoded within
the MHC may be due to the clustering of so many highly important immune response
genes within one area (Figure 6.19). Here, a single allele may not be enough to explain
the MHC-associated risk. Instead, we may consider the possibility that within these
extended haplotypes there is a cumulative effect on disease risk. Thus, patients with AIH
or PSC who have the ancestral 8.1 haplotype have inherited a particular risk portfolio that
includes several independent risk alleles. This multihit hypothesis offers an explanation

HLA class 2

TNFA Complement

Disease
risk

Non-MHC HLA class 1

genes

MICA

Figure 6.19: Multihit hypothesis. The figure shows the potential for multiple genes to increase disease
susceptibility. Here the figure shows five different MHC gene families or genes. However, these are not the only
risk genes for the traits discussed in this chapter. There is an indication that genes may interact, e.g. TNFA and
MHC class II. Thus, it is possible that polymorphisms only increase disease risk when other alleles are present
or overexpressed. Finally, the figure also allows for redundancy. In this case, the complement system is used
as an example. The gray arrow indicates that even in the presence of complement gene-encoded risk alleles,
possession of such an allele may not have an effect unless the biological system that the gene products serve is
stressed. Under such conditions having a null allele for C4A, for example, may lead to failure to produce adequate
complement levels and this may have downstream consequences in terms of immune complex clearance, etc..

2967_Ch06.indd 217 07/07/2015 11:57

 218 CHAPTER 6 The Major Histocompatibility Complex

of why certain haplotypes are associated with several diseases. For example, the extended
8.1 haplotype includes HLA-A*01–C*07:01–B*08–MICA*008–TNFA-308**–C4A*Q0–
C4B*1–C2*C–Bf*S–DRB3*01:01–DRB1*03:01–DAQ1*05:01–DQB1*02:01, one or any
combination of these individual alleles may increase or reduce a particular disease risk.
A combination effect is partly illustrated by the example of AIH above. This possibil-
ity may explain the predominance of this haplotype in autoimmune disease. Of course,
this is a common haplotype and most carriers do not develop autoimmune diseases, but
possession of this group of risk alleles under immune stress may be sufficient to tip the
balance in the wrong direction. The converse is also possible under certain circumstances,
for example during a major epidemic it may be beneficial to have this haplotype because
possession of the haplotype may result in a mild response to the infectious agent. This may
explain the high frequency of some haplotypes in the population.
According to this explanation there are multiple risk alleles within the MHC. Each hap-
lotype will carry multiple hits, but the same haplotype does not necessarily operate in the
same way in each disease. The same DR, DQ, and C4 polymorphisms may impact on
different diseases differently, each making a large, small, or no contribution to the risk
portfolio. The multihit model can be exclusive.

6.10 How to Compare and Critically Evaluate

Contrasting Studies
History tells us about our past and informs our future. Only when we know our past can
we consider our future. This is as true in science as it is in global politics. We learn from
each publication. Most scientists are focused on narrow areas, developing hypotheses and
research plans based on the findings of previous studies. Thus, the scientific past also
informs the scientific present and the future.

Knowing history is important when we critically review and design

studies
PSC provides an interesting example to consider. PSC is relatively rare. Consequently
there are few centers that have studied the genetic basis of PSC, and even less that have
considered HLA or the MHC. Most of the UK studies have taken place at one center,
often in collaboration with other centers, but nevertheless almost always as a single unit.
This means that with time, the studies that have been performed, have benefited from an
ever-increasing DNA bank as well as new technologies that have become available. The
same is true for the Norwegian series, albeit that there is strong collaboration with a major
Swedish center, this group can still be considered as a single unit. In many ways there is
great benefit in this because this evolution within the research units ensures that technical
advances are applied to the study series and also it ensures a greater degree of consistency
in the studies.
However, it is also a potential trap unless one keeps a critical eye on one’s own work as well
as that of others. This is because when reviewing and comparing studies of genetic associa-
tions in disease it is all too easy to consider earlier studies as incorrect or flawed compared
with more recent studies. However, this is not always appropriate. We are inclined to be
apologetic about early studies, but there is no need for apologies—early analyses were
most often based on the best available methods for the time. As methods have improved

2967_Ch06.indd 218 07/07/2015 11:57

 Conclusions 219

it has become possible to increase the level of resolution and to focus on areas previously
ignored, allowing hitherto undetected, often undetectable relationships to be identified.
Therefore in studies of the MHC we often see the focus of research move from one loca-
tion to another. We must remember that had there been no initial interest in HLA, these
associations would not have been identified and perhaps in many cases the idea that these
complex diseases would be worthy of genetic investigation would have gone unmarked.
Studies of HLA in disease go back at least as far as 1967, and flowered in the 1970s and
1980s. These were periods during which even with poor technology, great advances were
made in a range of diseases and these were the stepping stones on which present-day
advances in the MHC and elsewhere began.

Conclusions
This chapter contains many key messages for the student of the genetics of complex dis-
eases. The early history of the MHC illustrates the importance of having standardized
nomenclature for complex systems. This will be of increasing importance in the post-
genome era with high-resolution genotyping being applied to more and more genes, creat-
ing a potential for personalized naming systems to arise.
The MHC is very complex with extreme levels of linkage disequilibrium and polymor-
phism. The high levels of linkage disequilibrium can be both helpful and problematic. In
particular, problems arise when data based around a single gene are over-interpreted. The
MHC illustrates the importance of studying whole haplotypes and not focusing on single
genes until there is sufficient evidence to be confident that the candidate has been identi-
fied. However, systems interact and genetic interaction on extended haplotypes can turn
the most carefully formulated hypotheses into nonsense.
In a more positive frame of mind, the examples discussed enable us to see the link between
genotype and phenotype, and the way genetics informs the debate on disease pathogenesis
at the molecular level. The key among these is the role of HLA in the formation of the
immune synapse—a process that is essential in adaptive immunity.
Though a key example, it would be wrong to consider the formation of the T cell synapse
as the only immune process influenced by MHC genes. HLA alleles have important roles
in innate immunity through NK cell activation and regulation.
Genetic associations with the MHC are not all the result of HLA polymorphism. This large
region encodes many immune-regulatory genes with diverse functions. Polymorphism in
any of these non-HLA immune-regulatory genes has the potential to impact on disease
susceptibility. For example, complement is a key element in immune complex clearance
and bacterial destruction, MICA genes impact on innate immunity, and the list goes on.
Finally, the examples discussed in this chapter also allow us to consider the difficulties
and complexity of assessing genetic studies in this area, and how we can approach these
subjects with a critical mind.
In summary, the MHC is a large densely packed genetic region within the human genome.
The genes within this region are particularly important in immune regulation, and there
are very strong links between polymorphisms in these genes and many complex diseases.
The region has been the focus for a considerable amount of research; however, much
remains to be done and, despite some very promising findings and hypotheses, the link
between genotype and phenotype is yet to be fully established. Interestingly, with current

2967_Ch06.indd 219 07/07/2015 11:57

 220 CHAPTER 6 The Major Histocompatibility Complex

technologies we are beginning to unpick the MHC to a greater extent than before and
identify multiple associations within the region. For example, the WTCCC2 study in
2011 suggested that there may be an independent association with an HLA-A allele that
confers some protection from multiple sclerosis. These technologies will be discussed in
more detail later in the book.

2967_Ch06.indd 220 07/07/2015 11:57

 Further Reading 221

Further Reading
Books/Theses the original report of 1984 IHTW identifying
Donaldson PT. The influence of donor recipi- HLA DQ and DP for the first time.
ent histocompatibility and immunogenetics on Brown JH, Jardetzky TS, Gorga JC et al. (1993)
the outcome of clinical liver transplants. PhD Three dimensional structure of the human
Thesis, Faculty of Clinical Medical Sciences, class II histocompatibility antigen HLA-DR1.
King’s College London. 1995. This thesis gives Nature 364:33–39.
a very thorough history of the development of Donaldson PT (2011) Electrostatic modifica-
transplantation from 1829, with Blundell's first tions of the human leucocyte antigen DR P9
use of blood transfusion, up to 1995. peptide-binding pocket primary sclerosing
Hillert J & Fogdell-Hahn A (2000) HLA and cholangitis: Back to the future with human
neurological diseases. In HLA in Health and leucocyte antigen DRβ. Hepatology 53:1798–
Disease (Lechler R and Warrens A eds), pp 1800. This is the Editorial for the paper by Hov
219–230. Academic Press. This chapter is full et al. (2011) written by one of the authors of
of data on HLA, and both multiple sclerosis this book. The Editorial explains the history of
and narcolepsy. HLA and PSC.
Parham P (2009) The Immune System, Donaldson PT & Norris S (2002) Evaluation of
3rd ed. Garland Science. This is an excel- the role of MHC class II alleles, haplotypes and
lent basic immunology book for the student selected amino acid sequences in primary scle-
of human immunology including links to rosing cholangitis. Autoimmunity 35:555–564.
immunogenetics. Han F, Faraco J, Dong XS et  al. (2013)
Van Rood JJ (2000) The history of the discov- Genome wide analysis of narcolepsy in China
ery of HLA. In HLA in Health and Disease implicates novel immune loci and reveals
(Lechler R & Warrens A eds), pp 3–22. changes in association prior to versus after the
Academic Press. This is an excellent overview 2009 H1N1 influenza pandemic. PLoS Genet 9
of the early days of HLA, and the flowering of (10):e1003880.
HLA nomenclature and typing methods, by Horton R, Wilming L, Rand V et  al. (2004)
one of its founding fathers. Gene map of the extended human MHC Nat
Rev Genet 5:889–899.
Articles Hov JR, Kosmoliaptsis V, Traherne JA et  al.
Arnett KL, Haung W, Valiante NM et  al. (2011) Electrostatic modifications of the
(1998) The Bw4/Bw6 difference between HLA- human leukocyte antigen-DR P9 peptide-
B*08:02 and HLA-B*08:01 changes the pep- binding pocket and susceptibility to primary
tides endogenously bound and the stimulation sclerosing cholangitis. Hepatology 53:1967–
of alloreactive T cells. Immunogenetics 48:56–61. 1976. This is an excellent article that discusses
the past and present interpretations of HLA
Bauer KH (2007) Homiotransplantation von associations with PSC, introducing some very
epidermis bei eineiigen zwillingen. Bruns Beit up-to-date methods for data analysis and pro-
Klin Chir 141:442–447. This is an example of viding a new model to account for these genetic
the early work on histocompatibility and raises associations.
awareness that not all papers were published in
English. Landsteiner K (1900) Zur kenntniss der anti-
fermentativen lytischen und agglutinieren-
Blundell J (1829) Observations on transfusion
den. Wirkungen blutserums und der lymphe.
of blood. Lancet 12:321–324. A fascinating
Zbl Bakt 27:357–362. Once again this paper
insight into very early clinical investigation.
reminds the student that not all science was and
Bjorkman PJ, Saper MA, Samraoui B et  al. is published in English. We should endeavor to
(1987) Structure of the human class I his- be aware of this especially when we consider
tocompatibility antigen, HLA-A2. Nature that Landsteiner was awarded a Nobel Prize for
329:506–512. his work.
Bodmer WF, Albert E, Bodmer JG et  al. Liu JZ, Almarri MA, Gaffney DJ et al. (2012)
(1985). Nomenclature for factors of the HLA Dense fine-mapping study identifies new
system, 1984. WHO Bull 63:399–405. This is

2967_Ch06.indd 221 07/07/2015 11:57

 222 CHAPTER 6 The Major Histocompatibility Complex

susceptibility loci for primary biliary cirrhosis. mechanisms in multiple sclerosis. Nature
Nat Genet 44:1137–1141. 476:214–219.
Liu JZ, Hov JR, Folseaas T et al. (2013) Dense Todd JA, Bell JI & McDevitt HO (1987)
genotyping of immune-related disease regions HLA-DQβ gene contributes to susceptibility
identifies nine new risk loci for primary scleros- and resistance to insulin-dependent diabetes
ing cholangitis. Nat Genet 45:670–675. mellitus. Nature 329:599–604.
Norris S, Kondeatis E, Collins R et al. (2001) Wiencke K, Spurkland A, Schrumpf E &
Mapping MHC-encoded susceptibility and Boberg KM (2001) Primary sclerosing chol-
resistance in primary sclerosing cholangitis: The angitis is associated to an extended B8-DR3
role of MICA polymorphism. Gastroenterology haplotype including particular MICA and
120:1475–1482. MICB alleles. Hepatology 34:625–630. The
Siebold C, Hansen BE, Wyer JR et al. (2004) study by Norris et al., above and this study by
Crystal structure of HLA-DQ0602 that Wiencke et  al., show slightly different results.
protects against type 1 diabetes and confers This is common in association studies between
strong susceptibility to narcolepsy. Proc Natl competing centers. Confounding results may
Acad Sci USA 101:1999–2004. This original be based on many factors. In this example the
paper compares two diseases and discusses studies are both based on similar-sized patient
complex molecular models to explain the pools, but use different methods and different
HLA associations with them. Not for the populations. This is an example that illustrates
faint-hearted, but a great paper that requires many of the concepts of studies in complex dis-
some consideration. eases, and together these two papers provide an
excellent basis for critical appraisal.
Terasaki PI & McClelland JD (1964)
Microdroplet assay of human serum cytotox-
ins. Nature 204:998–1000. This original report Online sources
opened the door to the use of micro assays for
HLA typing—a method almost universally http://hla.alleles.org/antigens/index.html
adopted for HLA class I typing in the 1970s This is an excellent source for up-to-date infor-
and 1980s until molecular genetics techniques mation on HLA nomenclature.
became available. http://www.ebi.ac.uk/imgt/hla/stats.html
The International Multiple Sclerosis Genetics This is an excellent source for up-to-date infor-
Consortium & The Wellcome Trust Case mation on HLA allele numbers and nomencla-
Control Consortium 2 (2011) Genetic risk ture. The website includes text on the history of
and a primary role for cell-mediated immune HLA nomenclature.

2967_Ch06.indd 222 07/07/2015 11:57

 CHAPTER
7
Genetics of Infectious Disease

Infectious diseases account for the majority of morbidity and mortality in human popula-
tions. In a classical sense, we would not normally consider infectious diseases as genetic
diseases. However, variation in the human genome can lead to variation in resistance and
susceptibility to infectious disease, and in this sense it is correct to say that there is a genetic
component in infectious disease. In this chapter, we will consider how the evolutionary
pressure of infectious disease has led to particular genetic variants being maintained in a
human population and will examine molecular mechanisms that link particular polymor-
phisms with susceptibility to infection. Two key examples, malaria and human immuno-
deficiency virus type 1 (HIV-1), will be used to illustrate how different clinical aspects of
infectious disease are influenced by host genetic variation. In addition, we will discuss the
use of hypothesis-driven and genome-wide studies to identify specific human genes that
influence the severity of infectious disease.

7.1 THE INFECTION PROCESS AND DISEASE

Before embarking on a discussion of the genetics of infectious disease it is worth briefly
reviewing the infection process. Infectious diseases occur when a susceptible host is
exposed to and infected by a pathogenic microorganism. We can divide these microor-
ganisms broadly into five types: viruses, bacteria, fungi, and unicellular or multicellular
eukaryotic parasites. Viruses, as well as some bacteria and unicellular parasites, are intra-
cellular pathogens, meaning that at least a part of their life cycle takes place inside the host
cells. The remainder are extracellular pathogens and colonize the extracellular environ-
ment either inside the host organism or on its surface.

2967_Ch07.indd 223 07/07/2015 10:44

 224 CHAPTER 7 Genetics of Infectious Disease

Mechanisms of infection vary widely but common steps in the process

can be identified
Though the mechanisms through which infectious pathogens cause disease vary, there are
several key steps in this process (Figure 7.1). First, an individual must be exposed to the
pathogen, and the likelihood of this depends on a number of factors including the health
status of the individual, the behavior of the individual, route of transmission of the patho-
gen, population size and density, sanitation, and healthcare. Once exposed, the pathogen
must be able to propagate and colonize the host. This may take place only at the site of
infection, leading to a localized infection, or alternatively the pathogen may spread from
the initial site of infection via the blood and lymphatic system to infect other organs and
generate a systemic infection.

The immune response combats infectious disease

The efficiency with which a pathogen is able to establish an infection is strongly influenced
by the counter-effect of the host’s immune response. The innate and adaptive arms of the
immune response are both deployed in an attempt to resist infection and to eliminate the
pathogen from the body. The majority of infections are asymptomatic, usually because
the host immune response is successful in quickly containing and clearing the infection
before symptoms appear. However, not all infections are dealt with so efficiently and those
that are not eradicated may result in disease symptoms of varying severity. In some cases, a
pathogen may establish a persistent infection, continuing to live in the host for many years
with only mild or even no symptoms. It is worth noting that many common symptoms
of infection, e.g. fever and tissue damage, may not be a result of damage induced by the
pathogen, but be the direct effects of an active host immune response.

Response to
treatment

Disease outcome
Exposure Infection Pathogenesis • Symptom-free, uninfected
• Symptom-free, carrier
• Local/systemic infection
• Mild/severe disease
Pathogen
• Acute/chronic disease
• Alternative symptoms
Host Multiplication
• Death
of pathogen

Figure 7.1: The infection process. All infectious diseases follow a similar process. First, the host must be
exposed to the pathogen that may then infect the host, replicating either on host cell surfaces or inside cells. The
infection process often requires the pathogen to bind to a host cell receptor protein. The outcome of infection
varies enormously, depending upon the specific interaction between host and pathogen, and ranges from
asymptomatic infection to mild disease, severe disease, and death. Treatment of infection (either before or after
exposure) may inhibit the infection process at one or more steps in the pathway, and there is significant variation
in the response of the patient and the pathogen to treatment.

2967_Ch07.indd 224 07/07/2015 10:44

 Heritability of Resistance and Susceptibility to Infectious Disease 225

Individuals infected by the same pathogen may experience different

outcomes
The symptoms that result from an infection depend to a large extent on the phenotype of
the pathogen. More interestingly, in the context of this book, the outcome of infection
can also vary among different individuals infected by the same pathogen. We will briefly
consider two important key examples that illustrate this point:
• HIV-1 is a retrovirus that causes acquired immunodeficiency syndrome (AIDS).
HIV-1 is a major global health threat and was responsible for up to 1.6 million deaths
in 2012 according to the World Health Organization (WHO). Most people who are
repeatedly exposed to HIV-1 through, for example, unprotected sexual intercourse
will become infected and go on to develop AIDS within approximately 10 years. In
December 2013, there were an estimated 35.3 million cases of HIV-1 infection world-
wide. Some individuals, however, are repeatedly exposed to HIV-1 in the same way,
but never become infected.
• Malaria is another global killer responsible for an estimated 627,000 (range 473,000–
789,000) deaths in 2012, mostly among African children. Malaria is caused by
eukaryotic parasites (protozoa) of the Plasmodium genus. The total number of exposed
individuals exceeded 3.4 billion with 207 million infected cases in 2012. Some indi-
viduals infected with the Plasmodium parasite experience no symptoms, whereas some
experience severe malarial anemia and some develop cerebral malaria.
It is clear from these two examples that pathogens can have very different effects in dif-
ferent individuals. Non-genetic factors, including age, living conditions, health status,
and presence of other infections, may contribute to the range of outcomes. Some of
this diversity also arises due to novel variations in the genome of the pathogen. Most
microorganisms multiply quickly and in doing so accumulate mutations. Viruses with an
RNA genome accumulate mutations rapidly because RNA polymerases lack proofreading
ability. Variation in the pathogen genome may increase or decrease the pathogenicity of
the microorganism and any mutations that confer a survival advantage will be quickly
selected. Non-genetic factors and genetic variation in the pathogen genome are, however,
by no means the whole story and cannot account for all observed variation in the human
response to infectious disease. We must also consider the effect of variation in the genome
of the human host, which is the basis of many differences in individual susceptibility and
resistance to infectious disease.

7.2 Heritability of Resistance and

Susceptibility to Infectious Disease
What evidence is there that resistance and susceptibility to infectious disease is heritable?
Even before the development of modern techniques for molecular genetic analysis, histori-
cal and epidemiological studies provided clues that genetic variation in human popula-
tions affected their response to infection.

Different populations infected by the same pathogen may experience

different outcomes
As well as variation in individual responses to infection by a particular pathogen, variation
is evident at a population level. This is best illustrated by looking back in history, to a time

2967_Ch07.indd 225 07/07/2015 10:44

 226 CHAPTER 7 Genetics of Infectious Disease

when populations were more geographically isolated than they are today. In 1519, Spanish
Conquistadors led by Hernando Cortes landed in Mexico. Unknown to the Conquistadors,
they were taking with them a number of infectious agents, including measles, smallpox,
and the influenza viruses. They also carried bacteria, including Rickettsia, which can cause
typhus. The invading populations were relatively resistant to these pathogens, whereas the
indigenous South American populations, including the Mexican Aztecs, were highly sus-
ceptible to these infectious agents and died in vast numbers as a result. For example, small-
pox became epidemic in Mexico between 1520 and 1521, and may have killed an estimated
10–50% of the local population. Several factors are likely to have contributed to the dif-
ference in susceptibility between the invading and indigenous populations. The invaders
had already been exposed to these pathogens and would therefore have acquired immune
memory. In contrast, the indigenous populations who had not previously been exposed to
these pathogens had no acquired immune memory, making them more vulnerable to infec-
tion. However, host genetic differences between the two populations may also have played
a significant role. Pathogens such as smallpox had been common among European popula-
tions for many years, and individuals with allelic variants conferring resistance to smallpox
and other common pathogens would have had a survival advantage over those without such
alleles. As a consequence of this selective pressure, resistance alleles may have been present
at a higher frequency in the invading populations than in South American populations,
where no such selection pressure had occurred. Thus, a difference in the relative frequency
of resistance-associated alleles in the two populations may have contributed to the greater
susceptibility of the South American people to the imported pathogens.
There are many other examples in history of naïve populations being devastated by infec-
tious diseases brought to them by immigrant populations. Haldane, in 1949, suggested
that “Europeans have used their genetic resistance to such viruses as that of measles (rube-
ola) as a weapon against primitive peoples as effective as fire-arms”. Yet the reverse rule also
applies: Europeans, migrating to colonial regions, have found themselves more susceptible
to local infectious agents and in some cases early colonies were wiped out by exposure to
infectious disease.

Leprosy and tuberculosis were once believed to be inherited diseases

Leprosy and tuberculosis are diseases that we now know to be caused by infection with the
intracellular bacteria Mycobacterium leprae and Mycobacterium tuberculosis, respectively.
The distinct clustering of leprosy and tuberculosis in families, however, resulted in an early
assumption that both diseases were inherited. Gerhard Hansen challenged this view of
leprosy in 1873, when he identified M. leprae as the causative agent. At the time this was
the first bacterial agent of human disease to be identified. Hansen stated “your opinions
about leprosy are completely wrong. You believe that the disease is hereditary but not
infectious. The truth is that it is infectious but not hereditary.” This was followed by the
identification of M. tuberculosis by Robert Koch in 1882.
More recently, twin studies indicated that concordance rates of leprosy and tuberculo-
sis were higher in monozygotic twins than dizygotic twins, though the accuracy of the
tuberculosis study has been questioned. Although leprosy and tuberculosis are infectious
diseases, the familial clustering and concordance rates still provide a strong indication
that susceptibility to both diseases has a heritable component that is likely to include
host genetic variation. This suggests that Hansen was not entirely correct and this may be
why some thought these were heritable diseases. Altogether, this reminds us not to look
for simple answers to complex questions. Twin studies of susceptibility to polio, febrile
malaria, and hepatitis B virus carrier status, among others, add further support to the

2967_Ch07.indd 226 07/07/2015 10:44

 Heritability of Resistance and Susceptibility to Infectious Disease 227

hypothesis that susceptibility to infectious disease is at least partially determined by the

genetic makeup of the host.

Adoption studies indicate that susceptibility to infectious disease has

a heritable component
A study of Danish adoptees revealed that their relative risk of death due to infectious disease
was six times greater if their biological parent died of infectious disease at an early age (less
than 50 years old) compared with those whose biological parent was alive at 50 years old.
No increased risk was observed for the adoptee if the adopted parent succumbed to infec-
tious disease. A follow-up study revealed that the odds ratio (OR) for death of a full sibling
due to infection was greater than 9, compared with half siblings or adopted siblings, further
strengthening the case for a genetically determined susceptibility to infection.

Rare monogenic defects in immunity can cause primary immune

deficiencies
The ability to resist infection is dramatically impaired in some individuals, due to mono-
genic defects in genes involved in the immune response. More than 200 monogenic disor-
ders of this type, known as primary immune deficiencies, have been identified. Though
these are not the subject of this book, these Mendelian disorders do illustrate an important
concept. Depending on the specific mutation, a primary immunodeficiency may result in
an increased susceptibility to infectious disease in general, to a particular subset of infec-
tions, or to a single pathogen (Table 7.1).
Most primary immune deficiencies are rare Mendelian recessive traits occurring in less
than 1:50,000 births and, as such, these disorders are beyond the scope of this book,
which is about common genetically complex disease. It is, however, important to appreci-
ate the existence of these primary immune deficiencies, firstly in order to have a full pic-
ture of how the host genome influences the outcome of infection, and secondly because
more common polymorphisms in the same genes, with lower penetrance, may have a
more subtle and complex effect on the host response to infection.

Table 7.1 Selected examples of rare monogenic disorders that cause defects in the immune response
leading to increased susceptibility to infectious disease.

Name of syndrome Immune defect Susceptibility

Severe combined No T cells (and in some cases General
immunodeficiency B cells)
(Bruton’s) X-linked No mature B cells, therefore Extracellular bacteria and
agammaglobulinemia (XLA) no immunoglobulin viruses, especially pyogenic
bacteria and enteric viruses
Mendelian susceptibility Defects in IL-12– Mycobacterial infection
to mycobacterial diseases IFN-γ signaling circuit
(MSMD) (macrophage–T
cell–macrophage)
Epidermodysplasia Mutations in EVER1 and Human papilloma virus
verruciformis (disseminated EVER2
cutaneous warts)

2967_Ch07.indd 227 07/07/2015 10:44

 228 CHAPTER 7 Genetics of Infectious Disease

7.3 Identifying Alleles That Affect Risk of

Susceptibility and Resistance to Infectious
Disease
In complex disease we are dealing with elevations and reductions in risk rather than abso-
lute values. Having a risk allele is neither sufficient nor necessary to develop the disease.
Nowhere is this more obvious than in infectious disease where the primary determinant of
disease is the infectious agent itself.

Risk alleles can be identified using a hypothesis-driven or genome-

wide approach
As for complex diseases described previously in this book, there are broadly two strate-
gies for identifying alleles that modify the risk of susceptibility or resistance to an infec-
tious disease. The first is to adopt a hypothesis-driven approach (sometimes referred to
as a candidate gene approach) in which existing knowledge of the infectious process is
used to select candidate genes that are predicted to influence susceptibility to infectious
disease. Genes involved in the immune response to a pathogen are obvious candidate
genes. Hypothesis-driven studies have successfully identified a number of genetic vari-
ants associated with resistance and susceptibility to infectious disease, some of which are
discussed below. These candidate gene studies are, however, constrained by the limits of
current knowledge of disease pathogenesis, and have often proved difficult to replicate for
a variety of reasons and have tended to involve small study sizes with consequently low
statistical power.
The second strategy is to take a genome-wide approach. This usually involves either
genome-wide linkage studies (GWLS) or genome-wide association studies (GWAS), tech-
niques which are described in Chapter 3. The advantage of genome-wide approaches is
that they are not hypothesis constrained and do not rely on prior knowledge. As a conse-
quence, they have the potential to identify novel genetic variants that were not previously
known to play a role in the infectious process. Genome-wide approaches can thus open
up new avenues for research into infectious disease, but do require large numbers to reach
adequate levels of statistical power.
Studies on infectious disease resistance/susceptibility alleles have been commonly based
on a hypothesis-driven approach. The reasons for this are two-fold. (1) There are some
clear candidate genes to target based on a general knowledge of the infectious process,
including genes that encode receptor proteins and genes that are involved in the immune
response to infection. (2) Genome-wide studies require large numbers of well-phenotyped
infected individuals, which can be difficult to collect in developing countries where infec-
tious diseases are often most prevalent. Genome-wide studies in Africa, where infectious
disease is a significant burden, are particularly challenging. African populations are often
more genetically diverse than other populations and linkage disequilibrium can be less
prominent, leading to a number of practical difficulties. For example, false-positive asso-
ciations may arise as a result of population stratification and multicenter studies may be
compromised by differences within the haplotypes in different participating communities.
Improvements in high-throughput genotyping technology have led to increased use of
genome-wide association studies, in which thousands of tagged single nucleotide poly-
morphism (SNPs) and other genetic markers across the entire genome are tested for

2967_Ch07.indd 228 07/07/2015 10:44

 Malaria 229

association with a specific infectious disease. This has allowed researchers to cast a much
wider net in the search for relevant alleles.

The outcome of infectious disease being tested must be

clearly defined
As in all studies of complex disease genetics it is important when searching for risk alleles
that influence resistance and susceptibility to infectious disease, that the phenotype (out-
come) of the disease being tested for association is clearly defined. For example, is the
study testing for association with infection per se or for a specific clinical outcome in an
infected population? Lack of clear definition of the parameters of the groups being tested
may lead to important effects being masked, and will also present significant difficulties
when combining results in collaborative studies and meta-analyses. This is illustrated by
studies of malaria resistance, which may focus on either association with infection per se
or on disease severity following infection. Disease severity may be defined simply by hos-
pitalization or by more stringent clinical parameters. It is also important that the method
of testing the phenotype is clear. For example, in studies of HIV-1 infection, infection per
se may be defined by polymerase chain reaction (PCR) detection of the viral genome or
seroconversion and studies on the speed of progression to AIDS may define progression
based on CD4+ T cell count or viral load (measured as viral genome copies per milliliter).
Using different parameters in different studies to measure these traits makes direct com-
parison of different studies impossible.

7.4 Malaria
Malaria is caused by protozoa of the Plasmodium genus, comprised of four species:
Plasmodium falciparum, Plasmodium vivax, Plasmodium ovale, and Plasmodium malariae.
Of these P. falciparum and P. vivax are the most common, with P. falciparum causing more
severe disease than P. vivax. The Plasmodium parasite is carried by the female Anopheles
mosquito, which thrives in wet, humid areas.

The life cycle of the Plasmodium protozoa is complex

When bitten by a malaria-infected mosquito, sporozoites in the salivary gland of the mos-
quito are transmitted into the bloodstream of human hosts. On entering the bloodstream
these sporozoites travel to the liver. In the liver they mature and are released back into
the bloodstream as merozoites, which infect and destroy the host red blood cells (eryth-
rocytes). Inside erythrocytes the parasites differentiate into male and female gametocytes
that are taken up once more by a mosquito. The sexual stage of the parasite reproduction
takes place within the mosquito, culminating in the production of new sporozoites and
completing the life cycle (see Figure 7.2).
Symptoms of malaria typically include fever, chills, and anemia. These coincide with the
destruction of red blood cells by the parasite, which leads to loss of hemoglobin and pro-
duction of a surge of cytokines, including tumor necrosis factor (TNF)-α and interleukin
(IL)-1, which trigger the fever and chills. Severe malaria in children is often characterized
by severe anemia, respiratory distress (due to lactic acidosis), and neurological symptoms,
which may culminate in coma (cerebral malaria). In adults, severe symptoms include
organ failure, particularly renal failure.

2967_Ch07.indd 229 07/07/2015 10:44

 230 CHAPTER 7 Genetics of Infectious Disease

Gametocytes
ingested with Feeding mosquito
blood meal injects sporozoites and
ingests gametocytes
Exflagellation to
form male gametes

Fertilization Gametocytes
Sporozoites
migrate to
Zygote Sporozoites
salivary glands
enter blood,
via hemocoel
penetrate
Sporozoites liver cell
Female released
gamete
Oocyst undergoes
Liver stage
sporogony
Merozoites
Oocyst forms released as
Merozoites erythrocyte
released is lysed

Ookinete penetrates
gut epithelium Asexual
blood cycle
Motile ookinete formed

Figure 7.2: The Plasmodium life cycle. Plasmodium parasites that cause malaria have a complex life cycle
involving a mosquito insect vector. When an infected mosquito bites a human host, Plasmodium sporozoites may
be introduced into the blood and travel to the liver. Here, the Plasmodium enters the next phase in its life cycle,
producing merozoites that infect red blood cells. Gametocytes are released from the red blood cells and can be
ingested when the host is bitten by another mosquito. The sexual phase of the Plasmodium life cycle takes place
in the gut of the mosquito and ultimately new sporozoites are produced. These travel to the salivary glands of the
mosquito from where they can be transmitted to the next host. (From Loker ES & Hofkin BV [2015] Parasitology.
Garland Science.)

Hemoglobinopathies confer resistance to malaria

Mutations in the genes for hemoglobin were the first specific polymorphisms recognized
to be associated with resistance to an infectious disease. Hemoglobin is the protein that
packs erythrocytes and carries oxygen around the body (Figure 7.3). The most common
form of hemoglobin is hemoglobin A (HbA) which is a tetramer of four polypeptides,
consisting of two α-globin chains and two β-globin chains, each with a heme prosthetic
group that is able to bind oxygen. Hemoglobinopathies, which include sickle cell anemia
and thalassemias, result when mutations in the globin genes lead to changes in hemoglo-
bin structure or expression levels.
Sickle cell anemia
Sickle cell anemia is caused by a SNP in the β-globin gene, leading to a substitution of
valine instead of glutamate at amino acid 6 of the β-globin molecule. This causes polym-
erization of the hemoglobin at low oxygen concentrations, which in turn leads to the
affected erythrocytes developing a characteristic sickle shape. Individuals who are homo-
zygous for the sickle cell allele (HbS) suffer from debilitating symptoms associated with

2967_Ch07.indd 230 07/07/2015 10:44

 Malaria 231

β1 α1

α2 β2

Figure 7.3: Structure of hemoglobin. Hemoglobin is a tetrameric protein composed of four polypeptide units
that is found in red blood cells and is responsible for the transport of oxygen in the blood. The most common
form of hemoglobin in adults is HbA, which consists of two α polypeptides and two β polypeptides. Each of these
polypeptides forms a complex structure with an iron-containing heme prosthetic group, which is able to bind one
oxygen molecule (shown above). Diseases caused by mutations that affect the structure or expression levels of
hemoglobin polypeptides are called hemoglobinopathies.

sickle cell anemia. Heterozygotes (HbAS), however, have sickle cell trait and suffer few if
any symptoms.
Thalassemia
Thalassemia is caused by mutations in globin genes that reduce the expression of α-globin
(α-thalassemia) or β-globin (β-thalassemia). There are two α-globin genes, closely linked
on chromosome 16, and therefore two types of α-thalassemia: α+-thalassemia occurs when
one α-globin gene is defective and α0-thalassemia occurs when both are defective. There
are five possible allele combinations for α-thalassemia, as illustrated in Table 7.2.
β-thalassemia is caused by defects in the β-globin gene. β-thalassemia heterozygotes (β/–)
suffer mild anemia and morphological changes to red blood cells, whereas β-thalassemia
homozygotes suffer severe anemia that can be fatal without treatment.

Table 7.2 Thalassemia: genotypes and associated phenotypes.

Genotype Phenotype
Normal homozygote (αα/αα) Normal
α+-thalassemia heterozygote (–α/αα) No symptoms
α+-thalassemia homozygote (–α/–α) Mild anemia and small red blood cells with reduced
hemoglobin
α0-thalassaemia heterozygote (– –/αα) Mild anemia
α -thalassasemia homozygote (– –/– –)
0
Lethal

2967_Ch07.indd 231 07/07/2015 10:44

 232 CHAPTER 7 Genetics of Infectious Disease

Haldane’s malaria hypothesis proposed that thalassemia confers

protection against malaria
In 1949, the biologist and geneticist JBS Haldane proposed that resistance to infectious
disease is an important driving force in the evolution by natural selection of the human
genome. Haldane proposed that an important mechanism in the process is what we now
call heterozygote advantage, whereby the heterozygote is fitter than either homozygote. He
went on to suggest that the high frequency of thalassemia observed in Mediterranean coun-
tries such as Greece and Italy could be linked to the prevalence of malaria in those coun-
tries. Knowledge of the Plasmodium life cycle indicated that erythrocytes could be the link:
part of the Plasmodium life cycle takes place inside erythrocytes and thalassemia is caused
by defective production of the protein hemoglobin leading to erythrocyte abnormalities.
In what became known as the malaria hypothesis, Haldane proposed that erythrocytes of
individuals who were heterozygous for the thalassemia gene may be resistant to infection
by the Plasmodium parasite. In areas where malaria is endemic this would confer a survival
advantage, leading to the thalassemia allele being maintained at high frequencies in those
populations. The thalassemia heterozygote is therefore fitter than both the wild-type homo-
zygotes, i.e. those without the thalassemia mutation, who will lack the increased protection
against malaria, and the thalassemia homozygote, who suffers from severe or fatal anemia.

Allison demonstrated that sickle cell trait confers resistance to

P. falciparum
Haldane’s malaria hypothesis was at the time largely speculative, with no supporting
experimental evidence. Indeed, evidence for a link between thalassemia and resistance
to malaria only began to appear in the early 1970s, as discussed in the following section.
A link between malaria resistance and another hemoglobinopathy, sickle cell anemia, was
more quickly established.
In 1954, Allison recognized that the geographical distribution of sickle cell trait coin-
cided with that of malaria (Figure 7.4) and went on to demonstrate experimentally that

Percentage of
population
that has the
sickle cell allele
(HbS)
Endemic
>6 P. falciparum
2–6 malaria

Figure 7.4: Comparison of global distribution of the HbS allele and malaria. The global distribution of
the HbS sickle cell allele corresponds strikingly with the distribution of P. falciparum. This correlation was first
recorded in 1954 by Allison, who proposed that the HbS allele conferred resistance to P. falciparum and went on
to demonstrate that red blood cells from HbS heterozygotes were resistant to infection by P. falciparum. (From
Berg JM, Tymokczo JL & Stryer L [2007] Biochemistry, 6th ed. With permission from W. H. Freeman.)

2967_Ch07.indd 232 07/07/2015 10:44

 Malaria 233

the erythrocytes of individuals who are heterozygous for the sickle cell allele (HbAS) are
more resistant to infection by P. falciparum than wild-type (HbA) homozygotes. Allison
concluded that “those who are heterozygous for the sickle-cell gene will have a selective
advantage in regions where malaria is hyper-endemic.” His work provided strong evidence
to support the malaria hypothesis, although at the time Allison was unaware of Haldane’s
earlier work and had reached the same conclusions independently.
More recent studies on the distribution of sickle cell trait indicate that 9% of the popula-
tion carry the HbS allele in parts of Africa stretching from southern Ghana to northern
Zam­bia and a frequency of 18.25% was recorded in northern Angola. A case control study
in West Africa in 1991 indicated that children with the HbS allele had a 92% reduction in
the relative risk for severe malaria compared with those without the HbS allele.

Studies on Pacific Island populations provided experimental evidence

that thalassemia confers protection from malaria
Although Haldane proposed, as early as 1946, that thalassemias offered protection against
malaria, experimental evidence to support his hypothesis was slow to emerge. Studies
undertaken in the 1950s and 1960s to investigate the link between the two diseases were
inconclusive, partly because it was difficult, with the techniques available at the time, to
accurately diagnose the milder forms of thalassemia. Misdiagnosis of thalassemia (either
false positives or false negatives) is likely to have skewed the results of these early studies.
The advent of molecular diagnostic techniques in the 1970s allowed more accurate diag-
nosis and more convincing evidence began to emerge. For example, in 1997 the results
of a case control study of children in Papua New Guinea showed that the relative risk of
developing severe malaria was 0.4 for α+-thalassemia homozygotes when compared with
their non-thalassemic peers. The risk in α+-thalassemia heterozygotes was 0.7 compared
with those without the thalassemia mutation.

The mechanism of resistance to malaria conferred by

hemoglobinopathies is still not fully understood
Since the 1950s, numerous studies have investigated the mechanism for malaria resis-
tance associated with thalassemia and sickle cell trait, but no single clear mechanism has
emerged for either allele (see Table 7.3).
Both the innate and adaptive immune responses are important in destroying free
Plasmodium parasites and infected cells. In vitro studies indicate that Plasmodium-infected
erythrocytes from α-thalassemia and sickle cell patients may be more efficiently eliminated
by the immune response. A range of mechanisms have been proposed, including increased
phagocytosis, complement-mediated lysis, and antibody-dependent cytotoxicity.
A possible mechanism for Plasmodium resistance in sickle cell patients involves the
P. falciparum erythrocyte membrane protein-1 (PfEMP-1), a protein encoded by the
Plasmodium genome and expressed on the surface of infected erythrocytes. PfEMP-1
is an important antigen that is recognized as part of the antibody-mediated immune
response to Plasmodium. PfEMP-1 also mediates the binding of infected erythro-
cytes to nearby uninfected erythrocytes, forming rosettes, and to endothelial cells that
line the post-capillary venules. This cell-to-cell binding, known as cyto-adherence,
causes the infected cells to accumulate in the small blood vessels, thereby evading
immune clearance in the spleen and is often linked to development of severe malaria.
Plasmodium-infected HbAS erythrocytes express lower levels of PfEMP-1 on their
surface compared with infected erythrocytes from normal individuals; less PfEMP-1

2967_Ch07.indd 233 07/07/2015 10:44

 234 CHAPTER 7 Genetics of Infectious Disease

Table 7.3 Selected examples of hemoglobinopathies that confer susceptibility or resistance to

malaria and some of the proposed mechanisms of action.

Variant Protein affected Effect Proposed mechanisms

Hemoglobin S β-globin Reduced Reduced invasion of erythrocytes
(HbS variant (Glu6Val risk of severe by Plasmodium; reduced growth of
of HBB gene) substitution) malaria Plasmodium inside erythrocytes; enhanced
phagocytosis and clearance of infected
erythrocytes; reduced cyto-adherence of
infected erythrocytes
Hemoglobin β-globin Reduced Reduced cyto-adherence of infected
C (HbC (Glu6Lys risk of severe erythrocytes
variant of substitution) malaria
HBB gene)
Hemoglobin β-globin Reduced Reduced invasion of erythrocytes
E (HbE (Glu26Lys risk of severe by Plasmodium; reduced growth of
variant of substitution) malaria Plasmodium inside erythrocytes; enhanced
HBB gene) phagocytosis and clearance of infected
erythrocytes
α-thalassemia α-globin (reduced Reduced Reduced growth of Plasmodium inside
(variants of expression of risk of severe erythrocytes; increased antibody binding to
HBA1 and α-globin) malaria infected cells; reduced rosetting of infected
HBA2 genes) erythrocytes; increased susceptibility to
P. vivax in infancy, enhancing later immunity
to P. falciparum; increased number of small
erythrocytes means that less hemoglobin is
lost per infected erythrocyte
β-thalassemia β-globin (reduced Reduced Increased antibody binding to infected
(variants of expression of risk of severe cells; enhanced phagocytosis
HBB gene) β-globin) malaria

means less cyto-adherence, offering a clear mechanism for protection against severe
malaria. It has also been suggested that the polymerization of hemoglobin in HbAS
erythrocytes may limit the growth of the Plasmodium parasite within the cells.
Rather surprisingly, in 1996 Williams et al. found that α+-thalassemia children on the Pacific
island of Espiritu Santo were more likely to suffer from milder forms of malaria compared
with non-thalassemic children. The effect was particularly significant in younger children
and for malaria caused by P. vivax. A possible explanation for this apparent paradox is that
α+-thalassemia makes these children more susceptible to infection by the less dangerous P.
vivax at a very young age, when they are still protected by maternal antibodies, so that they
develop immune memory that may protect them from more severe malaria later in life.
Another rather counterintuitive suggestion is that the microcytic anemia experienced
by α+-thalassemic individuals may itself offer protection from more severe malarial ane-
mia. Malarial anemia is caused by destruction of Plasmodium-infected erythrocytes.
α+-thalassemic patients have a high number of small circulating erythrocytes, each with
a relatively low concentration of hemoglobin. The rationale for the proposed mechanism

2967_Ch07.indd 234 07/07/2015 10:44

 Malaria 235

is that less hemoglobin is lost when a microcytic erythrocyte is destroyed, compared with
destruction of a normal erythrocyte. The resulting malarial anemia is therefore less severe.

Resistance to malaria conferred by HbS and thalassemia is a complex

genetic trait
It is worth noting that sickle cell anemia and the thalassemias are diseases caused by reces-
sive alleles that follow a classical Mendelian pattern of inheritance. Resistance to malaria
conferred by HbS and thalassemia alleles is, however, a complex genetic trait—the alleles
increase the probability of resistance to malaria, but do not guarantee it. The polymor-
phisms that cause HbS and thalassemia therefore conform to the extended definition of
complex disease, based on that of Haines and Pericak-Vance (1998), as those where altera-
tions in more than one gene (allele) alone or in concert either increases or decreases the risk
of developing a trait (see Chapter 2)—the trait in this case being susceptibility to malaria.

Other malaria resistance alleles have been identified via

epidemiological or hypothesis-driven studies
In addition to the thalassemia and sickle cell alleles, a number of other polymorphisms
have been identified on the basis of epidemiological studies and existing knowledge of
malaria pathogenesis, examples of which are summarized in Table 7.4.

GWAS suggest that polymorphisms in immunity-related genes may

affect outcome of Plasmodium infection
GWAS have revealed significant linkage between the intensity of the outcome follow-
ing malaria infection and chromosomes 10p15.3–p14 and 13q. Suggested linkage has
also been identified for clinical malaria disease with chromosome 5p15 and chromosome
13q13, and for parasite density in asymptomatic infection with chromosome 5q31 and
chromosome 12q21. The specific genes involved have yet to be identified, but there is
some overlap between these regions and regions linked to asthma, suggesting that immune
related genes may be involved. The 5q31–q33 region on chromosome 5 includes many
genes involved in the immune response to infection, including those encoding granulo-
cyte colony stimulating factor (CSF2), colony stimulating factor 1 receptor (CSF-1R) and
the cytokines IL-3, -4, -5, and -9.

GWAS searching for malaria resistance alleles highlight the challenges

of GWAS in African populations
As discussed above, there are practical difficulties in carrying out GWAS in developing
countries where malaria is prevalent and particularly in African populations, which have
a relatively higher level of genetic diversity and lower level of linkage disequilibrium than
seen in other populations. In 2009, a GWAS study carried out by members of the Malaria
Genomic Epidemiology Network (MalariaGEN) addressed some of these difficulties. They
were able to test their methodology using the known association between HbS and malaria
resistance. If the methods were robust there should be a strong association between malarial
resistance and a SNP labeled rs334, which causes the critical substitution of valine instead
of glutamate in the β-globin chain. The initial study, on children from the Gambia, inves-
tigated 500,000 SNPs in 1060 children with severe malaria compared with 1500 controls.
Being aware that genetic variation in the African population could lead to false-positive
results in the study, the investigators first used principle components analysis (PCA) to
determine the population structure of the study subjects and found that the population

2967_Ch07.indd 235 07/07/2015 10:44

 236 CHAPTER 7 Genetics of Infectious Disease

Table 7.4 Selected examples of non-hemoglobin polymorphisms that confer susceptibility or

resistance to malaria and some of the proposed mechanisms of action.

Variant Protein affected Effect Proposed mechanisms

FY*O Duffy antigen (P. vivax Reduced FY*O Duffy-negative individuals
receptor protein) risk of severe express no Duffy antigen on the
malaria erythrocyte surface, therefore
P. vivax is unable to enter cell;
confers 100% protection against
P. vivax
G6PDH Glucose 6-phosphate Reduced Reduced growth of Plasmodium
dehydrogenase (enzyme risk of severe inside erythrocytes; increased cell
that protects against malaria damage due to oxidative stress;
oxidative stress), enhanced phagocytosis
reduced expression
CRI Complement receptor Reduced Reduced CR1 on surface of
1 (binds PfEMP-1; risk of severe uninfected erythrocytes leads to
mediates rosetting), malaria reduced rosetting
reduced expression
TNF-308A and TNF-α (pro- Increased TNF-α induces production of
TNF-376A inflammatory cytokine) risk of adhesion molecules and pro-
(promoter cerebral inflammatory cytokines; increased
polymorphisms) malaria levels of TNF-α implicated in
development of cerebral malaria
HLA-B*53 HLA-B (involved in Reduced Enhanced presentation of
antigen presentation to risk of severe Plasmodium antigens
cytotoxic CD8+ T cells) malaria

could be divided into four distinct genetic subgroups. This is a clear example of popula-
tion stratification discussed in the opening chapters of this book. Once aware of the sub-
populations, they could be taken into account in the analysis. A total of 19 regions were
identified that showed genetic association with malarial resistance, one of which was the
HbS locus. The association with marker SNPs in the HbS region was, however, relatively
weak (P < 4 × 10−7) which was puzzling, given the high frequency of the HbS allele in the
Gambian population and the strong protection that it confers against P. falciparum.
Could weak linkage disequilibrium between the HbS tagged SNPs and the HbS causal
SNP in the Gambian population be the reason for the unexpectedly weak association
signal? To address this question the investigators needed more detailed genotype informa-
tion and so they carried out high-resolution association mapping at the HbS locus, using
multipoint mapping/imputation. Imputation analysis is a statistical method used to
infer a genotype by combining information on sequence variation and haplotype struc-
ture—essentially it is a way of filling in missing genotype information. Once again the
investigators recognized that there was considerable local variation in haplotype structure
in Africa and therefore they sequenced the HbS region in 62 of the subjects and used this
as reference data for the imputation analysis, instead of using the HapMap samples from
the Yoruba people, which are more commonly used in African GWAS. The result was a far
more convincing association signal for the SNP rs334 (P = 4.5 × 10−14).

2967_Ch07.indd 236 07/07/2015 10:44

 HIV-1 237

The MalariaGEN GWAS provides an excellent example of the difficulties of carry-

ing out GWAS in African populations. The Wellcome Trust Case Control Consortium
GWAS (2007 study), which investigated seven diseases in the British population, used
P < 5 × 10−7 as a significance threshold. Without the multipoint imputation-based fine-
resolution mapping described above, the association of malarial resistance with the HbS
locus would only just meet this level of significance, based on the earlier study, and this
is despite the HbS allele being present at a high frequency and known to be a strong
resistance determinant. This type of approach may now be applied to future GWAS, but
it is clear that denser, population-specific genotyping arrays for GWAS or direct genome
sequencing are needed for successful studies in Africa and it is hoped that sequencing ini-
tiatives such as the 1000 Genomes Project will help to achieve this.
The aim of GWAS in malaria, of course, is not simply to confirm known risk alleles, but,
more importantly, to identify novel associations with genes not previously implicated in
malaria pathogenesis. The MalariaGEN study identified a number of novel associations,
the most significant being with rs1451375, close to SCO1 that encodes a protein involved
in cytochrome oxidase function, and rs7803788, which lies in an intron of DDC, the gene
for DOPA decarboxylase that is involved in synthesis of dopamine and serotonin. More
work is needed, however, to establish whether these and other newly identified loci play a
role in the progression of malaria.

7.5 HIV-1
HIV-1 is a retrovirus that infects macrophages and helper T cells (CD4+ T cells). The virus
is transmitted by close contact with infected blood and other body fluids, most commonly
during sexual intercourse, or via use of contaminated needles either in a healthcare setting,
intravenous drug abuse, piercing, or tattooing. It can also be passed from mother to child
across the placenta, during birth, and in breast milk. In December 2013, there were 35.3
million people infected with HIV-1 worldwide.
On entering the host, the gp120 glycoprotein on the surface of the virus binds to its pri-
mary receptor, the CD4 protein, on the surface of the CD4+ T cells and macrophages.
This initial binding event results in a conformational change in the gp120 protein, allow-
ing it to interact with a second receptor, or co-receptor. Further conformational changes in
the receptor proteins and viral glycoproteins facilitate fusion of the viral envelope with the
cell membrane and the viral nucleocapsid is released into the cytoplasm of the cell. Once
inside the cell, the viral genome is uncoated and the RNA genome is used as a template in
the synthesis of complementary DNA (cDNA) by the reverse transcriptase enzyme carried
in the virus particle. Immediately after this process, a second viral enzyme, integrase, cata-
lyzes the integration of the viral cDNA into the host cell chromosome. Once integrated,
the viral genes can be transcribed and translated in order to generate new virus particles,
which leave the T cell by budding out from the cell membrane (see Figure 7.5).

C-C chemokine receptor 5 (CCR5) acts as a co-receptor for HIV-1 in

the early stages of infection
CCR5 is a chemokine receptor, located predominantly on the surface of macrophage and
memory T cells. Though not its natural function, CCR5 acts as a co-receptor for HIV-1
in the early stages of infection. At this time the virus is largely limited to these CCR5+ cell
types and is referred to as macrophage or M tropic. M-tropic strains of HIV-1 predominate

2967_Ch07.indd 237 07/07/2015 10:44

 238 CHAPTER 7 Genetics of Infectious Disease

1 2 3 4 5 6 7 8
Virus particle Viral envelope Reverse Viral cDNA T cell RNA transcripts Early genes code Viral RNA and
binds to CD4 fuses with cell transcriptase enters induces are multiply for proteins that proteins are
and co- membrane copies viral nucleus and low-level spliced, amplify assembled
receptor on T allowing viral RNA genomes is integrated transcription allowing transcription of into virus
cell genome to enter into double- into host of provirus translation of viral RNA and particles,
the cell stranded cDNA DNA early genes transport viral which bud
RNA to the from the cell
cytoplasm

CD4 Viral cDNA

co-receptor

Provirus

Cytoplasm Nucleus Chromosomal DNA

Figure 7.5: The life cycle of HIV. HIV infects predominantly macrophages and T cells expressing CD4 on the
cell surface. The virus initially binds to CD4 (primary receptor) and to a co-receptor (CCR5 or CXCR4). The viral
envelope then fuses with the cell plasma membrane and the viral nucleocaspid enters the cytoplasm. After
uncoating, a complementary DNA (cDNA) copy of the viral RNA genome is synthesized by reverse transcriptase
and this cDNA is able to integrate into the host cell chromosome. The integrated genome is transcribed and
translated to produce viral proteins, which package newly synthesized viral RNA genomes to form new virus
particles that bud from the host cell. (From Strelkauskas AJ, Strelkauskas JE & Moszyk-Strelkauskas D [2009]
Microbiology: A Clinical Approach. Garland Science.)

during the asymptomatic phase of infection and are believed to be the strain commonly
transmitted by sexual contact. Over time, however, the viral RNA genome accumulates
mutations and changes in the gp120 protein alter its binding affinity such that it recog-
nizes the CXCR4 chemokine receptor instead of the CCR5 receptor (Figure 7.6). The
CXCR4 receptor is found on a wider range of CD4+ T cells, including naïve T cells, and is
associated with more rapid depletion of the CD4+ T cell population. The CXCR4-adapted
HIV-1 is known as the T-tropic strain and emerges in around 50% of infected individu-
als. CD4+ T cells are essential in both antibody-mediated and cell-mediated arms of the
immune response, so as more T cells become infected and their numbers fall, the host
develops deficiency in both antibody and cell-mediated immunity. This acquired immune
deficiency syndrome is called AIDS. In most cases, without treatment, infection with
HIV-1 is fatal within approximately 10 years. Fatality is not caused by the HIV-1 virus
directly, it occurs as a result of overwhelming opportunistic infections and tumors which
occur due to the failure of the host immune response, especially CD4+ T cell activity.

Some individuals are naturally resistant to HIV infection

Despite the usually high level of HIV-1 infectivity, some individuals who are repeatedly
exposed to HIV-1 are remarkably resistant to infection. Conversely, there are others who are
highly susceptible and progress to AIDS within as little as 1 year. Individuals who show either
unusual resistance or unusual susceptibility can be divided into four phenotypic subsets:
• Exposed uninfected (EU), also known as exposed seronegative: these people show no
sign of HIV infection even after many years of repeated, high-risk exposure to the virus.
• Long-term non-progressors: are infected with HIV-1, but show a delayed progression
to AIDS of several decades.

2967_Ch07.indd 238 07/07/2015 10:44

 HIV-1 239

(a) (b) (c) (d)

M-tropic HIV-1 M-tropic HIV-1 T-tropic HIV-1 T-tropic HIV-1

gp120 gp120 gp120 gp120

gp120 gp120 gp120 gp120

RANTES SDF-1

CD4 CD4 CD4 CD4

CCR5 CCR5 CXCR4 CXCR4

R5-HIV-1 infection Blocked entry X4-HIV-1 infection Blocked entry

Figure 7.6: Attachment and entry of HIV-1. (a) CCR5 is a chemokine receptor expressed on the surface of
macrophages and T memory cells. The gp120 surface protein binds to CD4, inducing a conformational change
that allows gp120 to bind to the co-receptor CCR5. Binding to the co-receptor initiates fusion of the viral envelope
with the cell membrane allowing the virus nucleocapsid of the M-tropic HIV-1 to penetrate the cell. (b) RANTES
chemokine is a natural ligand for CCR5 and has the potential to block attachment of M-tropic strains of HIV-1,
thereby inhibiting infection. (c) As infection progresses, accumulated mutations in the gp120 gene lead to a change
in binding affinity. The resulting T-tropic strains of HIV-1 bind CXCR4 protein, expressed on the surface of a wider
range of CD4+ T cells and macrophages, as a co-receptor instead of CCR5. (d) SDF-1 chemokine is a natural ligand
for CXCR4 and has the potential to block attachment of T-tropic strains of HIV-1, thereby inhibiting infection.
(From O’Brien SJ & Nelson GW [2004] Nat Genet 6:565–574. With permission from Macmillan Publishers Ltd.)

• Fast progressors: are unable to control the virus and develop AIDS in less than 2 years.
• Elite controllers (EC): are infected but able to control the viral replication to an
extremely low level (less than 50 genome copies/ml).
We have already seen how variation in the host genome can influence outcome in malaria.
Could this spectrum of extreme outcomes following HIV-1 infection be attributed to
genetic variation in the human hosts?

A 32-bp deletion in the CCR5 gene confers resistance to

HIV-1 infection
In the 1990s, three research groups, working independently, identified a key genetic vari-
ant in EU individuals. Based on knowledge of the HIV-1 life cycle, they proposed that
variation in one or more of the HIV-1 receptor proteins could reduce the efficiency with
which the virus binds to and penetrates the target cell. 1996 saw a flurry of publica-
tions testing this hypothesis: firstly, in early 1996, the newly identified CCR5 chemokine
receptor was shown to be the major co-receptor for M-tropic strains of HIV-1. Just a
few months later, it was reported that T cells from two EU individuals were resistant to

2967_Ch07.indd 239 07/07/2015 10:44

 240 CHAPTER 7 Genetics of Infectious Disease

infection by M-tropic HIV-1, but were susceptible to T-tropic strains. This result hinted
that CCR5 may be the key to the remarkable resistance of these two subjects and in
August 1996 it was reported that the same two EU subjects were homozygous for a 32-bp
deletion (Δ32) in the CCR5 gene (CCR5-Δ32). The deletion causes a frameshift that
leads to a premature truncation of the CCR5 protein such that it is no longer trafficked
to the cell membrane. Within a matter of weeks two further publications independently
reported that the CCR5-Δ32 homozygotes among EU subjects were entirely absent from
the HIV-infected populations tested. Further analysis of the HIV-infected subjects com-
pared the genotype of those who showed a delayed progression to AIDS (long-term non-
progressors) with those who progressed to AIDS at the usual rate and found CCR5-Δ32
heterozygotes were more frequent among the long-term non-progressors.
The overall conclusions at the end of this remarkably productive year were that the CCR5
polymorphism is a key player in infection by HIV-1. CCR5-Δ32 homozygotes are highly
resistant to M-tropic strains of HIV-1 because macrophage and T cells in these individuals
do not express CCR5 on the cell surface, and so the virus is unable to bind to and enter the
cell via the CCR5 route. Further work showed that CCR5-Δ32 heterozygotes have a 70%
reduced risk of infection compared with those without the Δ32bp deletion and confirmed
that they are more likely to show a delayed progression to AIDS.

Selection pressure by HIV-1 cannot account for the high frequency of

CCR5-Δ32 in the northern European population
The CCR5-Δ32 allele is present in approximately 10% of Europeans, but is very rare in
African and Asian populations. The deletion has been detected in DNA recovered from
skeletal remains up to 2900 years old from various parts of Europe. It seems likely that, at
some time in the past, CCR5-Δ32 provided a survival advantage to the population, pos-
sibly by offering resistance to an infectious pathogen or multiple pathogens. HIV-1 is very
unlikely to be responsible, as HIV-1 infection has only been recorded in human popu-
lations since the early 1980s—not long enough to exert selection pressure. Numerous
epidemics have struck European populations over the last 3000 years and could be impli-
cated, particularly hemorrhagic fever viruses and smallpox. There is circumstantial evi-
dence for a connection between smallpox and CCR5 polymorphism, e.g. myxoma pox
viruses are able to infect cells expressing CCR5 on the cell surface and there is a five-fold
reduction in the replication of M-tropic HIV-1 in lymphocytes from subjects vaccinated
against smallpox compared with non-vaccinated controls. There are, however, arguments
against smallpox being the driving force in selection of CCR5-Δ32, including the fact
that smallpox has in the past been endemic in parts of India and yet CCR5-Δ32 is rare in
Indian populations. It cannot be ruled out that selection advantage has not, in fact, been
conferred by CCR5-Δ32 itself, but by an alternative closely linked allele.

CCR5-Δ32 affects the outcome of infection by West Nile virus

As discussed in the previous section, CCR5 polymorphism plays a key role in infection
by HIV-1. The loss of CCR5 expression on the surface of macrophage and T cells, as seen
in CCR5-Δ32 homozygotes, provides a remarkable level of resistance to HIV-1. CCR5
therefore presents a potential target for therapeutic intervention: could CCR5 be blocked
with, for example, anti-CCR5 monoclonal antibodies in order to prevent HIV-1 infec-
tion? It is important to remember that, in addition to acting as a co-receptor for HIV,
the primary function of CCR5 is to act as a chemokine receptor and help to orchestrate
the immune response to infection. Blocking CCR5 action may therefore compromise the

2967_Ch07.indd 240 07/07/2015 10:44

 HIV-1 241

immune response. Intriguingly, CCR5-Δ32 homozygotes appear be entirely healthy and

generally show no signs of immune suppression. This is presumably due to redundancy
in the chemokine system, whereby if CCR5 is unavailable then its role is taken on by
other chemokine receptors, making CCR5 an attractive therapeutic target. Redundancy
in biological systems is very important when considering the genetics of complex disease.
Although not essential to the immune response in general, CCR5 does have an important
role in the immune response to West Nile virus (WNV). WNV is a flavivirus transmitted
by mosquitoes that pass the virus on to horses, birds, and humans. It was first identi-
fied in Africa in 1937 and cases have been reported across the world since then. 80% of
WNV infections are asymptomatic and the remaining 20% result in West Nile fever, with
patients suffering flu-like symptoms. As many as 1:150 infections result in more severe
disease. In these cases the virus infects the brain causing encephalitis, meningitis, or West
Nile poliomyelitis. In 1999, WNV cases began to appear in the USA, the virus having
been carried by flamingos imported into New York’s Bronx Zoo. This developed into a
serious epidemic with approximately 20,000 human symptomatic cases and 782 deaths,
prompting an increased interest in the biology of WNV.
In 2006, Glass et  al. reported a higher frequency of CCR5-Δ32 in patients admitted
to hospital with severe WNV infection than in uninfected controls. It was apparent
that the CCR5-Δ32 allele, rather than conferring resistance as in the case of HIV-1,
was actually making the patients more susceptible to WNV-related disease. Further
work established that the CCR5-Δ32 variant had no effect on WNV infection per se,
i.e. CCR5-Δ32 was found with the same frequency in WNV-infected patients as in
non-infected controls. The CCR5-Δ32 allele did, however, confer susceptibility to the
more severe, neuro-invasive form of WNV-related disease. CCR5 appears to play an
important role in the immune response to WNV in the brain. Infected neurons secrete
cytokines that bind to CCR5 on the surface of lymphocytes in the peripheral blood,
recruiting them to the site of infection. In CCR5-Δ32 individuals, who have reduced
levels of CCR5 expressed on the cell surface, this signaling process is compromised and
insufficient lymphocytes are recruited to the brain, allowing the WNV infection to take
hold and cause neurological symptoms (see Figure 7.7).
Although WNV is a relatively rare infection, this study does highlight the fact that CCR5
is not an entirely redundant chemokine receptor and that it may be important in the
immune response to other, as yet unidentified pathogens. For example, a subsequent study
in mice suggested that CCR5 may also be important in defense against Japanese encepha-
litis virus, another mosquito borne flavivirus. Therapeutic agents that block CCR5 func-
tion must therefore be designed and used with caution. In addition, even though there is
often redundancy in biological systems that may be sufficient to cope with everyday cir-
cumstances, these systems may break down when under stress. The compensation offered
by other elements of a pathway, which may normally make up for deficiency in one com-
ponent of the system, may then be insufficient to prevent morbidity.

CCR5-Δ32 cannot account for all HIV-1 resistance

It is clear that the CCR5-Δ32 variant cannot explain all observed HIV-1 resistance.
CCR5-Δ32 is present at very low frequencies in non-European populations and yet the
full spectrum of HIV-1 infection phenotypes, from exposed non-infected to fast progres-
sors, can be observed in all infected populations. Other genetic variants must contribute
to the overall heterogeneity in susceptibility to HIV-1 infection and some of these are
discussed in the following sections (see Table 7.5).

2967_Ch07.indd 241 07/07/2015 10:44

 242 CHAPTER 7 Genetics of Infectious Disease

WNV resistance HIV-1 susceptibility

CCR5
CD4
Chemokines
released

CCR5-Δ32
WNV HIV-1

WNV susceptibility HIV-1 resistance

Figure 7.7: The contrasting effect of CCR5-Δ32 on infection by WNV compared with HIV-1. CCR5 is a
chemokine receptor expressed on the surface of macrophages and some CD4+ T cells. It acts as a co-receptor
for HIV-1, facilitating entry of HIV-1 into the host cell. A 32-bp deletion in the CCR5 gene (CCR5-Δ32) inhibits
trafficking of CCR5 to the cell surface and this reduces susceptibility to infection by HIV-1. By contrast, CCR5-Δ32
increases susceptibility to infection of the brain by WNV. CCR5 appears to play an important role in the immune
response in WNV infection, detecting cytokines secreted in response to infection of the brain by WNV and
mediating recruitment of lymphocytes to the site of infection. Reduced expression of CCR5 in cells with the
CCR5-Δ32 allele leads to suppression of the immune response to WNV. (From Lim JK, Glass WG, McDermott DH
& Murphy PM [2006] Trends Immunol 27:308–312. With permission from Elsevier.)

CCR5 promoter polymorphisms affect HIV-1 control

A number of SNPs in the CCR5 promoter are associated with increased or decreased risk
of HIV-1 progression to AIDS. One example, identified as part of the Multicenter AIDS
Cohort Study (MACS), is the CCR5 59029A/G variant (also referred to in the literature as
-2459 A/G and -303 A/G). In a cohort of homosexual and bisexual men in the USA, those
with a 59029G/G genotype progressed to AIDS on average 3.8 years later than those with
the 59029A/A genotype. When the promoter activity of the 59029G variant was tested
in vitro it showed lower activity and lower levels of CCR5 expression on the surface of lym-
phocytes than the 59029A variant. These findings suggested that the slower disease progres-
sion observed in the 59029G/G individuals may be the result of reduced transcription of the
CCR5 gene leading to less CCR5 on the surface of the target cells. The same SNP was later
reported to influence resistance to HIV-1 infection in CCR5-Δ32 heterozygotes. Individuals
with just one copy of the CCR5-Δ32 allele are able to resist infection and display an exposed-
uninfected phenotype if the promoter of the wild-type CCR5 on the other chromosome has
the 59029G variant, resulting in reduced expression of the one functional CCR5 allele.

CCR5/CCR2 haplotypes have a complex effect on HIV-1 control

There is strong linkage between CCR5-Δ32 SNPs in the CCR5 promoter region and a
polymorphism in the CCR2 gene that encodes an isoleucine for valine exchange at amino
acid 64 (CCR2-V64I). CCR2 is another chemokine receptor and the CCR2 gene is located
less than 20 kb upstream of the CCR5 gene. The CCR2 protein is sometimes, though
rarely, used as a co-receptor in HIV infection and there is some evidence that CCR2
may form dimers with CXCR4, thereby reducing the amount of CXCR4 available to
bind HIV-1 in later stages of infection. Early studies indicated that CCR2-V64I was asso-
ciated with delayed progression to AIDS in African-Americans, but not in Europeans,

2967_Ch07.indd 242 07/07/2015 10:44

 HIV-1 243

Table 7.5 Some examples of genetic variations in the genes encoding the chemokine receptor and
chemokine ligands that influence the outcome following infection with HIV-1.

Protein affected Proposed resistance

Variant and role Effect mechanisms
CCR5-Δ32 C-C chemokine Δ32/Δ32: prevents Reduced CCR5 on surface
receptor 5 infection by HIV-1 of target cells leads to
(co-receptor for M-tropic strains; Δ32/+: reduced HIV-1 entry
HIV-1 M-tropic delayed progression to
strains) AIDS
CCR2 V64I C-C chemokine Delayed progression to Dimerization with
receptor 2 AIDS CXCR4, reducing
CXCR4 available for
HIV-1 binding; linkage
disequilibrium with CCR5
CCL5 In1.1C C-C chemokine Accelerated progression Reduced RANTES
ligand 5 to AIDS expression leads to reduced
(RANTES; natural competition with HIV-1
ligand of CCR5) for CCR5 binding sites
CXCL12 3′A C-X-C chemokine Delayed progression to Increased SDF-1
(3′ UTR) ligand 12 (SDF-1; AIDS expression leads to
natural ligand of increased competition
CXCR4) with HIV-1 for CXCR4
binding sites

but subsequent studies in various populations and ethnic groups have given very mixed
results—some indicating protection, some no effect, and some increased risk. It is there-
fore unclear whether CCR2-V64I has a causal effect on AIDS progression or whether the
association is due to linkage disequilibrium with the CCR5 gene.
A key paper published in 1999 by Gonzalez et  al. shed some light on the complexity
of associations between CCR5 and CCR2 genotypes and the outcomes of HIV-1 infec-
tion. The paper identified nine haplogroups, based on polymorphisms in CCR2 and
CCR5, and showed that the frequency of the haplogroups varied considerably between
ethnic and racial groups (Figure 7.8). Moreover, the effect of the various haplotype com-
binations on the outcome of HIV-1 infection also varied depending upon ethnicity. For
example, HHA and HHA/HHF*2 haplotypes were associated with delayed progression
to AIDS in African-Americans but had no effect in European-Americans. Conversely,
HHC haplotypes were associated with delayed progression to AIDS in Europeans and
European-Americans, but with accelerated progression in African-Americans. To add to
the complexity, haplotype pairing is also important. For example, accelerated progression
to AIDS in African-Americans with the HHC haplotype could be mitigated if combined
with the protective effects of HHA or HHF*2, thus HHC/HHA and HHC/HHF*2
African-Americans showed delayed progression to AIDS. Subsequent studies on the effect
of these haplogroups on different populations have added to the complexity of the story
but the important message was eloquently summarized by Gonzalez et al., who observed
that “analysis of a single mutation or haplotype in isolation may obscure the complexity
underlying CCR5 genotype–phenotype relationships”.

2967_Ch07.indd 243 07/07/2015 10:44

 244 CHAPTER 7 Genetics of Infectious Disease

CCR2 CCR5
Pu Pd 1 kb
Exon 3
CCR2-641

+1 ATG
CCR5-Δ32
CCR2-641 Exon 1 Exon 2

29(-2733)A/G

208(-2554)G/T
303(-2459)A/G

627(-2135)C/T
630(-2132)C/T
676(-2086)A/G

(927(-1835)C/T)
CCR5-Δ32
Haplotypes
HHG*1 V G G A C C A C –
HHG*2 V G G A C C A C +
HHF*1 V A G A C C A T –
HHF*2 I A G A C C A T –
HHE V A G A C C A C –
HHD V A T G T T A C –
HHB V A T G T C A C –
HHC V A T G T C G C –
HHA V A G G T C A C –

Figure 7.8: CCR5/CCR2 haplogroups. The gene for CCR5, a chemokine receptor that also acts as an important
co-receptor for HIV-1, is located on the short arm of chromosome 3. The gene for CCR2 (another chemokine
receptor) lies less than 20 kb upstream of the CCR5 gene. Nine CCR2/CCR5 haplogroups were identified by
Gonzalez et al., based on polymorphisms in the CCR5 gene, in the promoter region, and in the CCR2 gene. The
effect of each haplogroup on the outcome of HIV-1 infection was shown to vary depending upon the ethnic or
racial background of the population studied. (Adapted from Arenzana-Seisdedos F & Parmentier M [2006] Semin
Immunol 18:387–403. With permission from Elsevier.)

Polymorphisms in chemokine receptor ligand genes influence

HIV-1 control
The chemokine receptors CCR5 and CXCR4 have a number of natural chemokine
ligands (CCLs). These ligands are chemo-attractant molecules that bind to CCR5 and
CXCR4, recruiting CD4+ T cells to the site of an infection and initiating signaling path-
ways involved in the immune response. The ligands include RANTES (CCL5), MIP-1α
(CCL3), MIP-1αP (CCL3L1), and MIP-1β (CCL4), all of which are ligands for CCR5,
and SDF-1 (CXCL12), which is a ligand for CXCR4. Increased levels of these natural
ligands have been known for some time to compete with HIV-1 for the receptor targets
and thus to reduce HIV-1 infectivity both in vitro and in vivo. Based on this knowl-
edge, the genes that encode the chemokine ligands were investigated as candidate genes in
HIV-1 resistance and susceptibility.
Several SNPs in chemokine ligand genes have been identified that associate with HIV-1
resistance or susceptibility. For example, the In1.1C SNP within intron 1 of the RANTES
gene was found to be associated with accelerated progression to AIDS. In vitro studies sug-
gest that this SNP confers an altered affinity for nuclear-binding proteins and a four-fold
down-regulation of transcription. Therefore, a possible mechanism for the increased rate
of progression to AIDS in individuals carrying the In1.1C allele is that the production

2967_Ch07.indd 244 07/07/2015 10:44

 HIV-1 245

of RANTES is downregulated. Competition for the CCR5 receptor-binding site is con-

sequently reduced and HIV-1 is able to bind to CCR5 and enter the target cell more
frequently.
As alluded to in the previous section, however, it is important to consider the effects of
these chemokine ligand SNPs not only in isolation, but also in combination with CCR5
receptor polymorphisms. The effect of CCR5/CCR5-ligand haplotypes on HIV-1 infec-
tion is therefore an important ongoing area of study.

Polymorphisms in HLA genes affect outcome of HIV infection

As discussed in Chapter 6, the key function of the HLA is to present antigenic pep-
tides to CD4+ and CD8+ T cells, leading to the generation of pathogen-specific acquired
immunity. HLA works well in this process due to a number of factors: the relatively large
number of expressed HLA genes on each haplotype, the extensive polymorphism found
within populations, and the fact that each HLA molecule is capable of presenting a range
of different antigenic peptides and is not restricted to a single peptide antigen. Thus, it
stands to reason that, from the earliest days of HIV-1 research, the HLA genes were likely
to be important candidates for resistance and susceptibility to HIV. Polymorphisms in
HLA class I alleles were the first to be identified as influencing the outcome of HIV-1
infection (see Table 7.6).
HLA class I homozygosity is associated with accelerated progression to AIDS; the effect
is observed for all three HLA class I loci (HLA-A, HLA-B, and HLA-C) and is more

Table 7.6 Some HLA class I gene polymorphisms that have been shown to influence the outcome following
infection with HIV-1.

Protein affected and Proposed resistance

Variant role Effect mechanisms
HLA class I HLA class I (HLA-A, Accelerated Reduced variety of
homozygosity -B, and -C); antigen- progression to AIDS epitopes presented to
presenting protein cytotoxic CD8+ T cells
HLA class I HLA class I (HLA-A, Increased risk of Escape epitopes in mother
concordance -B, and -C); antigen- mother to child and (or infected partner) are
presenting protein sexual transmission able to escape cytotoxic T
cell response in child (or
newly infected partner)
HLA-B*35-Px HLA-B; antigen- Accelerated Reduced epitope binding
presenting protein progression to AIDS decreases cytotoxic T cell
response to HIV-1
HLA-B*57 HLA-B; antigen- Delayed progression to Enhanced antigen
presenting protein AIDS; increased risk presentation
of hypersensitivity to
abacavir
KIR3DS1 +  Killer immunoglobulin- Delayed progression to Enhanced activation
HLA-Bw4 like receptor 3DS1 +  AIDS of NK cells leading to
HLA-B; NK cell improved killing of HIV-1-
activation infected cells

2967_Ch07.indd 245 07/07/2015 10:44

 246 CHAPTER 7 Genetics of Infectious Disease

pronounced if two or three loci are homozygous. HLA class I proteins present viral anti-
gens to cytotoxic CD8+ T cells, leading to destruction of the infected cell. Homozygosity
in the HLA class I allele genotype means that the individual is able to present a reduced
number of viral epitopes compared with a heterozygous individual. Mutation of the
HIV-1 genome allows the virus to escape immune surveillance over time and this process
may happen more quickly in HLA class I homozygotes, which may explain the faster
progression to AIDS. When there is concordance in the HLA genotype of the mother and
child, there is also an increased risk of transmission from mother to child. Concordance in
this case means that both mother and child share the same HLA genotype, so that HIV-1
escape epitopes arising in the mother are also able to escape the cytotoxic CD8+ T cell
response in the child. Similarly, a high risk of transmission is reported to be associated
with concordance between sexual partners.

HLA class I homozygosity is not always bad news

All HLA-B (and some HLA-A) proteins display one of two motifs called Bw4 or Bw6,
and HLA-Bw4 homozygosity has been associated with delayed progression to AIDS. This
apparent contradiction to the accelerated progression described above may be explained
by the missing-self hypothesis which describes an important interaction between HLA
class I proteins and natural killer (NK) cells during the innate immune response to viral
infection. This missing-self hypothesis is not related to the genetic missingness discussed
elsewhere in this book. HIV-1 and many other viruses are able to suppress the cytotoxic
T cell response by downregulating the expression of HLA class I proteins on the surface
of the infected cell. To counter this NK cells are able to detect and destroy infected cells
displaying reduced HLA class I expression. This very specific killing relies on interac-
tion between killer immunoglobulin-like receptor (KIR) proteins, which are expressed
on the surface of NK cells, and HLA class I proteins on the infected cell. KIRs can be
classified as activating or inhibitory. The destruction of an HIV-1-infected cell by an NK
cell may be initiated by stimulation of an activating KIR or suppressed by an inhibitory
KIR (Figure 7.9). Thus, the lack of self-major histocompatibility complex (MHC) class
I molecules on the cell surface leads to NK destruction of infected cells – the missing-self

HLA-Bw4
Self-peptide
Inhibitory KIR Uninfected cell
NK cell

Activating KIR

Killing action
Infected cell

HIV/stress peptide Virus

Figure 7.9 Interaction between KIRs on NK cells and HLA-Bw4 on uninfected and HIV-1-infected cells. NK
cells express KIRs on the cell surface, which may activate or inhibit NK cell activity in response to ligand binding.
This interaction may be modified by the HLA molecule. HLA-Bw4 epitopes are found on some HLA-B proteins, as
well as some HLA-A proteins, and act as ligands for both inhibitory and activating KIRs. This process mediates the
destruction of HIV-1-infected cells by NK cells and may prevent destruction of non-infected cells. (Adapted from
Pelak K, Need AC & Fellay J [2011] PLoS Biol 9(11):e1001208. With permission from the Public Library of Science.)

2967_Ch07.indd 246 07/07/2015 10:44

 HIV-1 247

hypothesis. The HLA-Bw4 motif plays a key role in this process, acting as an impor-
tant KIR-ligand. HLA-Bw4 homozygosity may therefore offer protection from HIV by
enhancing NK control of HIV-1 infection, although the mechanism for this is complex
and not yet fully understood.
Other HLA class I polymorphisms consistently associated with delayed progression to
AIDS are HLA-B*57 and HLA-B*27. The proposed mechanisms underpinning these asso-
ciations are thought to be linked to the peptide-presenting role of the HLA class I proteins.
HLA-B*57 is believed to have a broad peptide-binding specificity, which makes it harder
for HIV-1 escape mutants to evolve. HLA-B*27 is known to recognize and present an epi-
tope from the HIV-1 gag protein and escape mutations with this epitope appear to reduce
the fitness of the virus. Again, it is therefore difficult for the virus to escape the CD8+ T cell
response. It is worth noting that, as discussed in Chapters 6 and 8, HLA-B*57 is also associ-
ated with hypersensitivity to the anti-retroviral drug Abacavir and therefore may also influ-
ence the outcome of HIV-1 infection by modifying the patient’s response to treatment.
Peptide presentation may not, however, be the whole story. Interestingly, both, HLA-B*27
and HLA-B*57 carry the Bw4 motif discussed above, and there is growing evidence that
interaction with KIRs may offer an alternative, or complimentary, protective mechanism.
The KIR genes themselves are highly polymorphic and a marked protective effect has been
reported, e.g. for individuals with high-expressing alleles of KIR3DL1 in combination
with HLA-B*57.
HLA-B*35 is associated with rapid progression to AIDS; the effect appears to vary depend-
ing on the ethnicity of the subjects, with significantly faster progression in European and
European-Americans, but not in African-Americans. HLA-B*35 individuals can be divided
into two subtypes based on peptide-binding preference: HLA-B*35-PY preferentially binds
peptides with a proline at position 2 and a tyrosine (Y) at position 9; HLA-B*35-Px also
binds peptides with a proline at position 2 and has no specific preference (x) at position
9, but will not accept tyrosine. This difference in peptide binding can be attributed to a
single amino acid change at position 116 of the HLA-B*35 allele. Rapid progression to
AIDS is associated with HLA-B*35-Px, but not HLA-B*35-PY. This may explain the ethnic
variation in this association, as HLA-B*35-PY is the most common HLA-B*35 variant in
African-Americans. It should be noted, however, that there are more than 60 listed HLA-
B*35 alleles and this relationship may be much more complex than first reported.

GWAS confirms the protective role of HLA-B in HIV-1 infection

As for malaria, the advent of improved technology has allowed GWAS to be employed
in the search for HIV-1 resistance and susceptibility alleles. The first GWAS of HIV-1
was carried out by Fellay et al. in 2007 and reported association between resistance and
the HLA complex P5 (HCP5) gene on chromosome 6p21.3, within the MHC very close
to the HLA-B locus where alleles predicted to protect against HIV-1 are encoded. Later
work, however, showed that the apparent protective effect of HCP5 could be attributed
to linkage disequilibrium with HLA-B*57. The HCP5 gene (now used as a proxy for
HLA-B*57) was identified in at least two other GWAS and HLA-C was also found to
be associated with protection against HIV-1 in at least four GWAS. There was there-
fore striking reproducibility between several of the GWAS, although this is perhaps less
surprising when taking into account the fact that all the studies mentioned above were
carried out on Europeans and those with European ancestry. Two GWAS of African and
African-American cohorts carried out in 2010 failed to identify any alleles reaching the
significance threshold. This once again highlights the challenges of conducting GWAS in

2967_Ch07.indd 247 07/07/2015 10:44

 248 CHAPTER 7 Genetics of Infectious Disease

the genetically diverse African ethnic populations. As discussed for malaria above, more
extensive SNP arrays, the application of imputation analysis, and more detailed sequence
data such as that provided by the 1000 Genomes Project, will help improve the success of
GWAS in these very important cohorts. As Africa bears by far the largest burden of HIV-1
infection, such developments will be critical in the fight against AIDS.

Amino acids in the HLA-B binding groove are associated with

HIV-1 control
A GWAS published in 2010 by the International HIV Controllers Study identified four
SNPs associated with control of HIV-1, including marker SNPs for HLA-B*57:01 and
HLA-C. The study went on to define the HLA-B amino acids most significantly associated
with protection to be 67, 70, and 97, with amino acid 97 showing the strongest impact.
Amino acids 62 and 63 were also found to be significant. These amino acids are all located
in the peptide-binding groove of the HLA-B protein (Figure 7.10), which again provides
convincing evidence that variation in the specificity and efficiency of peptide presentation
may play a central role in HIV-1 control.

GWAS revealed, for the first time, association of HLA-C with

HIV-1 control
As discussed above, HLA-B and HLA-C showed the most consistent association with
HIV-1 control in GWAS. However, whereas the HLA-B association confirmed previous
results from hypothesis-driven studies, the findings regarding HLA-C were new. HLA-C,
like the other HLA class I proteins, presents peptides to CD8+ T cells, but is not down-
regulated during HIV-1 infection, and interacts with KIR on NK cells to a greater extent
than the HLA-A and HLA-B antigens. The newly associated tagged SNP is located 35bp
upstream of the HLA-C gene, but is not proposed to be the causal polymorphism. Instead

62 70
67
63

97

Figure 7.10: A three-dimensional ribbon diagram of the peptide-binding groove of the HLA-B protein.
Variations at amino acid positions 62, 63, 67, 70, and 97, all located in the peptide-binding groove, have been
shown to affect HIV-1 resistance. Variation in amino acid 97, located in the floor of the peptide-binding groove,
has the greatest impact on HIV-1 resistance.

2967_Ch07.indd 248 07/07/2015 10:44

 Conclusions 249

the mechanism of control appears to involve a second SNP, 263D/I, located in a binding
site for micro-RNA in the 3′ untranslated region (UTR) of the HLA-C gene, which is in
linkage disequilibrium with the –35 SNP in Europeans. Binding of micro-RNA destabi-
lizes mRNA transcripts and is an important control mechanism for expression of many
genes. The 263D/I polymorphism affects the amount of HLA-C expressed on the cell
surface and presumably thereby moderates the immune response to HIV-1.

Some SNPs previously implicated in HIV-1 control have not yet been
confirmed by GWAS
As polymorphisms in the CCR5 gene and its chemokine ligands are clearly associated
with HIV resistance we would predict that association with these alleles would be identi-
fied in GWAS. Interestingly, only a weak association with CCR5/CCR2 was detected and
no significant association was seen with any of the other SNPs identified in hypothesis-
driven studies (except those in the MHC region). MHC variants have been proposed to
account for around 19% of the variability seen in viral load in HIV-1-positive subjects,
which rises to 23% when combined with the CCR5/CCR2 variants. There are clearly a
large number of as yet undetermined factors, probably both genetic and environmental,
which affect infection by HIV-1 and progression to AIDS.

Conclusions
Infectious disease, unlike most of the other complex diseases discussed in this book, is not
generally considered to be genetic in the common use of the term. However, it does have
a heritable component. It is important to remember that infectious disease has played a
major role in human evolution and genetic diversity. Indeed, one of the important scien-
tific issues that developed from present studies has been a better understanding of how the
pressure from infectious disease on human populations has played a significant role in the
evolution of the human genome and also the maintenance of particular alleles that confer
resistance to pathogens.
Heritability in infectious disease can be illustrated by considering human history. For
example, consider the impact of pathogens carried by the conquistadors on the indig-
enous populations of South America, and later scientific observations on the apparent
heritability of leprosy and tuberculosis. Both of these examples illustrate how the outcome
of exposure to a pathogen may be influenced by the host genome. There is an enormous
variety of pathogens that can challenge Homo sapiens, and new viruses and bacterial strains
arise frequently, e.g. the hepatitis viruses (A, B, and C) and HIV-1 are all quite recently
emerged human pathogens. In a genetically diverse population when new infectious
agents develop, diversity ensures that someone is likely to survive a pandemic, whereas in
a less diverse population the likelihood of survival will be reduced.
The genetics of infectious diseases can be more difficult to study than the genetics of non-
infectious chronic disease. The causative agents of most infectious diseases are known. In
some cases exposure almost always leads to infection, so investigating the role of genes in
infection per se is not possible or appropriate. The real points for genetic research often
lie with understanding response to infection, progression, outcome, and response to treat-
ment, and not initial susceptibility or resistance to infection, which may be governed
by other factors. However, this is not true for all infectious diseases, e.g. HIV-1 and the
CRR5-Δ32 deletion discussed above.

2967_Ch07.indd 249 07/07/2015 10:44

 250 CHAPTER 7 Genetics of Infectious Disease

The same processes that have been applied to non-infectious disease have been applied to
infectious disease and with great success in some cases, especially when the right question
is set. Substantial existing knowledge of the infectious process has allowed scientists to
use a hypothesis-driven approach to identify genes that confer susceptibility or resistance
to infectious disease. More recently, non-hypothesis-constrained approaches, particularly
GWAS, have been used to identify susceptibility and resistance alleles in infectious dis-
ease, and this is shedding new light on the mechanisms by which some pathogens cause
disease. There are, however, significant technical challenges to be overcome in carrying out
GWAS in non-European and mixed populations, especially in Africa, where the burden
of infectious disease is greatest. Despite considerable progress in our understanding of the
interplay between the human genome and infectious disease, there are as yet few examples
of this knowledge leading to effective preventions or treatments.
By concentrating on two selected examples of infectious disease, this chapter shows how
variation in the host genome plays a significant role in determining the disease phenotype.
The first example, malaria, illustrates how genetic variation associated with Mendelian
recessive traits can influence complex traits. In the second example, HIV-1, we consid-
ered how inherited variation in genes can directly influence infection with a virus. In this
example we also considered how genetic variation can influence the host response to a
virus and briefly how our genes may determine our response to treatment. The last point
is covered in more detail in chapter 8. All of this holds a clue to how we may one day con-
quer many different infections through a better understanding of disease pathogenesis and
the development of more efficient treatment programs, including better vaccines where
appropriate.
Future use of whole-genome sequencing, epigenetics, and expansion of existing databases
using imputation may produce new insights into the genetics of infectious disease. We are
making progress towards fulfilling some of the promises of the Human Genome Project
that is to improve patient management and care and improve our understanding of disease
pathology. However, as we have seen, and shall see again in the next chapter there is still
plenty of work to be done.

2967_Ch07.indd 250 07/07/2015 10:44

 Further Reading 251

Further Reading
Books to malaria. Malaria J 10:271–281. An overview
Engleberg NC, DiRita V & Dermody T of polymorphisms associated with susceptibil-
(2012) Schaechter’s Mechanisms of Microbial ity and resistance to malaria.
Disease, 5th ed. Lippincott Williams & Fellay J, Shianna KV, Ge D et  al. (2007) A
Wilkins. An accessible entry-level text whole-genome association study of major
book on microbial disease, including chap- determinants for host control of HIV-1. Science
ters on the immune response, retroviruses 317:944–947.
(including HIV-1), and protozoa (including
Glass WG, McDermott DH, Lim JK et  al.
P. falciparum).
(2006) CCR5 deficiency increases risk of symp-
Haines JL & Pericak-Vance MA (1998) tomatic West Nile virus infection. J Exp Med
Overview of mapping common and geneti- 203:35–40.
cally complex human disease traits. In Gonzalez E, Barnshad M, Sato N et al. (1999)
Approaches to Gene Mapping in Complex Race-specific HIV-1 disease-modifying effects
Human Disease (Haines JL & Pericak-Vance associated with CCR5 haplotypes. Proc Natl
MA eds), pp 1–16. Wiley. This book provides Acad Sci USA 96:12004–12009. This is a key
the best definition of complex disease and is a study that first defined and investigated CCR2–
good starter text on major issues in complex CCR5 haplogroups and their influence on HIV
disease. progression to AIDS.
Hill AVS, Allsop CEM, Kwiatkowski D et  al.
Articles (1991) Commonest African HLA antigens are
Allison AC (2004) Two lessons from the associated with protection from severe malaria.
interface of genetics and medicine. Genetics Nature 352:595–600.
166:1591–1599. This is an interesting auto- International HIV Controllers Study (2010)
biographical account of the discovery of the The major genetic determinants of HIV-1 con-
protective role of sickle cell anemia against trol affect HLA class I peptide presentation.
malaria. Science 330:1551–1557. This paper reports
An P & Winkler CA (2010) Host genes asso- a major collaborative GWAS that implicates
ciated with HIV/AIDS: advances in gene dis- HLA-C and the binding groove of HLA-B in
covery. Trends Genet 26:119–131. This is an the response to HIV-1 infection.
excellent overview of the field that includes a Jallow M, Teo YY, Small KS et  al. (2009)
discussion of genome-wide RNA interference Genome-wide and fine-resolution analysis of
screens, not included in this chapter. malaria in West Africa. Nat Genet 41:657–
Arenzana-Seisdedos F & Parmentier M (2006) 665. This paper describes a GWAS study
Genetics of resistance to HIV infection: role carried out by members of the MalariaGEN
of co-receptors and co-receptor ligands. Semin consortium that provides an excellent exam-
Immunol 18:387–403. ple of the challenges of GWAS in African
Carrington M & Walker B (2012) populations.
Immunogenetics of spontaneous control of Lim JK, Glass WG, McDermott DH &
HIV. Annu Rev Med 63:131–145. This is a clear Murphy PM (2006) CCR5: no longer a “good
and readable review on the effect of HLA geno- for nothing” gene – chemokine control of
type on response to HIV. West Nile virus infection. Trends Immunol
Chapman SJ & Hill AVS (2012) Human sus- 27:308–312. This is an interesting review
ceptibility to infectious disease. Nat Rev Genet of the role of CCR5 in the control of WNV
13:175–188. A comprehensive overview of infection.
recent developments in the field; includes a dis- Lopez C, Saravia C, Gomez A et  al. (2010)
cussion of single-gene variants associated with Mechanisms of genetically based resistance to
susceptibility and resistance to infectious dis- malaria. Gene 467:1–12. A review of poten-
ease, and a very useful glossary of terms. tial mechanisms by which polymorphisms
Driss A, Hibbert JM, Wilson NO et al. (2011) associated with susceptibility and resistance to
Genetic polymorphisms linked to susceptibility malaria exert their effects.

2967_Ch07.indd 251 07/07/2015 10:44

 252 CHAPTER 7 Genetics of Infectious Disease

Vannberg FO, Chapman, SJ & Hill AVS http://www.malariagen.net

(2011) Human susceptibility to intracellu- MalariaGEN Genomic Epidemiology
lar pathogens. Immunol Rev 240:105–116. A Network. A data-sharing community focusing
review focused on intracellular pathogen genet- on genes involved in resistance to malaria.
ics that provides an interesting range of patho-
gens beyond the examples given in this chapter. http://www.1000genomes.org
An ambitious project run by the European
Williams TN, Maitland K, Bennett S et  al. Bioinformatics Institute that aims to find most
(1996) High incidence of malaria in alpha- genetic variants that have frequencies of at least
thalassaemic children. Nature 383:522–525. 1% in 27 populations worldwide.
http://www.hivcontrollers.org
Online sources A collaborative study that aims to investigate
http://www.who.int why some people infected with HIV are able to
The WHO provides up-to-date information control the infection without medication.
about the infectious diseases discussed in this http://statepi.jhsph.edu/macs/macs.html
chapter. Multicenter AIDS Cohort Study.

2967_Ch07.indd 252 07/07/2015 10:44

 CHAPTER
8
Pharmacogenetics

Individuals vary considerably in their response to prescribed drugs and other foreign
compounds. Genetic factors have been shown to make an important contribution to
this variability. The subject of pharmacogenetics is concerned with this relationship.
Though considered to be a specialized discipline, pharmacogenetics can also be con-
sidered to be a sub-branch of complex disease genetics, because even though many
pharmacogenetic traits behave as Mendelian traits, some are complex traits with low
levels of penetrance and more than one risk allele. In this chapter, we will concentrate
on genetic factors that affect either drug metabolism or the interaction of the drug with
its target within the body and also with adverse drug reactions, which are unwanted
and unexpected effects of a prescribed drug that can sometimes be life-threatening to
the patient. A number of different drug classes will be included in the examples we
consider, but we will not consider pharmacogenetic aspects of individual physiologi-
cal systems or deal with particular medical specialties. Pharmacogenetics can also be
extended to other areas not concerned with response to prescribed medicines such
as genetic susceptibility to diseases where chemical exposure is a risk factor, includ-
ing cancer. This area will be considered briefly in the Chapter 9. Similarly, the cancer
genome as opposed to the patient genome may help predict the outcome of drug treat-
ment in cancer. This specialized aspect of pharmacogenetics will also be considered
separately in Chapter 9.

2967_Ch08.indd 253 07/07/2015 10:48

 254 CHAPTER 8 Pharmacogenetics

8.1 DEFINITION AND A BRIEF HISTORY OF

PHARMACOGENETICS
The term pharmacogenetics has been in use since the 1950s and can be defined broadly
as the study of genetic factors affecting the response to drugs. Box 8.1 summarizes the
development of the subject from the early twentieth century to the present day. There
is considerable overlap between pharmacogenetics and the more recent discipline of
pharmacogenomics and both terms are used interchangeably. Pharmacogenetics has
traditionally been concerned with single-gene effects, whereas pharmacogenomics is
described as the whole-genome application of pharmacogenetics. In this chapter, the
term pharmacogenetics will be used to cover both genetic and genomic examples.

BOX 8.1 MILESTONES IN PHARMACOGENETICS

1902 Garrod discusses chemical individuality in relation to the disease

alkaptonuria.
1932 Synder shows the ability to taste the bitter-tasting compound phenylthio-
carbamide is genetically inherited.
1952–1956 The antimalarial drug primaquine shown to give rise to acute hemo-
lysis in some individuals who lack the enzyme glucose 6-phosphate
dehydrogenase.
1953–1960 Metabolism of the antituberculosis drug isoniazid shown to vary between
individuals. Those who convert the drug slowly to acetylisoniazid are
more likely to develop peripheral neuritis, with this trait being inherited
recessively.
1956–1957 The adverse drug reaction succinylcholine apnea, where rare individuals
given this muscle-relaxing drug develop breathing difficulties, shown to
be due to an inherited defect in the enzyme pseudocholinesterase, which
metabolizes the drug.
1959 Term “pharmacogenetics” first used by Friedrich Vogel.
1975–1977 Working separately, researchers in the UK and Germany show that some
individuals are unable to oxidize the drugs sparteine and debrisoquine.
1980 Genetic polymorphism in metabolism of 6-mercaptopurine by thiopu-
rine methyltransferase described.
1984 New pharmacogenetic polymorphism affecting mephenytoin metabolism
described.
1988 CYP2D6 cloned and genetic polymorphisms responsible for poor metab-
olism of sparteine and debrisoquine identified.
1993 Gene duplication and other amplifications of the CYP2D6 gene shown to
occur in an early example of CNV. This results in increased rates of drug
metabolism (ultrarapid metabolism).

2967_Ch08.indd 254 07/07/2015 10:48

 Cytochrome P450 255

BOX 8.1 MILESTONES IN PHARMACOGENETICS  (Continued)

1994 CYP2C19 gene shown to code for the enzyme that metabolizes mephe-
nytoin and gene defects identified.
1995 Thiopurine methyltransferase gene cloned and polymorphisms respon-
sible for absence of activity identified.
1997 Term “pharmacogenomics” first used.
1998 Trastuzumab (Herceptin®) licensed as a targeted treatment for breast cancer.
2001 First draft of the Human Genome Map.
2004 VKOR gene cloned. Polymorphism in this gene is a key determinant of
warfarin dose requirement.
2007 First GWAS on complex diseases are followed by GWAS on pharmacoge-
netic traits.
2008 Clinical trial shows HLA-B*57:01 to be a highly sensitive predictor of
abacavir hypersensitivity. Genotyping for this allele prior to treatment
becomes common.
2009 IL28B genotype shown to be key predictor of interferon-α response in hepa-
titis C infection.

8.2 Cytochrome P450

There are two stages in drug metabolism: phase I and phase II. In phase I, reactions are car-
ried out by monooxygenases that work by adding an oxygen atom. The effect is to make it
easier for the second phase to be carried out. Phase II reactions involve transferase enzymes
that conjugate (add) small polar molecules such as glucuronic acid, sulfate, or glycine to
the drug. Phase I and phase II reactions are illustrated in Figure 8.1.
Early studies on drug metabolism showed that many phase I metabolic reactions occur in
the endoplasmic reticulum (ER) of cells in the liver. In 1962, Omura and Sato isolated a
protein they named cytochrome P450 from rat liver ER. This protein showed an absor-
bance peak at 450 nm and contained a molecule of heme. Further studies confirmed that
this protein was an enzyme with a role in the metabolism of a range of drugs. It was ini-
tially assumed that cytochrome P450 was a single enzyme, but evidence soon emerged to
suggest multiple cytochrome P450s, with at least 50 different human cytochrome P450s.

There is a clear relationship between genotype and phenotype for

several forms of cytochrome P450
In parallel with continuing biochemical studies on cytochromes P450 in the 1970s, studies
on the drugs debrisoquine and sparteine, showed that approximately 10% of Europeans
could not metabolize these drugs. The term poor metabolizer was used to describe this
phenotype. The deficiency in metabolism of both drugs was found to co-segregate in fami-
lies and the trait was found to be inherited recessively. Subsequently, the enzyme respon-
sible for metabolism of these two drugs was shown to be a member of the cytochrome
P450 family of enzymes, now referred to as CYP2D6. This is one of the most important

2967_Ch08.indd 255 07/07/2015 10:48

 256 CHAPTER 8 Pharmacogenetics

Lipophilic drug

Monooxygenases

FMO CYP450
Phase I
Mostly oxidation
reactions, notably Esterases,
aromatic hydroxylation hydrolases

BCHE, CES PON EPHX

etc.

Water-soluble
intermediate with Ac
reactive “handle” Me

NAT
TPMT
Phase II
Conjugation reactions
in which a transferase UDPG Glucuronyl
adds a chemical group
GST Glutathionyl
ST

Excretable SO42–
product

Figure 8.1: Phase I and phase II drug metabolism. Phase I drug reactions create more polar drugs with a
reactive group that prepares the drug for phase II. They involve a variety of enzymes, including CYP450. The
phase II reactions involve addition of chemicals to the molecule to facilitate excretion of the final product.
Note that phase II reactions do not always occur after phase I reactions; they can occur without a phase I
reaction. BCHE, butyrylcholinesterases; CES, carboxyesterase; EPHX, epoxide hydrolases; FMO, flavin-containing
monooxygenases; GST, glutathione S-transferases; NAT, N-acetyltransferases; PON, paraoxonases; TPMT,
thiopurine methyltransferases; UGT, UDP-glucuronosyltransferases; ST, sulfotransferases. (From Strachan T,
Goodship J & Chinnery PF [2014] Genetics and Genomics in Medicine. Garland Science.)

cytochrome P450s involved in drug metabolism, contributing to metabolism of around

20% of all of the drugs metabolized by the cytochrome P450 family. A large number of
different polymorphisms in CYP2D6 associated with the poor metabolizer phenotype have
been described, but the most common variant involves a base substitution at an intron/
exon boundary that affects RNA splicing, leading to production of a truncated protein
with no enzyme activity. In addition to the poor metabolizer phenotype, there is also an
ultrarapid metabolizer phenotype associated with higher than normal CYP2D6 activity
due to the presence of one or more additional copies of the CYP2D6 gene adjacent to the
wild-type gene in germ-line DNA. The finding of extra copies of CYP2D6 in germ-line
DNA was one of the first reports of copy number variation (CNV) in the human genome.
It is particularly significant that copy number is related to function in this example as this
is not always the case when CNV is observed in other genes. Additional polymorphisms in
the CYP2D6 gene that give rise to amino acid changes that decrease, but do not abolish,
enzyme activity have been identified. Individuals homozygous for these alleles or hetero-
zygous for both an absence of activity allele and a decreased activity allele fall into the cat-
egory of intermediate metabolizers. As shown in Figure 8.2, there is a clear relationship
between drug-metabolizing activity for the antidepressant nortriptyline and the number

2967_Ch08.indd 256 07/07/2015 10:48

 Cytochrome P450 257

(a) Number of active Allele

CYP2D6 genes
CYP2D6
n=1 Normal (N)
n=1 Reduced function
(partially active) mutation ( )

n=0 Inactivating
mutation ( )

n=0 Deletion ( )
Simple
n=2
duplication
Multiplex
n = 3–13 duplication

(b)
Multiple
Genotype N,N or N,other or or or or
CYP2D6 genes
Ultrarapid Extensive Intermediate Poor
Phenotype metabolizers metabolizers metabolizers metabolizers
Frequency
5–10% 65–80% 10–15% 5–10%
(Caucasians)

90
80

70
Number of patients

60
50
40
30
20
10

0
0.01 0.1 1 10 100
Metabolic ratio
Nortriptyline dose requirement (mg/day)
>250–500 150–100 20–50
Nortriptyline (mg)

Figure 8.2: CYP2D6 alleles and relationship between genotype and metabolism of the antidepressant
drug nortriptyline. (a) CNV in CYP2D6 genes is common in some populations and creates different
phenotypes according to the copy number. There are also deletions, inactivation, and reduced function
mutations. (b) CNV results in four phenotypes: ultrarapid metabolizers, extensive metabolizers, intermediate
metabolizers, and poor metabolizers. The figure shows urinary concentration of the drug substrate, in this
case the antidepressant nortriptyline, with high metabolic ratios representing poor metabolizers and low
metabolic ratios representing ultrarapid metabolizers. Extensive and intermediate metabolizers are shown in
between these two extremes. (Adapted from Meyer UA [2004] Nat Rev Genet 5:669–676. With permission from
Macmillan Publishers Ltd.)

2967_Ch08.indd 257 07/07/2015 10:48

 258 CHAPTER 8 Pharmacogenetics

Table 8.1 Properties of selected cytochromes P450 important in human drug metabolism.

Examples of polymorphisms
(including rs number where Effect of
Isoform Substrates appropriate) polymorphism
CYP2D6 Debrisoquine, tricyclic Splice site (rs3892097) No activity
antidepressants, codeine
Deletion No activity
Missense (rs165852) Decreased activity
Duplication Increased activity
CYP2C19 Clopidogrel, diazepam, Splice site (rs4244285) No activity
omeprazole
Promoter region base substitution Increased activity
(rs12248560)
CYP2C9 Warfarin, diclofenac, Missense (rs1057910) Decreased activity
ibuprofen
CYP3A4 Midazolam, nifedipine, Missense (rs55785340) Decreased activity
cyclosporin, tacrolimus
Intronic base substitution Decreased activity
(rs35599367)
CYP3A5 Tacrolimus Splice site (rs776746) No activity
CYP2A6 Nicotine Missense (rs1801272) No activity
Deletion No activity

of active CYP2D6 genes. The CYP2D6 polymorphisms are associated with metabolizing
activity for a number of different drugs that use the same pathway.
Though CYP2D6 was the first example of a cytochrome P450 gene showing functionally
significant polymorphism, there are now a number of similar examples involving other
members of the cytochrome P450 family of genes. Table 8.1 lists the most important
examples with details of typical drug substrates and the main polymorphisms involved
together with their overall effects. Polymorphisms in genes encoding enzymes that affect
drug metabolism such as cytochrome P450 may result in circulating drug levels that are
either too low or too high to achieve the normal therapeutic response. It is possible to over-
come this problem by either changing the dose or prescribing a drug that is metabolized
by a different enzyme. Three drugs where the cytochrome P450 genotype appears to be
clinically relevant are considered in more detail below: codeine, warfarin, and clopidogrel.

The conversion of the analgesic drug codeine, which is administered

as a pro-drug and is activated to morphine by CYP2D6, is of clinical
importance
Codeine is used widely as an analgesic, but it is a pro-drug that needs to be converted to
morphine by CYP2D6 for the analgesic effects to occur. This means that it is generally not
an effective analgesic in those who lack CYP2D6 activity. This is illustrated in Figure 8.2

2967_Ch08.indd 258 07/07/2015 10:48

 Cytochrome P450 259

in relation to a different drug (nortriptyline) that is metabolized by this same enzyme.

Studies of this phenomenon for codeine have been relatively limited, partly because it is
difficult to accurately define the severity of pain. However, there are a number of recent
reports that suggest that in addition to lack of or low levels of CYP2D6 activity being a
problem when codeine is administered individuals who have additional CYP2D6 copies
(ultrarapid metabolizers) may have high levels of CYP2D6 and hence high levels of mor-
phine. Under certain conditions high levels of morphine generated in those with multiple
copies of the CYP2D6 gene may have significant consequences, with serious, even poten-
tially fatal, adverse outcomes when codeine is administered. This seems to be a particular
problem with infants and children, though there are also some reports of adverse reactions
in adults. One relatively recent study from Koren et al. (2006) in Canada described the
case of a breast-fed baby who died 13 days after birth. Further investigations found that
the mother’s breast milk, given to the baby, contained a high level of morphine. This high
level of morphine was generated by elevated CYP2D6 activity in the mother. Genetic
investigations found that the mother carried more than one copy of the CYP2D6 gene
and was therefore an ultrarapid metabolizer for codeine. The baby had a normal CYP2D6
genotype. Subsequently there were a number of similar reports of serious problems follow-
ing treatment with codeine where either children or adults were ultrarapid metabolizers.

The cytochrome P450 CYP2C9 metabolizes warfarin – a very widely

used drug
Warfarin, which was developed originally as a rat poison, is an anticoagulant drug that acts
by inhibiting the enzyme vitamin K epoxide reductase (VKOR) (see Section 8.4). Vitamin
K is essential in the activation of several factors in the blood coagulation cascade. If VKOR
is unable to regenerate reduced vitamin K from vitamin K epoxide produced during the
clotting process (see Figure 8.3), the normal coagulation process is disrupted. In patients
who are at risk of developing blood clots due to some types of cardiovascular disease, dis-
rupting coagulation can be help prevent the formation of clots that might lead to strokes
and heart attacks. However, if there is too much inhibition of VKOR, the patient is at risk
of developing uncontrolled and potentially fatal bleeding due to the inability of blood to
clot. This means that the dose of warfarin given to the patient needs to be titrated by mon-
itoring the drug response on a regular basis and especially during initial treatment. The
dose needed by different patients to achieve the required level of anticoagulation varies.
As shown in Figure 8.3, warfarin is metabolized to the inactive 7-hydroxywarfarin by the
cytochrome P450 CYP2C9. Two common polymorphisms which each result in amino
acid substitutions have been detected in the CYP2C9 gene. Individuals who have one or
two of the variant alleles show slower than normal warfarin metabolism. Some years ago,
it was demonstrated that patients being treated with warfarin who were positive for one
or more CYP2C9 polymorphisms generally needed lower doses of warfarin than patients
who were homozygous wild-type. This finding has been confirmed in a large number of
subsequent studies and there is general agreement that CYP2C9 genotype is an important
determinant of warfarin dose requirement, especially in those of European or African
ethnicity.

CYP2C19 activates clopidogrel – a drug widely used to prevent strokes

and heart attacks
Clopidogrel is a drug used in the prevention of strokes and heart attacks in high-risk
patients, especially those who have already suffered such events and have undergone

2967_Ch08.indd 259 07/07/2015 10:48

 260 CHAPTER 8 Pharmacogenetics

Vitamin K-dependent coagulation factors:

II, VII, IX, X

Biologically
Precursor
Carboxylase active
Oxidative
deactivation

Vitamin K Vitamin K
(reduced) (epoxide)

CYP1A2
CYP2C9
CYP3A4
VKORC1

S-warfarin R-warfarin

Warfarin

Figure 8.3: The mechanism of action of warfarin and role of VKORC1 in the pathway. The role of VKORC1
in the conversion of vitamin K epoxide to oxidized vitamin K is shown. The carboxylase enzyme transfers carboxyl
groups to glutamic acid residues of selected clotting factors with a requirement for oxidized vitamin K. Warfarin
inhibits the VKORC1 reaction. There are two chemically distinct enantiomers of warfarin: the R and S forms. The
S enantiomer is more biologically active than the R form and is also metabolized only by CYP2C9, whereas other
P450 isomers metabolize the R form.

arterial stenting. Cytochrome P450 genotype may be important in predicting indi-

vidual responses to clopidogrel, which is a pro-drug that needs to be metabolized to
have biological activity. One of the enzymes involved in this metabolism is the cyto-
chrome P450 CYP2C19. CYP2C19 activity is completely absent in around 3% of
white Europeans and 20% of people from East Asia, who are often referred to as poor
metabolizers. CYP2C19 deficiency is an autosomal recessive trait that is most com-
monly due to a splice site mutation resulting in a truncated protein with no enzyme
activity (Table 8.1). A number of studies suggest that patients treated with clopidogrel
who are homozygous for CYP2C19 variant alleles associated with absence of activity
have an increased risk of death and myocardial infarction, though not all studies agree
on this. There is also more limited data suggesting that those carrying one or more
copies of the CYP2C19 allele CYP2C19*17 (ultrarapid metabolizers) are at increased
risk of suffering bleeding when treated with clopidogrel due to higher than normal
circulating levels of the activated drug.
The practical consequences of these findings are such that the US Food and Drug
Administration (FDA) now refers to the CYP2C19 polymorphism on their approved label
for clopidogrel. Those homozygous for CYP2C19 poor metabolizer alleles could be given
an alternative drug as there are several drugs with similar properties to clopidogrel that do
not require activation by CYP2C19.
These examples provide a direct link to the promises of the Human Genome Project
(HGP), where it was suggested that we would in future use genetics to develop individu-
alized therapies for patients and in patient management and care. Several of the findings
described above, however, predate the publication of the first complete human genome
sequence.

2967_Ch08.indd 260 07/07/2015 10:48

 Other Drug-Metabolizing Enzymes and Transporters 261

8.3 Other Drug-Metabolizing Enzymes and

Transporters
Though the cytochrome P450 enzyme family is perhaps the most important enzyme fam-
ily in drug metabolism, other enzyme families also contribute. Several genetic polymor-
phisms affecting drug metabolism by conjugation (phase II metabolism) have been well
studied.
In addition to phase I and phase II, several proteins known as drug transporters affect
entry and efflux of drugs into a variety of cell types, including enterocytes, hepatocytes,
and renal epithelial cells. These contribute to drug absorption, distribution and excretion
within the body, and are also subject to genetic polymorphisms relevant to drug treat-
ment. The overall effects of these polymorphisms are on drug levels, broadly similar to
those described for the cytochromes P450 in Section 8.2.

For phase II conjugation reactions, the UDP glucuronosyltransferase

family makes the largest contribution
The UDP glucuronosyltransferases (UGTs) are one of the major players in phase II conju-
gation reactions. As with the cytochrome P450 family, a number of polymorphisms with
functional effects have been detected in the UGT gene superfamily, but one of the most
studied is the UGT1A1 gene. In addition to a role in drug metabolism, UGT1A1 contrib-
utes to metabolism of endogenous compounds by conjugating bilirubin, produced during
heme degradation with glucuronic acid, which increases its solubility and facilitates biliary
excretion. Bilirubin is a toxic compound whereby higher than normal plasma levels result
in jaundice, which can lead to toxic effects in a number of organs including the brain.
Approximately 5% of Europeans show mildly elevated plasma bilirubin levels because of
slower than normal conjugation with glucuronic acid in a condition called Gilbert’s syn-
drome, which is generally considered benign. This syndrome is due to lower than normal
levels of UGT1A1 activity. This decreased activity is associated with homozygosity for
a T > A insertion in the promoter region of the UGT1A1 gene close to the TATA box
resulting in lower than normal transcriptional activity. UGT1A1 has an important role in
the metabolism of irinotecan—an anticancer drug used in the treatment of certain forms
of advanced cancer, especially colon cancer and lung cancer. Individuals homozygous for
the UGT1A1 promoter region polymorphism have been demonstrated to be more likely
to develop neutropenia, a serious adverse reaction where neutrophil counts are very low,
if treated with irinotecan. In the USA, it is recommended by the FDA that patients under-
going chemotherapy with irinotecan should be genotyped for this promoter region poly-
morphism and a lower dose of the drug used in those homozygous for this variant.

Methyltransferases are also important in phase II drug metabolism

Conjugation with a methyl group is another type of phase II drug metabolism reaction. It
does not occur as commonly as conjugation with glucuronic acid and there are a number
of different methyltransferase enzymes that carry out these reactions. One of the most
frequently studied pharmacogenetic examples from this group is thiopurine methyltrans-
ferase (TPMT), which has a rather specialized role in the conjugation of 6-mercaptopurine.
6-mercaptopurine is commonly used in the treatment of childhood leukemia. Azathioprine
is a pro-drug precursor of 6-mercaptopurine and is quite widely prescribed for the treat-
ment of common diseases with an immune etiology, such as inflammatory bowel disease,

2967_Ch08.indd 261 07/07/2015 10:48

 262 CHAPTER 8 Pharmacogenetics

(a)
Percent of subjects per 0.5 units of activity TPMTL/ TPMTL
TPMTL/ TPMTH
TPMTH/ TPMTH
10

0
0 5 10 15 20
TPMT activity (units/ml RBCs)

(b)
1 2 3 4 5 6 7 8 9 10
TPMT*1
(wild-type)
VNTR

TPMT*3A

VNTR G460A A719G

Ala154THr Tyr240Cys
TPMT*3C

VNTR A719G
Tyr240Cys

Figure 8.4: TPMT: relationship between levels of the enzyme in red blood cells and genotype. (a) Activity
of TPMT in red blood cells (RBCs) from 298 European blood donors. Presumed genotypes for the TPMT gene
polymorphisms are also indicated. TPMTL and TPMTH are designations for alleles resulting in “low” and “high”
activity, respectively. These allele designations were used before the molecular basis for the polymorphism
was understood. (Adapted from Weinshilboum RM & Sladek SL [1980] Am J Hum Genet 32:651–662. With
permission from Elsevier.) (b) Common TPMT alleles. TPMT*1 is the most common allele (wild-type). TPMT*3A,
with two non-synonymous coding SNPs, is the most common variant allele in Caucasian European subjects.
TPMT*3C is the most common variant allele in East Asian subjects. Rectangles represent exons with the gray scale
representing the open reading frame.

eczema, and rheumatoid arthritis. Both 6-mercaptopurine and azathioprine are thiopu-
rine drugs that were originally developed by Gertrude Elion—a pioneering biochemist
who received the Nobel Prize for Medicine in 1988. TPMT is polymorphic (Figure 8.4).
Approximately 1:300 individuals lack TPMT activity. This appears to be a recessive trait
usually due to the presence of one or more non-synonymous mutations. These individuals
will develop severe bone marrow toxicity if prescribed the normal recommended dose of
either 6-mercaptopurine or azathioprine. It is therefore recommended that patients should
be genotyped for TPMT before they are treated with 6-mercaptopurine or azathioprine

2967_Ch08.indd 262 07/07/2015 10:48

 Drug Targets 263

and those predicted to lack activity should either be given a lower drug dose or treated with
another drug that is not metabolized by TPMT.
This is probably the best current example of a pharmacogenetic polymorphism where
genotyping is routinely performed prior to drug prescription and the result used to guide
treatment. The example once again illustrates the potential role for pharmacogenetics in
clinical practice.

Polymorphisms in drug transporters also play a role in

pharmacogenetics
Hepatocytes are the main cell type in the liver and an important site for drug metabolism.
Some drug targets are also located in these cells. Though some drugs can enter the hepa-
tocyte by passive diffusion, most require active transport that is achieved using transporter
proteins located on the cell membrane of the hepatocytes facing the sinusoids which form
the main blood vessels within the liver. Several different transporter families exist. For
anionic drugs, the SLCO family is the main group of transporters. SLCO1B1 has a role
in transport of a range of different drug classes, including statins and antimicrobial agents.
Several common polymorphisms result in decreased transport by SLCO1B1. One of these
appears to increase the risk of statin-induced myopathy, a serious adverse drug reaction
(see Section 8.5), and may also be relevant in the response to methotrexate, a widely used
immunosuppressant and anticancer drug.

8.4 Drug Targets

In addition to the genetic factors affecting the entry and elimination of drugs within the
body discussed in Sections 8.2 and 8.3, genes encoding the targets for drugs may also
encode some common polymorphisms that may affect response to treatment. Depending
on the individual drug, the target gene product may be an enzyme, a receptor, or a mem-
brane transporter. Polymorphisms affecting drug targets may affect the response of the
individual to a particular drug, so the overall effect will be different to that seen for poly-
morphisms in the genes encoding metabolizing enzymes discussed in Sections 8.2 and 8.3.
A higher or lower than normal drug dose may be needed to achieve the desired response.
In some patients, the presence of a particular polymorphism in the gene encoding the
drug target may make the drug unsuitable for clinical use.
Among drug targets, one of the most consistent pharmacogenetic associations reported
has been that for the gene encoding the VKOR with coumarin anticoagulants. Evidence
for other pharmacogenetic associations with drug targets is more limited. However, there
is some positive data regarding an association between the genotype for ADRB2, which
encodes the β2-adrenergic receptor, a target for β-adrenergic receptor agonists that are used
in the treatment of asthma.

The relationship between VKOR and coumarin anticoagulants is one

of the most consistently reported genetic associations involving drug
targets unrelated to cancer
As described in Section 8.2 and in Figure 8.3, VKOR regenerates reduced vitamin K
from vitamin K epoxide produced during γ-carboxylation of coagulation factors in the
blood clotting cascade. The human enzyme is an integral membrane protein within

2967_Ch08.indd 263 07/07/2015 10:48

 264 CHAPTER 8 Pharmacogenetics

the endoplasmic reticulum. VKORC1, the gene encoding the enzyme VKOR in
humans, was only identified quite recently. Coumarin anticoagulants including war-
farin (see Section 8.2) inhibit VKOR enzyme activity directly. A very small proportion
of patients treated with coumarin anticoagulants fail to respond at doses within the
normal range and are termed coumarin resistant. This lack of response appears to be
due to rare non-synonymous mutations in VKORC1 that appear to affect binding of
the anticoagulant to the enzyme. Some coumarin-resistant individuals cannot be given
coumarin anticoagulants successfully, but others can be stabilized on a very high dose.
Polymorphisms in the VKORC1 coding sequences are rare, but several non-coding poly-
morphisms, including -1639G > A  (rs9923231) affect levels of VKORC1 expression.
The A variant, for example, is associated with lower transcription than the G variant.
VKOR protein levels appear to directly determine the required dose of an anticoagulant.
A clear association between -1639G > A genotype and warfarin dose requirement has
been reported in a large number of independent studies, and similar associations for
the other coumarin anticoagulants such as acenocoumarol and phenprocoumon have
also been detected. There is ethnic variation in the frequency of the -1693A variant.
Though the -1639A allele is less common than -1693G in Europeans, it occurs at a high
frequency in East Asians and, as a result, the average dose of coumarin anticoagulant
required in East Asian patients is significantly lower than that required in Europeans.
Approximately 20–30% of variability in coumarin anticoagulant dose requirement can
be explained by VKORC1 genotype. Anticoagulant dosage is routinely individualized by
measurement of coagulation rate, but it is possible that problems with excessive bleeding
or lack of response during the initial treatment period may be decreased by genotyping for
VKORC1 and for CYP2C9 (see Figure 8.3). Genotype and other patient-specific factors
can be used to calculate an individualized dose. The FDA suggests that genotyping for
VKORC1 and CYP2C9 should be considered in patients receiving warfarin treatment, and
clinical trials to assess the value of genotype-based dosing are in progress.

The efficacy of β-adrenergic receptor agonists widely used in the

treatment of allergies may also be genetically determined
β2-Adrenergic receptors are G-protein-linked receptors that are expressed in smooth mus-
cle. Activation involves elevation of intracellular cAMP. ADRB2, a member of the fam-
ily of β-adrenergic receptors, appears to be unusually polymorphic compared with other
adrenergic receptors, especially in the coding sequence. As a result of this, and also because
β-adrenergic agonists are widely used to treat asthma and other common lung diseases,
there has been considerable interest in assessing the relevance of these polymorphisms to
drug response.
There are at least five polymorphisms in the ADRB2 coding region with four resulting in
amino acid substitutions. In addition, there is a single non-synonymous polymorphism
in the region encoding a leader peptide and several upstream polymorphisms close to the
transcription start site (Figure 8.5). Four common haplotypes with frequencies less than
5% have been described but the frequency of these varies between ethnic groups. Among
Europeans, three haplotypes are common. Two of these differ mainly at codon 16 with
one having a sequence encoding glycine, while the other encodes arginine at this position.
The glycine-encoding form can be further subdivided into two haplotypes: one encoding
glutamine at position 27 and the second, which is less common, encoding glutamic acid.
These three haplotypes are also common in other ethnic groups, though overall frequen-
cies differ to those seen in Europeans. A fourth haplotype is also common in African-
Americans. Most clinical studies to date have focused on the codon 16 polymorphisms.

2967_Ch08.indd 264 07/07/2015 10:48

 Drug Targets 265

A F
S L
G L
6
N A
G P
P N 15
16
Q R Arg16Gly
G S
H2N– M H
A
β2-adrenoceptor
P
D
H
D Ile164Thr
V
T
C Y A N
Q N E
27
Q Gln27Glu I T
R A C Arg175Arg
D E C
E Q D
V T F G H F
W N
W M F T F
V K W A T N L I
D R
Extra I V II M III C IV R 175 V N VI Q VII K
L A Q V I E
G M G A H I T W
F E
M H WQ
Y
A Y V H I Y V
V I S M G A I S I S A I L
N L
I L P F V D L P I I S I V
N
I W
L V A V V C L T S F S V P F
F Y G
I A G L T V A G L Y F L N V S
V M S S P V C W G
G F V I V L L F
V N D L T E W I I V F T P N
V L C A A C L M V L V M G T M I L
T I S L I V I V F I I C Y
I A I T V A R V I S Y L G S R
A F D A R T P Intra
K Y R K V K D
F N Y N F L F
E T F K Q A R
R L V A T E K I
Q T
I L A H A
T L K E F
S S R K Q
P Q Q L E
F K Y E K N C
L C L Q E R S
Q F L E N G T
K K C V K Q N
I S L H L S D
D S R Y L D S
K R R G C I L
S L –COOH
R S S E N
E L S Q D D
G G L E L S
R H K G P P
F G A T G V
H T Y N T T
351
V R G G E G Gly351Gly
Q G N N D Q
N D G S F H
L Q Y S V G
S Q V E

Figure 8.5: ADBR2 gene polymorphisms. The figure illustrates main polymorphism in the ADBR2 gene. The
β2-adrenoreceptor (ADBR2) gene is located on chromosome 5q31–q32 and consists of only one exon. There are
five common variants to the coding sequence indicated in the figure. The R16G polymorphism (rs1042713), with
the Arg (R) allele occurs in 25–40% of all individuals and the Q27E variant (rs1042714), with the Glu (Q) allele
occurs in 50–60% of all white individuals. The I164T variant (rs1800888) is associated with a strong and stable
phenotype. Additional common variants of unknown functional relevance are R175R (rs1042718) and G351G
(rs1042719). (From Rosskopf D & Michel MC [2008] Pharmacol Rev 60:513–535. With permission from The
American Society for Pharmacology and Experimental Therapeutics.)

Data on the functional significance of the codon 16 polymorphism is limited but the
arginine-16-containing form of ADRB2 appears to be associated with increased agonist-
induced downregulation and with a poorer response to inhaled short-acting β-agonists.
This response to β2-adrenergic receptor agonists appears to be specific to short-acting

2967_Ch08.indd 265 07/07/2015 10:48

 266 CHAPTER 8 Pharmacogenetics

agonists as a large study involving patients treated with longer-acting agonists failed to
show any difference in outcome based on the codon 16 genotype. In general, it is now
accepted that the codon 16 genotype does appear to affect response to short-acting ago-
nists. However, this may not be important in clinical practice as current treatments mainly
involve use of longer-acting agonists.

8.5 Adverse Drug Reactions

Polymorphisms affecting drug metabolism and the normal function of drug targets can
lead to adverse drug reactions. This may occur if, for example, as discussed above in ear-
lier sections, the plasma drug concentration is unusually high due to slow metabolism or
the effect on the target is greater than expected due to a low concentration of the target
protein. In these cases toxic levels of the drug may build up. However, in addition to these
predictable and generally concentration-dependent effects, other forms of less predictable
adverse drug reactions may occur. These reactions, which are often referred to as idiosyn-
cratic adverse drug reactions, have a less obvious relationship with drug concentration,
may involve more than one genetic risk factor, and often involve the immune system.
Idiosyncratic adverse drug reactions are generally rare, but are quite problematic for a
number of reasons:
• Rarity makes it less likely they will be seen during clinical trials.
• There are generally no appropriate cell- or animal-based models that can be used to
screen new drugs for propensity to cause these reactions early in drug development.
• As they are rare, they can be difficult to recognize in the initial stages. Stopping the
drug will often improve symptoms, but idiosyncratic adverse drug reactions can be
fatal or require major clinical interventions, such as an organ transplant.
There have been a number of high-profile cases during the last 20 years where new drugs
have been withdrawn from the market relatively soon after licensing because of reports of
serious adverse drug reactions.
Idiosyncratic adverse drug reactions can affect a number of different organs, including
the liver, skin, kidney, heart, and muscle, and with some drugs, more generalized hyper-
sensitivity reactions can occur. HLA genotype is an important predictor of some though
not all idiosyncratic adverse drug reactions. The development of genome-wide association
studies (GWAS) has been useful in increasing our understanding of genetic susceptibility
to idiosyncratic adverse reactions because of the limitations of exploring biologically plau-
sible candidates. The basis for many idiosyncratic adverse drug reactions appears to involve
covalent binding of a chemically reactive metabolite or possibly the actual drug to cellular
proteins. This may cause an inappropriate immune response, direct cellular damage, or
affect normal cellular function. The reason why particular drugs are associated with one
type of organ toxicity remains unclear, though the location at which reactive metabolites
are created may explain organ-specific toxicity.

HLA genotype is a potent determinant of susceptibility to several

different types of adverse drug reactions
HLA genotype has been associated with susceptibility to many forms of hypersensitivity
including hypersensitivity to drugs. Hypersensitivity is an inappropriate immune reaction to
an otherwise non-toxic agent. Type 1 hypersensitivity is normally seen as a skin rash and can

2967_Ch08.indd 266 07/07/2015 10:48

 Adverse Drug Reactions 267

be mild or severe. In pharmacologically induced hypersensitivity, the mildest form of reaction

involves a skin rash without additional symptoms and drug withdrawal normally results in
rapid improvement. Blistering skin reactions are considered the most severe manifestation of
drug-induced hypersensitivity reactions affecting the skin and are potentially fatal. A number
of other symptoms are also manifest in severe cases, including fever and effects on organs such
as the liver, kidney, lungs, and bone marrow. However, the extent of involvement of other
organs varies considerably between patients even when treated with the same drug.
Liver injury is a common symptom of drug-induced hypersensitivity and can in
extreme cases be fatal. The genetic associations with HLA are currently among the best-
established inherited risk factors for serious idiosyncratic adverse drug reactions, though
other genetic risk factors are also likely to contribute. It is also important to realize that
HLA genotype is unlikely to contribute to all types of adverse reaction. HLA genes
and alleles, and their relationship with other forms of complex diseases and traits, are
discussed in depth in Chapter 6.

The anti-human immunodeficiency virus (HIV-1) drug Abacavir gives

rise to hypersensitivity in some patients
In terms of hypersensitivity, a particularly well-studied HLA association is that with the
anti-HIV-1 drug abacavir. Hypersensitivity to abacavir affects approximately 5% of all
patients treated. This reaction can be fatal, particularly if a patient is re-exposed to the drug
after initial withdrawal and resolution of the original symptoms. An association between
abacavir hypersensitivity and the HLA-B*57:01–HLA-DRB1*07–HLA-DQB1*03 haplo-
type was initially demonstrated in two small independent studies. The association was
then confirmed in a large randomized controlled trial by Mallal et al. in 2008. The clini-
cal trial demonstrated that over 50% of patients positive for HLA-B*57:01 will develop
symptoms of hypersensitivity if exposed to abacavir, which means that genotyping for this
allele to decide whether to treat with abacavir has a high sensitivity and specificity. Testing
for HLA-B*57:01 is now routine prior to treatment in many countries and patients found
to have the risk allele are offered alternative therapies. In 2008, the FDA added informa-
tion to the drug label recommending that individuals be tested for abacavir sensitivity
prior to prescription. This recent development is another example of the translation of
pharmacogenetic findings into clinical practice.
This is not the only strong association between HLA and adverse drug reactions. A large
number of HLA alleles and haplotypes have now been shown to be associated with adverse
drug reactions (Table 8.2), but there are some limitations. The associations reported for
some of these other adverse drug reactions are not as strong as that reported for abacavir
and HLA-B*57:01, which means that in practice testing is not as sensitive or specific when
predicting toxicity. However, there are some exceptions. The HLA-B*15:02 allele is an
excellent predictor of susceptibility to carbamazepine-induced skin injury in certain eth-
nic groups, including the Han Chinese and Thai populations. HLA typing for this allele
is frequently performed in individuals from these ethnic groups prior to carbamazepine
prescription. Another HLA allele, HLA-A*31:01, is a risk factor for skin injury due to
carbamazepine in both Japanese and Europeans, though the specificity is not as high as
that for HLA-B*15:02.

Drug-induced liver injury is a rare, but clinically important problem

Many different drugs can cause drug-induced liver injury (DILI), with the precise
pattern of injury varying between drugs. DILI reactions are usually classified as

2967_Ch08.indd 267 07/07/2015 10:48

 268 CHAPTER 8 Pharmacogenetics

Table 8.2 HLA associations with serious adverse drug reactions.

Type of reaction Drug involved HLA allele associated

Generalized hypersensitivity/ Abacavir B*57:01
skin rash
Allopurinol B*58:01
Liver injury Co-amoxiclav A*02:01,
DRB1*15:01–DQB1*06:02
Flucloxacillin B*57:01
Lapatinib, ximelagatran DRB1*07:01–DQB1*02:02
Lumiracoxib DRB1*15:01–DQB1*06:02
Ticlopidine A*33:03
Skin injury Carbamazepine, phenytoin B*15:02a
Carbamazepine A*31:01b
a
Han Chinese, Thai, Korean, and other East Asian populations.
b
Japanese and European populations.

hepatocellular when the injury is focused on the hepatocyte and cholestatic when the
damage occurs at the hepatocyte canalicular membrane or further downstream in the
biliary tree (Figure  8.6). The underlying mechanism by which DILI develops may
involve both direct toxic effects by the drug, e.g. involving oxidative stress or cellular
damage, and formation of reactive intermediates resulting in an inappropriate immune

Esophagus

Liver
Left hepatic duct

Right hepatic Common hepatic

duct duct

Stomach
Gall bladder

Cystic duct
Common
bile duct
Pancreatic Pancreas
duct
Duodenum

Figure 8.6: The biliary tree. When the liver cells secrete bile, it is collected by a system of ducts that flow from
the liver through the right and left hepatic ducts to the common hepatic duct, which joins with the cystic duct to
form the common bile duct. Approximately half of the bile formed flows into the duodenum or small intestine
and half into the gallbladder. The stored bile is used to break down fats following the ingestion of food. (From
Ahmed N, Dawson M, Smith C & Wood E [2006] Biology of Disease. Garland Science.)

2967_Ch08.indd 268 07/07/2015 10:48

 Adverse Drug Reactions 269

30

25

20
–log10P

15

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 20 22 X
Chromosome

Figure 8.7: A Manhattan plot for a GWAS of flucloxacillin-induced liver injury. This Manhattan plot
indicates a number of statistically significant associations with flucloxacillin-induced liver injury. In particular,
the –log10P value shows that a number of polymorphisms in the MHC region have genome-wide significance.
The study involved 51 cases of DILI and 282 population controls. The polymorphism (rs2395029) showing the
most significant difference in frequency between cases and controls is in complete linkage disequilibrium with
the HLA class I HLA-B*57:01 allele. (From Daly AK, Donaldson PT, Bhatnagar P et al. [2009] Nat Genet 41:816–
819. With permission from Macmillan Publishers Ltd.)

response. Up to 10% of patients with DILI develop liver failure, which may be fatal
unless a liver transplant is performed, though the majority of patients recover after the
causative drug is discontinued. There is sometimes overlap between hypersensitivity
reactions as discussed above and DILI.
Several different HLA alleles are risk factors for DILI, but the strongest effect for any
HLA allele reported up to the present is that with HLA-B*57:01, which has been
associated with flucloxacillin-induced liver injury (see Figure 8.7). Though this is the
same risk allele as that identified for abacavir hypersensitivity and the overall specific-
ity is high, sensitivity is low because most individuals who are prescribed flucloxacillin
(a commonly used antibiotic, especially in the UK) and are B*57:01-positive do not
appear to develop liver injury. The underlying mechanisms for toxicity with abaca-
vir and flucloxacillin may also differ. Another example where the same HLA alleles
and  haplotypes appear to be risk factors for liver injury induced by different drugs
(Table 8.2) is lapatinib and ximelagatran, two drugs with very different chemical struc-
tures that share the same association with the HLA-DRB1*07:01–DQB1*02:02 hap-
lotype and the development of liver toxicity. A common feature of many of these
associations discussed here is that compounds with very different chemical structures
show the same HLA association.

There are many other susceptibility factors for serious adverse drug
reactions
Based on a number of studies, it appears that liver injury associated with some widely used
drugs such as statins and isoniazid is not HLA-associated. Risk factors for these forms of
DILI may include:
• Genotypes for enzymes affecting drug metabolism, e.g. N-acetyltransferase 2 (NAT2),
which metabolizes isoniazid.

2967_Ch08.indd 269 07/07/2015 10:48

 270 CHAPTER 8 Pharmacogenetics

• Genes relevant to oxidative stress, e.g. superoxide dismutase (SOD2), and genes
relevant to the innate immune response, e.g. signal transducer and activator of
transcription 4 (STAT4), which encodes a transcription factor that regulates T cell
responses.

Adverse reactions to commonly used statins provide a key example of

non-HLA-related adverse drug reactions
Statins are widely prescribed drugs that are highly effective in lowering cholesterol.
However, they can cause a toxic reaction involving muscle pain in some individuals. A
number of different drugs can give rise to myopathy—a general term used to describe
adverse reactions involving the muscle. These reactions can range from muscle weakness
or pain, through myositis where the enzyme creatine phosphokinase is elevated, to rhab-
domyolysis where creatine phosphokinase is greatly elevated and breakdown of muscle
fibers can lead to renal damage. Most cases are not serious and are reversible through drug
withdrawal. Statins are a relatively common cause of muscle toxicity. The ability of differ-
ent statins to cause myopathy varies, but an overall estimate of an incidence of 0.1% has
been made, though some say as many as 2% of all patients encounter muscle pain.
Individual statins have been shown to vary in their ability to cause myopathy, with
cerivastatin being the most toxic followed by simvastatin, lovastatin, pravastatin, ator-
vastatin, and fluvastatin. Drug interactions where patients are taking more than one
drug seem to be an important contributor to statin-induced myopathy. A GWAS on a
subgroup of patients suffering from elevated creatine phosphokinase in a clinical trial
involving use of simvastatin found a signal for a tagged single nucleotide polymorphism
(SNP) (rs419056) close to the SLCO1B1 gene (see Section 8.3) and in complete linkage
disequilibrium with the SLCO1B1*15 valine-174-alanine allele. This valine for alanine
exchange has been shown previously to affect transport activity with several statins. This
association has been confirmed for other statins, but possession of the variant allele only
explains around 18% of the risk. The SLCO1B1 risk (wild-type) allele has been associ-
ated with a four-fold risk in heterozygous individuals and a 16-fold risk in homozygotes
according to some authors. Additional genetic risk factors possibly relating directly to
the muscle rather than drug levels determined by SLCO1B1 may also contribute.

Cardiotoxicity reactions to drugs do not appear to involve an immune

or inflammatory response
Drug-induced cardiotoxicity usually involves prolongation of the QT interval
(Figure 8.8), which is the time taken for one complete cycle of cardiac ventricular depo-
larization and repolarization, and can be measured using an electrocardiogram. A number
of different drug classes can cause this type of toxicity, which can result in sudden death
in susceptible individuals and it is a frequent reason for new drugs in development being
withdrawn before they are licensed. Many of the causative drugs do not act primarily on
the cardiovascular system and range from antimicrobial agents to drugs used in psychia-
try, though drugs such as amiodarone that are used to treat cardiac arrhythmias can also
cause QT prolongation. At present, genes found to predict drug-induced QT prolonga-
tion include those encoding proteins, such as ion channels expressed in heart tissue and
the nitric oxide synthase 1 adaptor protein (NOS1AP), which interacts with nitric oxide
synthase 1 to accelerate cardiac repolarization by inhibition of calcium channels. These
risk factors overlap with those reported for syndromes associated with long QT interval
independent of drug exposure.

2967_Ch08.indd 270 07/07/2015 10:48

 Conclusions 271

(a) One beat

QT
R R

P T P T

Q Q
S S

Long QT
R R

P T P T

Q Q
S S

(b)

Post-ectopic pause Abnormal T/U wave Torsades de pointes

Short Long Short

Figure 8.8: Drug-induced prolongation of the cardiac QT interval and torsades de pointes. (a) The cardiac
depolarization and repolarization cycle. The QT interval is shaded; it starts with the onset of QRS and terminates
at the end of the T wave. This is the time taken for one complete cycle. (b) Some drugs can prolong the QT
interval and induce rapid heart beating, which is seen in the torsades de pointes (TdP; polymorphic ventricular
tachycardia). The abnormal TU wave preceding the TdP indicates cardiac arrhythmia, which can potentially be
fatal. (From Strachan T, Goodship J & Chinnery PF [2014] Genetics and Genomics in Medicine. Garland Science.)

Conclusions
There are now a large number of well-described and replicated pharmacogenetic associa-
tions. Recent developments in genomics such as the widespread application of GWAS
have helped increase knowledge in this area. Despite the large body of knowledge, there
has been less than hoped for progress in the clinical implementation of these find-
ings, with a few important exceptions. Currently, genotyping for HLA-B*57:01 prior
to abacavir prescription and TPMT genotyping or phenotyping prior to azathioprine
or 6-mercaptopurine prescription are the main examples of tests that are performed
routinely in many countries. For many of the other examples described above, the sen-
sitivity of the test may not be sufficient to make genetic testing useful in diagnosis or
treatment choice. As there are no data based on some of these randomized clinical tri-
als, assessing the clinical utility of genetic associations is not yet possible. In the case of
warfarin and other coumarin anticoagulants, several clinical trials to assess the value of
genotyping for CYP2C9 and VKORC1 prior to prescription have now been completed,

2967_Ch08.indd 271 07/07/2015 10:48

 272 CHAPTER 8 Pharmacogenetics

but unfortunately the findings are inconsistent with the two largest studies disagree-
ing on whether genotyping is useful. Even larger studies may be needed to resolve the
discrepancy. There are also additional examples where clinical trials might be helpful,
such as testing CYP2C19 and clopidogrel and CYP2D6 and tamoxifen. There are also
additional examples where patient response to commonly used treatments is variable
and impossible to predict. In these cases, response may have a genetic component. This
includes response to antihypertensive drugs and to aspirin. Clearly, there is much more
work yet to be done in this area, but some of the results so far look promising. The
increasing availability of genome-wide sequencing help us to better understand the
complex genotype–phenotype interactions that exist in pharmacogenetics.
The application of new technologies will speed up the process of genotyping in phar-
macogenetics as well as in other areas of genetic research. The further development of
large collaborative groups and integration of data sets and large collections will enable
larger, better quality studies to be performed. Altogether, this means faster, better,
and more detailed studies. The potential to identify major genetic links in pharma-
cogenetics and develop better pharmacological agents using that information, or to
identify genetic links to an individual’s response to commonly used drugs, and then
produce personalized therapy plans, are only two of the goals that the new genetics
offers. The future use of present and developing technologies holds great possibilities
in pharmacogenetics.

2967_Ch08.indd 272 07/07/2015 10:48

 Further Reading 273

Further Reading
Books Mallal S, Phillips E, Carosi G et  al. (2008)
Coleman MD (2010) Human Drug HLA-B*5701 screening for hypersensitivity to
Metabolism. An Introduction. Wiley-Blackwell. abacavir. N Engl J Med 358:568–579.
Maitland-van der Zee A-H & Daly AK (eds) Marsh S & Van Booven DJ (2009) The increas-
(2012) Pharmacogenetics and Individualized ing complexity of mercaptopurine pharma-
Therapy. Wiley. cogenomics. Clin Pharmacol Ther 85:139–141.
Strachan T, Goodship J & Chinnery PF (2014). Meyer UA (2004) Pharmacogenetics – five
Genetics and Genomics in Medicine. Garland decades of therapeutic lessons from genetic
Science. diversity. Nat Rev Genet 5:669–676.
Niemi M (2010) Transporter pharmacogenet-
ics and statin toxicity. Clin Pharmacol Ther
Articles 87:130–133.
Brodde OE (2008) Beta-1 and beta-2 adreno- Relling MV, Gardner EE, Sandborn WJ
ceptor polymorphisms: functional importance, et  al. (2011) Clinical Pharmacogenetics
impact on cardiovascular diseases and drug Implementation Consortium guidelines for
responses. Pharmacol Ther 117:1–29. thiopurine methyltransferase genotype and thio-
Daly AK (2004) Pharmacogenetics of the purine dosing. Clin Pharmacol Ther 89:387–391.
cytochromes P450. Curr Topics Med Chem Rubboli A, Becattini C & Verheugt FWA
4:1733–1744. (2011) Incidence, clinical impact and risk of
Daly AK & Day CP (2012) Genetic associa- bleeding during oral anticoagulation therapy.
tion studies in drug-induced liver injury. Drug World J Cardiol 26:351–358.
Metab Rev 44:116–126. Scott SA, Sangkuhl K, Gardner EE et al. (2011)
DeGorter MK, Xia CQ, Yang JJ & Kim Clinical Pharmacogenetics Implementation
RB (2012) Drug transporters in drug effi- Consortium guidelines for cytochrome
cacy and toxicity. Ann Rev Pharmacol Toxicol P450-2C19 (CYP2C19) genotype and clopido-
52:249–273. grel therapy. Clin Pharmacol Ther 90:328–332.
Guillemette C, Levesque E, Harvey M et  al. Uetrecht J (2007) Idiosyncratic drug reactions:
(2010) UGT genomic diversity: beyond gene current understanding. Annu Rev Pharmacol
duplication. Drug Metab Rev 42:24–44. Toxicol 47:513–539.
Jonas DE & McLeod HL (2009) Genetic and van Schie RM, Wadelius MI, Kamali F et  al.
clinical factors relating to warfarin dosing. (2009) Genotype-guided dosing of coumarin
Trends Pharmacol Sci 30:375–386. derivatives: the European pharmacogenetics of
Killeen MJ (2009) Drug-induced arrhythmias anticoagulant therapy (EU-PACT) trial design.
and sudden cardiac death: implications for the Pharmacogenomics 10:1687–1695.
pharmaceutical industry. Drug Discov Today
14:589–597.
Online sources
Koren G, Cairns J, Chitayat D et  al. (2006)
Pharmacogenetics of morphine poisoning in http://www.cypalleles.ki.se
a breastfed neonate of a codeine-prescribed This is a useful website which provides informa-
mother. Lancet 368:704. tion of CYP450 gene nomenclature and alleles.

2967_Ch08.indd 273 07/07/2015 10:48

2697_Prelims.indd 2 09/07/2015 09:35
 CHAPTER
9
Cancer as a Complex Disease:
Genetic Factors Affecting
Cancer Susceptibility and
Cancer Treatment

Cancer is a major cause of morbidity and mortality worldwide. Statistics from the USA
and UK reported in 2013/14 showed that 1,690,000 and 331,487 new cancer cases were
diagnosed in the period 2012 and 2011 in the USA and UK, respectively. More than
500,000 American and 159,178 UK citizens die from cancer each year. In the USA and
UK, lung cancer is the leading cause of cancer-related deaths in both men and women,
accounting for 28 and 22% of all cancer mortalities, respectively. In the USA, breast
cancer is the second-leading cause of cancer-related deaths for women (approximately 23
deaths per 100,000 women annually) and prostate cancer is the second-leading cause for
men (approximately 20 deaths per 100,000 men annually), followed by colorectal cancer
for both sexes (15 deaths per 100,000 population annually). In the UK, figures for breast
cancer and prostate cancer are similar to those reported in the USA; however, in the popu-
lation overall bowel cancer accounted for 10% of all cancer-related deaths, whereas breast
and prostate cancer accounted for 7% in 2011.
The examples above are all examples of solid tumors, but it is important to remember
that not all cancers are solid tumors (e.g. leukemias). Some cancers are caused by viruses,

2967_Ch09.indd 275 08/07/2015 14:42

 276 CHAPTER 9 Cancer as a Complex Disease

e.g. human papilloma virus is linked to cervical cancer and human herpes virus 8 is
associated with Kaposi’s sarcoma, which is also common in human immune deficiency
virus (HIV-1)/acquired immune deficiency syndrome (AIDS). The viruses that can cause
tumors are often referred to as oncoviruses.
This chapter will address selected examples of recent advances in the field of cancer
genetics, but will not attempt to cover all cancers and genetic risk factors. However,
before we start looking at the selected few it is worth considering the basic statistics on
other forms of cancer. Table 9.1 presents selected currently available data for cancer
deaths per 100,000 individuals for a range of different countries for men and women
published by the World Health Organization (WHO) in 2014 (based on data from
2008; http://apps.who.int/gho/data/node.main.A864). This shows some interesting
contrasts, both between countries and between males and females. In most cases, the
incidence of death from cancer is lower in women compared with men. The size of this
difference can vary from country to country, with figures being twice as high in men as
in women. In some countries, the figures are very similar for men and women, particu-
larly where the overall incidence is low, for example in Namibia and Kuwait. Looking
in more detail at rates for different cancers, Figure 9.1 illustrates the 10 most common
cancers in men and women in the UK in 2011 as published by Cancer Research UK in
2014 (http://www.cancerresearchuk.org/home/). These clearly show the impact of the
selected examples on the UK population only.
However, we cannot discuss all forms of cancer here. In the selected examples, we will
look at genetic risk factors in three categories: novel genetic risk factors i.e. those that have

Table 9.1 Mortalities from cancer per 100,000 head of population

in 2014.

Country Male Female

Australia 141 93
Austria 154 95
China 182 105
France 183 94
Germany 156 99
Italy 158 91
Japan 150 77
Kuwait 62 70
Namibia 64 50
Panama 111 98
Qatar 101 84
Thailand 115 96
UK 155 114
USA 141 104

2967_Ch09.indd 276 08/07/2015 14:42

 Defining Cancer 277

Breast cancer

Prostate cancer

Lung cancer

Bowel cancer

Uterine cancer

Ovarian cancer

Men
Malignant melanoma
Women
Non-Hodgkin’s lymphoma

Brain tumor

Pancreatic cancer

Kidney cancer

Bladder cancer

Esophageal cancer

Leukemia

Other

0 10,000 20,000 30,000 40,000 50,000 60,000

Incidence

Figure 9.1: Incidence of the 10 most common cancers in women and men in UK in 2011. This figure shows
the reported incidence of each form of cancer in 2011 in the UK. Note that some of the cancers do not occur
in both men and women (for example ovarian and prostrate cancer), or are very rare in one sex compared to
the other (for example, breast cancer is rare in men). In addition, some though common, are not found to be in
the 10 most common for both groups. For example pancreatic cancer is listed in women but does not make the
top 10 in men for this period. Whereas in women where ovarian, uterine, and pancreatic cancer are in the top 10,
bladder, esophageal cancer, and leukemia are not listed. The less common cancers are all included in the group
labeled “other.” Breast cancer accounts for 30% of cancers in women in this figure, while prostate cancer accounts
for 25% of cancers in men, with bowel and lung cancer making up approximately 25% of cancers in both men
(28%) and women (23%). The frequencies for the other listed cancers range from around 5 to 2% in descending
order. (Data from Cancer Research UK.)

2967_Ch09.indd 277 08/07/2015 14:42

 278 CHAPTER 9 Cancer as a Complex Disease

been detected only by genome-wide association studies (GWAS), risk factors detected by
GWAS affecting more than one type of cancer, and risk factors previously detected by can-
didate gene studies that have been confirmed by GWAS. In addition, the role of genetics
in understanding response to therapy and advances in treatment of cancers as a result of
genetic research will also be considered in detail in this chapter.

9.1 Defining Cancer

A cancer or a tumor is by definition an uncontrolled growth of abnormal cells in any
part of the body, that is tissues developing outside the normal regulatory boundaries or
at unrequired levels within those boundaries. Rapid growth is common. Some tumors
follow simple Mendelian patterns of inheritance, but the majority are sporadic. Estimates
of the percentage of total cancers that show Mendelian inheritance vary from 5 to 15%,
though the genes responsible in some Mendelian cases may still be unknown. One well-
studied example is breast cancer. Here, it is well established that inherited mutations in
genes such as BRCA1, BRCA2, TP53, ATM, and CHEK2 account for approximately 25%
of the genetic risk. Current data suggests that these genes are mainly relevant in familial
disease. Examples of other common types of cancer where heritability is relatively high
include colorectal cancer and prostate cancer, with estimates of 35% and 42% heritability
based on twin studies from Scandinavia. Despite these findings, the majority of cancers,
like most other diseases, can be considered to be genetically complex. In the last 20 years,
our understanding of cancer as a complex genetic disease and the role that common poly-
morphisms play in cancer susceptibility has increased. As with other complex diseases,
both genes and the environment play a role. For example, smoking is the main environ-
mental risk factor for lung cancer, but a number of other environmental factors may be
important contributors in cancer development (Figure 9.2). When considering the term
“heritability” it is also important to remember that heritability is not constricted to genetic
­inheritance—other factors are heritable.
Cancer predominantly affects the aging population. For example, as shown in Figure 9.3,
the greatest number of new breast cancer cases in women in the UK is in the 60–65 years
age group and the disease is very rare below age 40 years. This is also true in other devel-
oped countries. This means that with an increasingly aging population, the incidence of
many cancers is increasing. The relatively late age of onset also means that performing
genetic studies on families in cancer is challenging as parents and grandparents of affected
individuals are less likely to be available for testing.
Before considering individual genetic risk factors for cancer in more detail, it is impor-
tant to consider that in cancer, as with infectious disease and unlike most other complex
diseases, we are dealing with two different genomes—that of the normal host and that
of the cancer cells. There will be important differences between the germ-line DNA and
tumor DNA. The differences between germ-line and tumor DNA are not inherited dif-
ferences, but they are differences that develop during carcinogenesis. During this pro-
cess by which a somatic cell is changed to a cancer cell, a variety of changes can occur in
the normal host DNA, including point mutations, loss or gain of DNA, and epigenetic
changes involving altered DNA methylation or other DNA or histone modifications (see
Chapter 12 for more detail on epigenetics). Such changes lead to altered gene expression
with the result that the patterns of both mRNA and protein expression will be differ-
ent in a tumor cell compared with the equivalent somatic cells. Powerful new methods
such as whole-genome sequencing have increasingly enabled the detection and study

2967_Ch09.indd 278 08/07/2015 14:42

 Defining Cancer 279

Occupation

Meat consumption
Women
UV radiation
Men

Infection

Alcohol

Low consumption of
fruit and vegetables

Overweight

Smoking

0 5 10 15 20 25
Percentage

Figure 9.2: Fraction of cancer attributable to selected lifestyle and environmental factors. The percentage
of all cancers where specific risk factors contribute. Men and women are considered separately. The data is for
the UK and is based on information collected for 2010. (Data from Parkin DM, Boyd L & Walker LC [2011] Br J
Cancer 105 (Suppl 2):S77–S81.)

7000 500
450
6000 Female cases
Average number of cases per year

Female rates 400

5000 350
Rate per 100,000

300
4000
250
3000
200

2000 150
100
1000
50
0 0
0-4
5-9
10-14
15-19
20-24
25-29
30-34
35-39
40-44
45-49
50-54
55-59
60-64
65-69
70-74
75-79
80-84
85+

Age at diagnosis (years)

Figure 9.3: Average age at diagnosis of breast cancer. The histogram shows the average number of cases per
year in the UK in 5-year age ranges. The solid line shows the rate per 100,000 women plotted against age. (Data
from Cancer Research UK.)

2967_Ch09.indd 279 08/07/2015 14:42

 280 CHAPTER 9 Cancer as a Complex Disease

of these changes in tumor material. However, the emphasis in this chapter is on cancer
susceptibility and individualized treatment of cancer, not the process by which normal
cells become tumor cells.

9.2 Cancer as a Complex Disease

In general, sporadic cancers, usually defined as cancers occurring in individuals without
a family history of cancer, can be considered as typical complex diseases with both a
strong contribution from the environment and a genetic component. Data from families
indicate a two- to three-fold increased risk in first-degree relatives for a number of
common cancers and there is higher concordance between monozygotic twins compared
with dizygotic twins. Despite this, attempts to identify specific genetic risk factors for
most common (usually sporadic) cancers have been relatively unsuccessful. However, the
recent application of large-scale, high-resolution GWAS has led to considerable progress
in this area.

Early studies of cancer found evidence of genetic associations

with risk
The risk for a number of common cancers is strongly associated with environmental factors
including smoking, occupational exposure to chemicals, and diet. Candidate gene-based
case control studies performed between 1985 and early 2000 led to the discovery of a large
range of apparently significant genetic associations with polymorphisms in genes that con-
tribute either to the activation or to the detoxification of proven or possible carcinogens.
However, many of the initial findings were not reproduced in subsequent studies. Most
of these studies shared the same limitations, including use of relatively small numbers of
cases and controls, and the selection of single genes and single polymorphisms for study.
There are particular problems in studies based on small numbers, which are related to
statistical power. In these early studies, false-positive results were common and false nega-
tives equally so. By restricting the analysis to selected candidates, no thought was given to
other polymorphisms. Though the polymorphisms studied were frequently biologically
plausible candidates in relation to the cancer under investigation and in some cases bone
fide associations were identified, most of these early association studies lacked consistency.
Improvements in study design from 2000 onwards, particularly the use of better
characterized case groups, larger numbers of cases and controls, and the introduction of non-
hypothesis-constrained analysis (e.g. GWAS), has helped to identify gene polymorphisms
involved in susceptibility and resistance to cancer. One of the important features of recent
work has been increased international collaboration to achieve very large case cohorts,
e.g. the COGS (Collaborative Oncology Gene-environment Study) consortium that has
focused specifically on identifying genetic risk factors for breast, ovarian, and prostate
cancers (http://www.cogseu.org). Coupled with major technological advances, this type
of approach has revolutionized the search for alleles associated with cancer as it has in
many other diseases. Overall, it appears that the role of the polymorphisms identified
in early studies may be less important than originally suggested, with a few notable
exceptions. These include, for example, an association between aerodigestive cancers and
alcohol dehydrogenase genes and an association between bladder cancer susceptibility, and
polymorphisms in the GSTM1 and NAT2 genes that code for enzymes that contribute to
the detoxification of chemicals (see Section 9.5).

2967_Ch09.indd 280 08/07/2015 14:42

 Genetic Risk Factors for Particular Cancers Detected by Gwas 281

GWAS has revolutionized the search for cancer-promoting alleles in

non-familial cancers
The application of GWAS to cancer has resulted in the identification of more novel alleles that
are not obvious candidate risk alleles for cancer, but in many cases are still biologically plau-
sible. As with other complex diseases where GWAS has identified new risk factors, the impact
tends to be small with typical odds ratios (ORs) being lower than 2. In addition, GWAS has
identified large collections of different risk alleles in different forms of cancer as well as other
diseases. It is generally not feasible, based on current knowledge, to predict an individual’s risk
of developing a cancer by genotyping for these newly identified risk alleles and mutations, but
the detection of novel risk variants is important in understanding the biological basis of these
diseases and may also enable new treatments to be developed in the future.
In cancer, as with most complex diseases, good quality phenotypic data is very important in
studies aimed at identifying genetic risk factors. Tumors can be categorized according to their
phenotype. Tumors in particular organs show considerable biological heterogeneity and the
cells affected may also differ from one individual to another. Detailed information on tumor
pathology is particularly useful in studies of cancer. Other types of phenotypic data that are
collected in studies typically include age of onset, clinical stage, and response to treatment.

9.3 Genetic Risk Factors for Particular

Cancers Detected by Gwas
Various cancers have featured strongly in the list of diseases studied by GWAS approaches.
Breast cancer, lung cancer, and prostate cancer have been particularly well studied, and a
few specific examples of individual genetic risk factors for these cancers will be considered
here to illustrate the value of the GWAS approach.

GWAS has identified a number of biologically plausible genetic risk

factors for breast cancer
Though there are clear Mendelian cases of breast cancer, most cases (75–80%) are sporadic
with no family history. It has been observed that the incidence of breast cancer is higher in
women with affected siblings and the onset in this group is often earlier, pointing to genetic
susceptibility. However, because it is a common condition, affecting up to 1:12 women in the
developed world, there is often concordance for breast cancer in families by simple coincidence
rather than by inheritance. Therefore, it is important to be aware of this before embarking
on any studies of breast cancer. It is also important to note that breast cancer is not restricted
to women and male breast cancer (not discussed here) does occur, though it is relatively rare.
Breast cancer was one of the first diseases to be studied by GWAS and since 2007 a series
of independent GWAS have identified more than 70 different genetic signals [mostly
tagged single nucleotide polymorphisms (SNPs)] that are significantly associated with risk
of breast cancer, though the precise genes relating to these tagged SNP signals have not
yet all been identified. Of those that have been identified, the majority that contribute
to the increased risk of sporadic breast cancer code for proteins that have functions in
cell growth and cell signaling (see Table 9.2 for further examples). Of course it is not the
genes themselves that increase the risk, as these are present throughout the population,
but it is possession of specific sequence variations in the form of mutations and polymor-
phisms that confer risk. The FGFR2 gene on chromosome 10q26 codes for the fibroblast

2967_Ch09.indd 281 08/07/2015 14:42

 282 CHAPTER 9 Cancer as a Complex Disease

Table 9.2 Selected examples of identified potential susceptibility loci for breast cancer and the
gene function.

Gene Biological function

Fibroblast growth factor receptor Receptor that has tyrosine kinase activity and is located on
2 (FGFR2) (OR = 1.26) the plasma membrane; binds fibroblast growth factors that
have a diverse range of biological effects including roles
in angiogenesis, mitogenesis, cellular differentiation, cell
migration, and tissue injury repair
TOX high mobility group Codes for a protein containing an HMG box that may
(HMG) box family member modify chromatin structure by bending and unwinding
3 (TOX3) (also referred to DNA
as TNRC9 and CAGF9)
(OR = 1.11)
Estrogen receptor 1 (ESR1) Codes for nuclear receptor for estrogen that acts as a
transcriptional activator of a range of target genes following
estrogen binding
Mitogen-activated protein kinase Codes for a serine/threonine protein kinase enzyme that
kinase kinase 1, E3 ubiquitin is part of some signal transduction cascades, including the
protein ligase (MAP3K1) ERK and JNK kinase pathways, and the NF-κB pathway
(OR = 1.13)
Caspase 8 (CASP8) Caspase 8 is a member of the cysteine-aspartic acid protease
(caspase) family, and is involved in the programmed cell
death induced by Fas and various apoptotic stimuli
Cytochrome c oxidase assembly Cytochrome c oxidase is a mitochondrial protein complex
homolog 11 (COX11) that transfers electrons from reduced cytochrome c to
molecular oxygen; COX11 codes for a protein which is not
a structural subunit, but may be a heme A biosynthetic
enzyme involved in cytochrome c oxidase formation
Solute carrier family 4, sodium This plasma membrane protein transports sodium and
bicarbonate co-transporter, bicarbonate ions; these processes may be important in
member 7 (SLC4A7 ) regulating intracellular pH
RAD51 homolog B (RAD51B) RAD51 family members are proteins with roles in DNA
(also known as RAD51L1; repair by homologous recombination
REC2; R51H2)

growth factor receptor 2 and in European populations has been shown to have one of the
strongest genetic associations of all the risk variants reported up to now. However, the per
allele odds ratio of approximately 1.26 even for the SNP showing the strongest association
(rs2981578) is relatively modest, though statistically significant. As we may expect, there
are several other SNPs within the gene showing roughly equivalent effects and the molecu-
lar basis for this association is still not completely clear. However, the SNP rs2981578 is in
a site for transcription factor binding that is conserved among species and may affect gene
expression. It is important to note that the SNPs in FGFR2 that were found to be associ-
ated with breast cancer risk were all more strongly related to estrogen receptor-positive
cancer. Differences in associations depending on the estrogen receptor status of the tumor
have been found to extend to other genetic risk factors for breast cancer, though one of

2967_Ch09.indd 282 08/07/2015 14:42

 Genetic Risk Factors for Particular Cancers Detected by Gwas 283

those listed in Table 9.2, TOX3, which encodes a transcription factor, shows a similar risk
for both estrogen receptor-positive and -negative tumors.

Novel insights into lung cancer involving the target for nicotine were
detected by GWAS
Prior to the advent of GWAS, genetic susceptibility to lung cancer had been studied quite
widely using the candidate gene case control approach, though there was considerable
inconsistency between reports. The impact of this cancer worldwide and the availability of
existing DNA collections enabled GWAS studies to be performed and data collected from
large numbers of cases. However, the number of loci found to be significantly associated
with the disease has been smaller than for some other cancers, including breast cancer
(summarized in Table 9.3). The most significant association reported for lung cancer is in
a region of chromosome 15 where several genes encoding nicotinic acetylcholine receptors
(CHRNA3 and CHRNA5), the targets for nicotine, are located. It has been proposed that
these genes contribute to an increased risk of becoming addicted to nicotine through the
expression of these receptors in the brain. In addition, expression of these receptors in the
epithelial cells in the lung has been implicated in the development of lung cancer. More
recently, studies of lung cancer in individuals who have never smoked have shown that
this region is not associated with disease development. This observation provides further
evidence for a direct effect of the presence of nicotine in the lungs, which is unlikely to
happen in non-smokers, playing a key role in lung cancer development.

A large number of genetic risk factors for prostate cancer have been
revealed by GWAS
Prostate cancer is another cancer that has been well studied by GWAS. Up to now, GWAS
studies on this cancer have led to the identification of the largest number of genetic risk
factors for any common cancer, accounting for approximately 30% of total familial risk of
the disease. Some of the most established associations with specific gene loci are listed in
Table 9.4. One of the stronger associations described is with MSMB, which codes for the
prostate-specific protein β-microseminoprotein (PSP94). The gene product is secreted into
seminal plasma after synthesis in the prostate epithelial cells. MSMB appears to function

Table 9.3 Selected potential susceptibility loci for lung cancer and their function.

Gene Biological function

CHRNA3/5 These genes encode nicotinic acetylcholine receptor subunits; these
receptors are members of a superfamily of ligand-gated ion channels
that mediate fast signal transmission at synapses
BCL2-associated BAG6 codes for a nuclear protein that is cleaved by caspase 3 and
athanogene 6 (BAG6 ) involved in the control of apoptosis; MSH5 codes for a member of
[also called BAT3/mutS the mutS family of proteins that are involved in DNA mismatch
homolog 5 (MSH5)] repair and meiotic recombination
TERT–CLPTM1L More general risk factor for several cancers (see Table 9.4)
Tumor protein p63 This gene codes for a member of the p53 family of transcription
(TP63) factors

2967_Ch09.indd 283 08/07/2015 14:42

 284 CHAPTER 9 Cancer as a Complex Disease

Table 9.4 Selected potential susceptibility loci for prostate cancer and their function.

Gene Biological function

β-Microseminoprotein Prostate-specific protein that is secreted into seminal plasma
(MSMB) (PSP94 )
Kallikrein 3 (KLK3) Codes for prostate-specific antigen (PSA), a protein produced
exclusively by the prostate
NKX3.1 (human Androgen-regulated transcription factor that is specific to the
homolog of the prostate
Drosophila NK3 gene)
Integrin α6 (ITGA6 ) First integrin gene to be associated with human cancer susceptibility
Hepatic nuclear Transcription factor shown previously to be associated with maturity
factor-1β (HNF1B) onset diabetes of the young (MODY); haplotypes associated with
diabetes appear to be protective against prostate cancer

as a tumor suppressor gene, it is believed that the SNP is associated with prostate cancer
risk and it affects transcription, possibly leading to decreased expression. Several other
genes that are expressed in the prostate only, are also risk factors. Though the number
of different loci associated with prostate cancer is relatively high, the biological basis of
these associations is not as well understood as for some of the breast cancer risk factors
discussed previously. Several separate loci on chromosome 8q24 have been identified that
may also contribute to prostate cancer risk (see Figure 9.4). It is possible that the genetic
risk is outside the exome as a large amount of genetic variation is encoded within intronic
sequences. It is also possible that epigenetics plays a significant role in this and some other

PrCa5 PrCa2/CLL PrCa4/Breast PrCa3/CR PrCa1 Bladder

MYC

rs10086908 rs2456449 rs13281615 rs4242382 rs9642880

rs13254738 rs620861 rs7014346

rs16901979 rs6983267

Figure 9.4: Multiple cancer susceptibility loci on 8q24 defined by cancer GWAS. Approximate locations of
GWAS-reported cancer susceptibility loci are indicated by vertical arrows on the linkage disequilibrium pattern of
the 1000 Genomes Project “CEU” (Northern Europeans from Utah) data (November 2010 release, chr8:127,878–
128,880 kb genomic region, reference build 37.1). The arrowheads indicate probable recombination hotspots as per
HapMap 1 and 2. Five distinct regions have been associated with prostate cancer risk (regions 1–5). Region 3 is also
conclusively associated with colorectal cancer. Region 4 also harbors a breast cancer susceptibility locus rs13281615,
and a bladder cancer susceptibility locus rs9642880 is telomeric to region 1 and approximately 30 kb centromeric to
the MYC oncogene. (From Chung CC & Chanock SJ [2011] Hum Genet 130:59–78. With permission from Springer.)

2967_Ch09.indd 284 08/07/2015 14:42

 General Cancer Risk Loci Detected by Gwas 285

cancers. Finally, some but not all of the associations appear to be with polymorphisms that
act as risk factors in several cancers and they are not organ-specific risk alleles confined to
one organ or one system (see Section 9.4).

9.4 General Cancer Risk Loci Detected by Gwas

A number of different genome regions have been found to associate with susceptibility to
more than one form of cancer (Table 9.5). The major histocompatibility complex (MHC)
on chromosome 6p21.3 has long been established as a region that includes risk alleles for
several cancers. More recent GWAS have identified other chromosomal regions that are
associated with a general increase in risk of cancer. Two examples that will be considered
in more detail are the 8q24 region, discussed in Section 9.3 in relation to prostate cancer,
and the region 5p15.33, which includes the TERT and CLPTMIL genes.
The 8q24 region is relatively close to the MYC oncogene, but the SNPs associated with
cancer are over 300 kb away from this gene, are not in linkage disequilibrium with SNPs
within MYC, and there are no other known genes nearby. There are five independent loci
in this region associated with prostate cancer, and some of these loci are also associated
with chronic lymphocytic leukemia, colorectal cancer, and breast cancer. There are also
additional loci associated with bladder cancer and ovarian cancer in this area. It is possible
that the disease associations relate to long-range interactions through DNA looping with
the MYC oncogene. Some evidence that this might occur has been obtained.
Two genes, TERT and CLPTMIL, are located in the 5p15.33 region. TERT encodes a
subunit of the telomerase gene and has been associated with some rare genetic diseases in
addition to cancer. Telomerase and the control of telomere length are intimately linked to
the process of tumor genesis in humans. Telomerase appears to contribute to tumor pro-
gression and metastasis by activation of the glycolytic pathway and suppression of cancer

Table 9.5 Selected potential general susceptibility loci for cancer.

Chromosomal region Biological function and types of association

6p21 (HLA region) Associations with pancreatic cancer, non-Hodgkin’s lymphoma,
nasopharyngeal cancer, and hepatocellular carcinoma; some, but not
all, relate to cancers where viral infection is known to contribute to
disease etiology
8p24 (in general area of Associations with prostate, breast, colorectal, bladder, and ovarian
MYC oncogene) (OR in cancer together with CML reported; may involve long-range
breast cancer = 1.08) interactions with MYC oncogene
5p15.33 Associations with lung adenocarcinoma, pancreatic, testicular, bladder,
(TERT–CLPTM1L) brain, and testicular cancer, as well as basal cell carcinoma, melanoma,
cervical cancer, prostate cancer, breast cancer, and several leukemias;
associations may involve increased or decreased susceptibility for the
same polymorphism depending on the individual cancer; effects may
relate to the telomerase gene, but not proven
11q13 Associations with prostate, renal, and breast cancer reported;
biological basis unclear

2967_Ch09.indd 285 08/07/2015 14:42

 286 CHAPTER 9 Cancer as a Complex Disease

cell differentiation. Abnormal telomere length has also been demonstrated in many cancers.
The biology of CLPTMIL is less well understood, but there are limited data suggesting a
role for involvement in apoptosis following genotoxic stress. The cancers that show asso-
ciations with the TERT–CLPTMIL region include lung, brain tumors involving glial cells
(glioma), bladder, testicular, and pancreatic cancer together with basal cell carcinoma and
melanoma, which affects the skin. The relationship between polymorphisms in this region
and susceptibility to particular cancers is quite complex. A single SNP, rs401681, has been
identified as a risk factor for lung cancer and basal cell carcinoma. This SNP has also been
suggested to be protective against another skin cancer, melanoma, and pancreatic cancer.
For lung cancer, several different SNPs across the entire 100 kb region show associations.
One of these, rs2736100, lies within TERT and appears to be a risk factor for lung cancer in
both smokers and non-smokers, with the effect strongest in adenocarcinoma a tumor type
that is particularly common in non-smokers. It is also a risk factor for glioma.
Discovery of these common genetic risk factors for various cancers is an interesting devel-
opment and should add to our understanding of cancer biology. It is useful to consider
the situation in research into the genetics of immune-mediated diseases where many of the
risk loci are shared between diseases. This may indicate common pathways in disease and
illustrates how understanding genetics of disease, even when the risk ratios are small, can
help us to understand disease pathogenesis.

9.5 Previously Established Cancer Risk Factors

Confirmed by Gwas
Candidate gene case control studies on cancer predated GWAS by approximately 20 years
and were generally unsuccessful in detecting robust genetic associations for the reasons
cited above and elsewhere in this book. Early studies also expected (mostly incorrectly) to
identify alleles with large effects. The cost of genetic testing and lack of information about
the human genome and gene polymorphisms were major restrictions for geneticists seeking
associations in complex disease. Consequently, the literature is generously peppered with
papers reporting apparently plausible associations that were not confirmed. Despite these
limitations, occasionally studies struck gold and some associations originally reported
in candidate gene association studies have been subsequently confirmed in GWAS. The
cancer-related genes detected as risk factors in these earlier studies are generally those
concerned with carcinogen metabolism and so are different biologically to, for example,
those identified recently as risk factors for breast and prostate cancer. In this section,
polymorphisms relevant to carcinogen metabolism and to risk of upper aerodigestive
cancer and bladder cancer will be considered (Table 9.6).

Alcohol, smoking, and chemical exposure increase the risk of cancer

Cancer of the oral cavity, larynx, and esophagus
Upper aerodigestive cancers comprising squamous cell carcinomas of the oral cavity, phar-
ynx, larynx, and esophagus are known to be associated with both alcohol consumption
and tobacco smoking. The possibility that polymorphisms in the genes encoding alcohol
dehydrogenases, which convert alcohol to acetaldehyde, and aldehyde dehydrogenases,
which convert acetaldehyde to acetate, could affect susceptibility to a large number of
diseases associated with high intake of alcohol has received considerable attention. Up to
the present, the main association reported by several independent research groups is that

2967_Ch09.indd 286 08/07/2015 14:42

 Previously Established Cancer Risk Factors Confirmed by Gwas 287

Table 9.6 Genetic associations relevant to carcinogen metabolism detected by GWAS and candidate
gene approaches.

Gene Associations
Alcohol dehydrogenases Both contribute to ethanol metabolism; SNPs within these genes
ADH1B and ADH7 have been shown to contribute to risk of upper aerodigestive
cancer by both GWAS and candidate gene studies
N-acetyltransferase 2 Polymorphisms in this gene associated with absence of enzyme
(NAT2) activity are risk factors for bladder cancer development in both
GWAS and candidate gene studies
Glutathione S-transferase Null allele is associated with increased risk of bladder cancer
M1 (GSTM1)

individuals positive for variant alleles in two alcohol dehydrogenase isoforms, ADH1B and
ADH7, show a decreased risk of upper aerodigestive cancer. Original findings based on
candidate gene case control studies have now been confirmed by GWAS.
NAT2 and bladder cancer
Tobacco smoking is an important risk factor for development of bladder cancer.
N-acetyltransferase 2 (NAT2) is an enzyme that conjugates certain chemicals (including
some prescribed drugs) with acetyl groups. Typical substrates include the aromatic amines
that are found in tobacco smoke. It has been known since the 1950s that NAT2 activity is
absent in approximately 50% of individuals, who are often referred to as slow acetylators.
The genetic basis of this recessive defect is now well understood with a number of poly-
morphisms in the coding gene (NAT2), including several non-synonymous SNPs, shown
to be associated with absence of activity. From the late 1970s onwards, there were isolated
reports that slow acetylators were over-represented among cases of bladder cancer. A recent
large well-controlled study in the USA found that slow acetylators who smoked heavily
were over-represented among bladder cancer cases compared with those with one or two
wild-type alleles, though the overall increase in risk was small. The association between
tagged SNPs predicting the NAT2 slow acetylator phenotype and increased risk of bladder
cancer in smokers has been confirmed by GWAS.
In 1982, a small study in the UK based on cases of bladder cancer among former workers
in the dye industry, who were likely to have been exposed to the carcinogen benzidine,
showed that the slow acetylator phenotype was significantly over-represented. However,
slow acetylators without occupational exposure to benzidine also appeared to have a
slightly increased risk of developing bladder cancer. The occupational exposure reported in
the 1982 study dated back to the 1950s. It is likely that subsequent improved regulations
on occupational exposure to chemicals mean that the currently lower incidence of bladder
cancer, which is down by 14% in women and 18% in men in the UK, may be partly due
to these regulatory changes. Nevertheless, the slightly increased risk of this tumor among
slow acetylators continues to be reported.
GSTM1 and bladder cancer
There is also report of a relationship between the GSTM1 genotype and increased risk of
bladder cancer. The GSTM1 gene encodes glutathione S-transferase M1—an enzyme that
conjugates and detoxifies chemicals. Approximately 50% of many populations lack this
enzyme because they are homozygous for a large deletion (null allele) in the GSTM1 gene.

2967_Ch09.indd 287 08/07/2015 14:42

 288 CHAPTER 9 Cancer as a Complex Disease

There are a relatively large number of published studies showing that those homozygous
for the null allele are at increased risk of bladder cancer development. This observation was
confirmed in the same GWAS study from the USA as that mentioned above for NAT2,
though the overall effect for GSTM1 was smaller, with a 1.28-fold increased risk for those
predicted to have two null GSTM1 alleles on the basis of genotyping using tagged SNPs
specific for the null variant compared with other genotypes. However, a separate GWAS
has shown that the GSTM1 null genotype is a risk factor for bladder cancer in non-
smokers only. This is a slightly unexpected finding, but there is at least partial agreement
between the early candidate gene studies and more recent GWAS.
In summary, the importance of polymorphisms in genes relevant to carcinogen metabolism
as risk factors for cancer has been overestimated in the past with a small number of
exceptions, including those discussed above. There are a number of possible explanations
for this smaller than anticipated role, one of which is redundancy in biological systems,
whereby the existence of a number of different proteins with an ability to carry the same
function makes up for the deficiency created by the non-expressed or absent gene. This
compensation acts as a safeguard against extinction. It is possible that the biological effect
(the trait or disease) is only seen when the system is stressed to a point where compensation
alone is not enough to make up for the deficiency.

9.6 Individualizing Drug Treatment Based

on Tumor Genotype
As discussed in detail in Chapter 8, genetic factors are important in determining response
to drug treatment. In particular, some patients treated for cancer with drugs such as
6-mercaptopurine and irinotecan will develop toxicity, but this can be predicted by geno-
typing prior to treatment. In treating cancer, however, there is an added dimension. For
a number of drug treatments, especially those developed recently, the tumor genotype or
phenotype needs to be considered in addition to the germ-line genotype of the patient. In
the case of the newer targeted therapies, knowledge of the tumor genotype is essential to
determine whether the patient will benefit from the treatment. Targeted anticancer drugs
can be subdivided into small molecules similar to traditional drugs and antibodies. A few
representative examples of each class where tumor phenotype or genotype is determined
before initiation of treatment will be considered in more detail below.

Newly developed drugs inhibit the function of mutated proteins

in cancer cells
A well-established example of targeted therapy relates to the use of endocrine hormone
therapy for breast cancer. To benefit from this treatment, the tumor needs to express
the estrogen receptor-α (estrogen receptor-positive phenotype). The original drug used to
treat estrogen receptor-positive breast tumors was tamoxifen, though specific aromatase
inhibitors are now used in preference to tamoxifen. In contrast to estrogen receptor-
positive breast cancer, it is generally accepted that there is little benefit in treating estrogen
receptor-negative breast cancer with hormone therapy. This is an early example of targeting
of therapy to a particular cancer phenotype (phenotype-guided cancer treatment).
Approximately 95% of cases of chronic myelogenous leukemia (CML) are positive for
leukemic cells that have undergone a chromosomal rearrangement whereby there is recip-
rocal translocation between chromosomes 9 and 22, which is known as the Philadelphia

2967_Ch09.indd 288 08/07/2015 14:42

 Individualizing Drug Treatment Based on Tumor Genotype 289

9 22 der (9) Ph1

BCR BCR–ABL1

ABL1 ABL1–BCR

Figure 9.5 The Philadelphia chromosome. The Philadelphia chromosome is created by a translocation
between chromosome 9 and chromosome 22. The transformed chromosomes are labeled as “der (9)” for
chromosome 9 and “Ph1” for chromosome 22. Ph1 is the Philadelphia chromosome. The translocation joins exons
of the breakpoint cluster region (BCR) gene and the ABL1 oncogene. (From Strachan T, Goodship J & Chinnery P
[2014] Genetics and Genomics in Medicine. Garland Science.)

chromosome. Chromosome 9 becomes longer while chromosome 22 is shortened in this

process. This rearrangement results in a hybrid gene that encodes a fusion protein of ABL,
which codes for a protein kinase, with the breakpoint cluster region or BCR, termed BCR–
ABL (Figure 9.5). This results in a constitutively active tyrosine kinase, with much higher
enzyme activity than that observed for the normal ABL gene product, and this in turn
results in increased and poorly controlled cell proliferation. The drug imatinib was devel-
oped to specifically inhibit the BCR–ABL tyrosine kinase and has proved to be extremely
successful as a treatment for CML. The Philadelphia chromosome targeted by imatinib
is present in the vast majority of CML cases, so no additional test would normally be
required to assess suitability of imatinib treatment. The main limitation of imatinib is that
resistance can develop, but in such cases new tyrosine kinase inhibitors are now available
and can provide suitable alternatives to imatinib. Imatinib is also an inhibitor of the KIT
gene product, another receptor with tyrosine kinase activity. The c-Kit protein encoded
by KIT is overexpressed on the cell surface of gastrointestinal stromal tumor cells and
imatinib is a useful treatment for many, though not all, cases.
Another tyrosine kinase inhibitor, gefitinib, targets the epidermal growth factor receptor
(EGFR). This drug was originally developed as a general treatment for non-small cell lung
cancer (NSCLC), but was subsequently found to be effective only when the EGFR gene
was mutated at certain positions. In practice, few NSCLC tumors are found to be positive
for these mutations and the prescription of gefitinib is only recommended if tumors are
found to be positive by DNA sequencing for these activating mutations. Similarly mela-
nomas may have mutations at certain positions in the BRAF gene. The drug vemurafenib
interrupts the B-Raf/MEK step on the B-Raf/MEK/ERK cell signaling pathway leading
to apoptosis or programmed cell death, but only if the BRAF gene in the tumor has the
common V600E mutation. As this drug may stimulate growth of melanoma cells with the
wild-type or other BRAF sequence, determining the BRAF   V600E genotype of the tumor
is essential prior to use of vemurafenib.

Specific antibodies can target tumor-specific proteins and inhibit

tumor growth
Tumor cells often express different proteins on their cell surfaces compared with non-
malignant cells and specific targeting of these proteins with antibodies has been used as

2967_Ch09.indd 289 08/07/2015 14:42

 290 CHAPTER 9 Cancer as a Complex Disease

an experimental technique ever since methods to prepare monoclonal antibodies in mice

became available. Initial attempts to treat human cancers with these antibodies were gen-
erally unsuccessful, but the development of chimeric and humanized antibodies has
increased success rates. To predict the likelihood of a response to antibody treatment, testing
of tumor material by either phenotypic or genotypic methods is often performed before use.
Treatment with targeted antibodies is often combined with conventional cancer treatments.
Trastuzumab is the most widely used targeted antibody treatment in cancer and was the
first drug to be approved by regulatory authorities for use in the treatment of advanced
cases of breast cancer in association with a diagnostic test that determines whether the
tumor expresses the target protein HER2 or ERBB2, encoded by the ERBB2 gene, and
is thus responsive to this monoclonal antibody. ERBB2 is a cell surface growth factor
receptor that is overexpressed in 20–25% of invasive breast cancers. When combined with
conventional chemotherapy, trastuzumab treatment of ERBB2-positive tumors results
in an objective response rate of approximately 20% for an average of 1 year. Though
originally licensed for treatment of advanced disease, recently there have been reports of
good responses in ERBB2-positive early-stage tumors. Currently, the licensed tests for
ERBB2 expression involve analysis of tumor material by either immunohistochemistry
or fluorescence in situ hybridization (FISH). A major limitation of trastuzumab treatment
is that patients generally become resistant after or during treatment at about one year. The
underlying mechanism for this resistance is still not completely understood.
Cetuximab is a monoclonal antibody that targets EGFR. EGFR is a member of the same
group of growth factor receptors as HER/ERBB2. It was developed as a treatment for
advanced colorectal cancer where other treatments had failed, and is only suitable for use
as a treatment for tumors that express EGFR at a high level. It is also now used as a treat-
ment for advanced head and neck cancer. Receptor expression is assessed by immunohis-
tochemistry. However, in addition to assessing expression of EGFR in the tumor prior to
treatment, the tumor KRAS gene sequence is also determined. Certain mutations in the
KRAS gene are common in colorectal cancer and result in constitutive KRAS activity. In
these cases, blocking EGFR is ineffective at preventing cell proliferation and cetuximab is
therefore not an effective treatment.

Epigenetic changes in the tumor involving methylation may affect

response to conventional drug treatments
In studies on the clinical response of gliomas to treatment with the drugs carmustine
(BiCNU®) and temozolomide, which modify DNA by alkylation, it was found that the
best outcome was seen in patients with tumors where the promoter region of the DNA
repair gene O6-methylguanine-DNA methyltransferase (MGMT) was methylated, lead-
ing to absence of expression or gene silencing. It may be possible to inhibit MGMT where
methylation has not occurred as a means of improving treatment outcome. Analysis of
methylation patterns in tumors may therefore be useful in determining the most appro-
priate treatment. In breast and ovarian tumors, the BRCA1 gene, which codes for another
DNA repair protein, may also be subject to methylation and silencing. There is some evi-
dence that a better response to treatment with poly(ADP) ribose polymerase inhibitors
is seen in patients with methylated BRCA1.

Gene expression profiling may enable personalized cancer treatment

The examples discussed in this section up to now have focused on single genes. However,
increasingly there are possibilities for individualizing patient treatment for cancer on the

2967_Ch09.indd 290 08/07/2015 14:42

 Individualizing Drug Treatment Based on Tumor Genotype 291

basis of the transcriptional profile of the tumor for expression of a number of genes. Clinical
trials of several different gene expression profiling protocols for breast tumors have been
performed. The number of different genes whose expression is profiled is typically between
20 and 70, and the main use of such a profile is currently to assist with decisions on whether
patients should receive chemotherapy in addition to other treatments, such as endocrine
therapy and radiotherapy. Use of this profiling is currently under investigation in the treat-
ment of several cancers, and in the future, designing individual care plans based on ques-
tions, such as which chemotherapeutic agents to apply, whether or not to use radiotherapy,
and which other drugs to prescribe, could become part of standard clinical practice.

Before we close the chapter on cancer it is important to recognize that

there are many forms of this disease
As stated at the beginning of the chapter, it is not possible to discuss all the different types
of cancer in one chapter. Even with this restriction, the chapter gives only a snippet of cur-
rent data based on three selected common examples. However, there are other examples
and other data are available. Cancer patterns are changing. Some forms are becoming
more common, while others are becoming less common. This is illustrated for 20 com-
mon cancers in men and women in the Figures 9.6 and 9.7. Though the data presented

Percentage change, 2000–2002 versus 2009–2011

Decreasing Increasing
All cancers 7%
Breast 7%
Lung 13%
Bowel 6%
Uterus 23%
Ovary –10%
Malignant melanoma 39%
Non-Hodgkin’s lymphoma 15%
Brain tumors –4%
Pancreas 11%
Kidney 38%
Leukemia 4%
Cervix 3%
Bladder –14%
Esophagus –8%
Stomach –28%
Oral 33%
Myeloma 8%
Thyroid 65%
Liver 33%
Vulva 10%

Figure 9.6 Percentage change in cancer frequency in women 2000–2002 versus 2009–2011. The figure
shows how frequencies of some common cancers have changed in the new millennium. There are some
differences compared with men (see Figure 9.7). These are mostly based on anatomy, but some are surprising.
Thus, while bladder and stomach cancer have fallen in both groups, lung cancer has risen in women and fallen in
men. These clearly reflect lifestyle changes, whereas others are more difficult to pinpoint. By comparing sexes we
may be able to identify sex-specific risk factors in common cancers. (Data from Cancer Research UK.)

2967_Ch09.indd 291 08/07/2015 14:42

 292 CHAPTER 9 Cancer as a Complex Disease

Percentage change, 2000–2002 versus 2009–2011

Decreasing Increasing
All cancers 3%
Prostate 16%
Lung –14%
Bowel 5%
Bladder –18%
Non-Hodgkin’s lymphoma 14%
Malignant melanoma 57%
Kidney 27%
Esophagus 3%
Leukemia 2%
Brain tumors –1%
Stomach –32%
Oral 32%
Pancreas 4%
Liver 45%
Myeloma 13%
Testis 6%
Mesothelioma 3%
Larynx –13%
Hodgkin’s lymphoma 11%
Thyroid 69%

Figure 9.7 Percentage change in cancer frequency in men 2000–2002 versus 2009–2011. The figure shows
how frequencies of some common cancers have changed in the new millennium. Bladder cancer, stomach
cancer, lung cancer, and cancer of the larynx have fallen, while there have been major increases in liver, thyroid,
skin, prostate, oral and kidney cancers, as well as both Hodgkin’s lymphoma and non-Hodgkin’s lymphoma.
Some of these clearly reflect lifestyle changes, whereas others are more difficult to pinpoint. (Data from Cancer
Research UK.)

are based on the UK data, similar figures are seen throughout the developed world. Figures
are different in less developed countries and can vary quite widely, as we can see looking at
the WHO site (http://apps.who.int/gho/data/node.main.A864). The figures for the UK
that are shown illustrate important shared changes in male and female cancer, and also
some important differences between males and females. In particular, the increased fre-
quency of lung cancer in women (up 13%) compared with the reduction in men (down
14%). It is interesting to speculate on why this may have occurred, for example is this
due to an increase in smoking in young women? The data presented also indicate the low
frequency of non-solid tumors, which do not count in the top 10. Changes in the inci-
dence of various cancers in the new millennium reflect the impact of the environment and
societal changes on disease, and not just on cancers. They are unlikely to reflect genetic
changes in our population.

Conclusions
There are several different forms of cancer – some cancers are familial Mendelian cancers,
others are sporadic cancers. Genetics plays a role in both types. However, unlike other

2967_Ch09.indd 292 08/07/2015 14:42

 Conclusions 293

diseases, there are two genomes to consider in cancer—the host genome and the tumor
genome. Looking at both early and more recent studies of cancer indicates that host genes
are important in all forms of cancer. Genetic markers in the form of tagged SNPs and
copy number repeat sequences have been used in GWAS to detect possible cancer genes
and their location across and throughout the genome. Some, though not all, of the genetic
risk factors previously detected by candidate gene analysis have also been confirmed by
GWAS. This research is important because not only does it enable us to identify poten-
tial cancer-causing genes and networks, it may also help us to use genetics to understand
the response to therapy in cancer and to develop better (more personal), novel treatment
options.
The last 10 years has seen considerable progress in our understanding of the genetic basis
of susceptibility and resistance to cancer. The application of GWAS in cancer has resulted
in the identification of some well-replicated risk factors for a number of cancers; however,
slightly disappointingly, the effect sizes remain small. Some may question the use of this
new knowledge. However, there is hope. These findings help to inform the debate on
the genesis of cancer and it is possible that, in the future, developments in the areas of
genome sequencing and epigenomics may enable additional risk factors to be identified
that will help us to develop more advanced screening programs. The future lies in inte-
grating knowledge of both the patient’s biology and the tumor biology at the levels of
DNA, RNA, protein, and other molecules, so that the likely response to a wide range of
treatments can be modeled mathematically. Ultimately, this information may be used to
inform clinical management of the cancer. In addition, the information gathered will help
to make systems biology a reality.
Further advances will require larger more extensive studies and the application of alter-
native technologies, such as exome sequencing and whole-genome sequencing, using a
variety of next-generation technologies. In 2013, studies in breast cancer and prostate
cancer, reported by the International Cancer Group (iCOGS), published the largest cancer
genetic association study to date (approximately 200,000 SNP probes in approximately
200,000 case/control samples for breast, prostate, and ovarian cancer). If these numbers
seem large there are plans underway to repeat the iCOGS with the larger OncoArray,
which has approximately 600,000 SNP/copy number variation probes using approxi-
mately 400,000 case control samples. Lung and colorectal cancer may also be included in
the next study. The iCOGS study represents an example of how the design in association
studies has developed over the years and how there will be increasing reliance on large
international collaborative networks.
Probably the most promising use of genetics in cancer so far has been in its use to inform
the development of personalized drug treatments for the disease. This has been generally
successful and is being used increasingly, especially in the treatment of advanced cancers
that may be refractory to other treatments. It is likely that personalized approaches will
become more common at all stages of cancer in the future.

2967_Ch09.indd 293 08/07/2015 14:42

 294 CHAPTER 9 Cancer as a Complex Disease

Further Reading
Books Hosking FJ, Dobbins SE & Houlston RS
Spector T (2012) Identically Different: Why (2011) Genome-wide association studies for
You Can Change Your Genes. Weidenfeld & detecting cancer susceptibility. Br Med Bull
Nicholson. 97:27–46.
Maitland-van der Zee A-H & Daly A (2012) Parkin DM, Boyd L & Walker LC (2011) The
Pharmacogenetics and Individualized Therapy. fraction of cancer attributable to lifestyle and
Wiley. environmental factors in the UK in 2010. Br J
Cancer 105 (Suppl 2):S77–S81.
Strachan T, Goodship J & Chinnery P (2014)
Genetics and Genomics in Medicine. Garland Rothman N, Garcia-Closas M, Chatterjee
Science. N et  al. (2010) A multi-stage genome-wide
association study of bladder cancer identi-
Weinberg RA (2007) The Biology of Cancer. fies multiple susceptibility loci. Nat Genet
Garland Science. 42:978–984.
Varghese JS & Easton DF (2010) Genome-
Articles wide association studies in common cancer-
Byrne HM (2010) Dissecting cancer through what have we learnt? Curr Opin Genet Dev
mathematics: from the cell to the animal 20:201–209.
model. Nat Rev Cancer 10:221–230. Wheeler HE, Maitland ML, Dolan ME et  al.
Chung CC & Chanock SJ (2011) Current sta- (2012) Cancer pharmacogenomics: strategies
tus of genome-wide association studies in can- and challenges. Nat Rev Gen 14:23–34.
cer. Hum Genet 130:59–78.
Dowsett M & Dunbier AK (2008) Emerging
biomarkers and new understanding of tra- Online sources
ditional markers in personalized therapy for http://apps.who.int/gho/data/node.main.
breast cancer. Clin Cancer Res 14:8019–8026. A864
Figueroa JD, Ye Y, Siddiq A et  al. (2014) A good source of data on cancer and other
Genome-wide association study identifies mul- diseases worldwide from the World Health
tiple loci associated with bladder cancer risk. Organisation (WHO).
Hum Mol Genet 23:1387–1398. http://www.cancerresearchuk.org/home/
Hanahan D & Weinberg RA (2000) The hall- This is an excellent source of data for cancer in
marks of cancer. Cell 100:57–70. the UK.
Heyn H & Esteller M (2012) DNA methyla- http://www.cogseu.org
tion profiling in the clinic: applications and This is the study database quoted in the text
challenges. Nat Rev Genet 13:679–692. section above regarding a Collaborative
Hindorff LA, Gillanders EM & Manolio TA Oncology Gene-environment Study consor-
(2011) Genetic architecture of cancer and other tium that has focused specifically on identify-
complex diseases: lessons learned and future ing genetic risk factors for breast, ovarian, and
directions. Carcinogenesis 32:945–954. prostate cancers.

2967_Ch09.indd 294 08/07/2015 14:42

 CHAPTER
10
Genetic Studies on
Susceptibility to Diabetes

Diabetes has a clear and demonstrable genetic component. There are two major forms
of diabetes with similar signs and symptoms: type 1 diabetes (T1D) and type 2 diabetes
(T2D). There is a considerable degree of clinical overlap between these two diseases. There
is also some clinical overlap with other diseases. In this chapter, we will look at these two
major forms of diabetes specifically focusing on their genetics. We will also briefly con-
sider the less common form of diabetes referred to as maturity onset diabetes of the young
(MODY). By comparing and contrasting the different forms of diabetes, we will be able to
demonstrate how genotype can be used to illustrate phenotype and to help us understand
disease pathogenesis.

10.1 Diabetes Mellitus

Diabetes is amongst the biggest killers in the developed world. To put this into c­ ontext,
accor­ding to the World Health Organization (WHO) (http://www.who.int/diabetes/
facts/en/), a total of 57 million people died in 2008. Of these deaths, 63% were due
to non-communicable diseases (NCDs). The major causes of these NCDs were cardio-
vascular disease (17 million), cancer (7.6 million), respiratory disease (4.2 million), and
diabetes (1.3 million) (Figure 10.1). The WHO figures show that prevalence is highest
in higher-income groups (10% versus 8% in the lower-income groups). The healthcare
burden in terms of resources for diabetes is considered to be two to three times higher.

2967_Ch10.indd 295 07/07/2015 10:55

 296 CHAPTER 10 Genetic Studies on Susceptibility to Diabetes

18

16

14

12

10

0
Cardiovascular Cancer Respiratory Diabetes

Figure 10.1: WHO data for deaths due to NCDs in 2008. The figure shows the impact of three major disease
groups on mortality in 2008, reported in 2014. Though the diabetes column appears small compared with
cardiovascular disease and cancer, these are very broad disease categories compared to diabetes, which is more
specific. (Data from WHO.)

In terms of morbidity, diabetes is a major cause of non-traumatic lower limb amputations,

which are 10 times more common in diabetes. Loss of sight and visual impairment are also
more common in diabetics, especially in developing countries.
Diabetes mellitus is a medical term describing a group of metabolic diseases that can
be characterized by higher than normal blood glucose levels due to either lack of
­insulin or failure to respond normally to insulin. There are two major forms of dia-
betes referred to as T1D and T2D; in addition, there is a group of mostly mono-
genic forms of diabetes referred to as MODY. T1D is much less common than T2D,
affecting approximately 0.5% of the population compared with approximately 6% for
T2D—a figure which is now being quoted in many developed countries. Overall T2D
accounts for nearly 85% of diabetes cases, with T1D making up 10% and around 5%
having MODY (Figure 10.2). Most cases of MODY tend to be Mendelian disorders,

T1D
T2D
MODY

Figure 10.2: Worldwide incidence of T1D and T2D. Based on WHO data, there are approximately 347 million
cases of diabetes worldwide with the majority being T2D (approximately 85%), as indicated in this pie chart, and
the remainder being mostly T1D (10%) and the minority having MODY (5%).

2967_Ch10.indd 296 07/07/2015 10:55

 Early Genetic Studies in T1D 297

Islets of Langerhans

Pre-pro-insulin
(signal peptide cleavage)

Pro-insulin
(further protease cleavage of 35 amino acids)

Mature insulin

Figure 10.3: Three phases of insulin production. Insulin produced in the islet cells goes through three phases
before it is functional.

while both T1D and T2D are genetically complex. To date, as with a number of other
genetically complex diseases, including most autoimmune diseases and some cancers,
only a proportion of the overall genetic risk for T1D and T2D has been accounted for.
The amount of progress that has been made varies between the different types of dia-
betes, for example excellent progress has been made with respect to T1D, but progress
has been slower for T2D.

10.2 Genetics of T1D

T1D occurs as a result of the immune destruction of the pancreatic β cells where insu-
lin is produced (Figure 10.3). This autoimmune reaction begins early in life. Evidence
for a genetic basis for T1D can be seen in studies of siblings of affected cases where
the sibling relative risk (λ) may be as high as 15. This indicates a 15-fold increase in
the risk of T1D for siblings of patients with this disease and suggests a considerable
genetic component in T1D. Though this figure is lower than that for Crohn’s disease,
it is higher than for many other genetically complex diseases and on the same level as
most other autoimmune disorders (Table 10.1). Studies of twins also show higher con-
cordance rates in monozygotic twins than for dizygotic twins with values of 40 or more
for monozygotic twins and less than 10 for dizygotic twins. However, caution needs to
be applied in twin studies. Long periods of follow-up are important as the age of onset
may vary between twins. If a study is performed for a short period only, the diagnosis
of diabetes in the paired twin may fall outside the study period and this may be missed,
giving false results.
Among complex diseases in general, the genetics of T1D is reasonably well understood.
This is due to both the strength of the genetic component and the nature of the dis-
ease. T1D compares well with other autoimmune diseases and those where the immune
response is known to play a major role in causality.

10.3 Early Genetic Studies in T1D

Early studies indicated a strong role for genetics in T1D. Studies in the 1970s first identi-
fied associations with a number of human leukocyte antigen (HLA) antigens; however,

2967_Ch10.indd 297 07/07/2015 10:55

 298 CHAPTER 10 Genetic Studies on Susceptibility to Diabetes

Table 10.1 Sibling relative risk (λ) values for different diseases.

Disease Estimated λ
T1D 15
T2D 3
Celiac disease 50
Ulcerative colitis 7–17
Crohn’s disease 13–36
Multiple sclerosis 20–50
Primary biliary cirrhosis 10.5 (rising to 58 in daughters of affected mothers)

the first study on HLA found no association. This later proved to be incorrect, and asso-
ciations with B7, B8, B15, B18, Cw3, DR2, DR3, and DR4 were soon identified. Most
of these studies were classical case control association studies, though there were excep-
tions (Figure 10.4). In parallel with this work on the major histocompatibility complex
(MHC), others were busy looking outside the MHC looking for genetic linkage using
multi-case families.

HLA class II genotype is the strongest genetic risk factor for T1D
Early studies reported associations with HLA class I antigens, but in the 1980s the focus
shifted to the HLA class II antigens and DR in particular. Studies identified associations
with DR3 and DR4 as major risk factors, and DR2 (later DR15) as a protective factor.
The development of better technology for HLA typing, including the introduction of
both restriction fragment length polymorphism (RFLP) analysis and polymerase chain
reaction (PCR) analysis (Chapter 6) led to the confirmation of the associations with DR2,

60

50 0
1
40 2
Percentage

30

20

10

0
Sibling pairs T1D sibling pairs

Figure 10.4: Sibling pair analysis in T1D looking at HLA haplotype sharing. The figure shows haplotype
sharing in two groups: a group of non-diabetic siblings and a group of diabetic sibling pairs. This study illustrates
the strong impact of HLA in T1D. This type of study is unusual, but it can be very useful as it does not identify
specific alleles but concentrates on haplotype sharing. (Adapted from Parham P [2009] The Immune System, 3rd
ed. Garland Science.)

2967_Ch10.indd 298 07/07/2015 10:55

 Early Genetic Studies in T1D 299

DR3, and DR4 at a molecular level (now labeled DRB1*03, DRB1*04, and DRB1*15). In
addition, the application of PCR enabled the associations with specific DQB1 and DQA1
alleles carried on the DRB1*03, DRB1*04, and DRB1*15 haplotypes to be identified for
the first time. Genotyping studies rapidly confirmed these associations in different cohorts
of T1D patients. In 1987, Todd et al. published a groundbreaking study suggesting that
HLA-encoded susceptibility to T1D is due to a specific amino acid at position 57 of the
DQβ-chain. The amino acid aspartate is found on DQB1 alleles that are associated with a
reduced risk of T1D. In contrast, DQB1 alleles that are associated with an increased risk
encode alanine, valine, or serine at position 57 (see Chapter 6). This study was the first to
consider HLA associations from a functional perspective. The study was published in the
same year that the crystal structure of HLA-A2 was published in Nature and the authors of
the T1D paper were able to show that this subtle structural difference determined whether
a salt bridge would form over the antigen-binding groove of the expressed DQ molecule.
Formation of the salt bridge involved the interaction between position 57 on the DQβ-
chain and arginine at position 79 on the DQα-chain. This interaction may be critical in
determining which antigens are preferentially presented to the T cell receptor (TCR) in
the formation of the immune synapse and the orientation of the bound peptide antigen
in that process.
This is certainly not the whole story for HLA and T1D. The associations established
in the 1980s and 1990s have mostly been confirmed. In 1994, an early genome-wide
study on affected sibling pairs using microsatellite markers confirmed a strong signal from
the MHC. The 2007 Wellcome Trust Case Control Consortium (WTCCC1) study also
found a very strong signal for the MHC [odds ratio (OR) = 5.49, P = 2.42 × 10−134]. The
HLA class II haplotypes that are positively associated with T1D include DRB1*03:01–
DQA1*05:01–DQB1*02:01 (the HLA 8.1 ancestral haplotype) and DRB1*04:01–
DQA1*03:01–DQB1*03:02. Approximately 95% of patients of European ancestry with
T1D are positive for one or both of the two haplotypes. In contrast, only 40% of non-­
diabetics of European ancestry have one or both of these haplotypes. The HLA class II hap-
lotype DRB1*15:01–DQA1*01:02–DQB1*06:02 is significantly less common in affected
individuals. The estimated contribution of HLA to genetic risk for T1D is between 30%
and 50%. Interestingly, these three haplotypes, which are all common, are also associated
with a wide variety of different diseases, especially immune-mediated and autoimmune
diseases. The risk alleles do not always have the same effect in different diseases—the
DQB1*06:02 haplotype, which is protective in T1D, is associated with an increased risk
of both multiple sclerosis and severe narcolepsy.

Not all of the risk for T1D above may be associated with the DQB
allele or HLA class II
These extended haplotypes carry a number of other potential risk genes. For example,
the HLA 8.1 ancestral haplotype carries MICA*008, which has been associated with an
increased risk of the autoimmune liver disease primary sclerosing cholangitis and is involved
in killer cell activation. The 8.1 haplotype also carries a null allele for C4A, thought to
be the main factor in MHC-encoded susceptibility to sporadic cases of systemic lupus
erythematosus. Therefore, it is possible that some of these haplotypes carry more than one
risk allele. Such a multihit hypothesis would explain the strong effect of the MHC in T1D
and other autoimmune diseases.
Finally, more recent evidence suggests an additional HLA class I-encoded contribution
to susceptibility to T1D that is independent of HLA class II. This was reported in a
study involving more than 5000 cases. Association signals from HLA-A and HLA-B

2967_Ch10.indd 299 07/07/2015 10:55

 300 CHAPTER 10 Genetic Studies on Susceptibility to Diabetes

were seen that were independent of HLA class II and not due to linkage disequilibrium.
Based on this study, HLA-B*39 appears to be the strongest independent class I risk
allele, but additional HLA-A and HLA-B alleles also appear to contribute and may
be important as determinants of age of onset of disease. Interestingly, an association
with HLA-B*39 was not listed in the compendium of early studies of HLA under the
subtitle diabetes.

Other genetic risk factors for T1D include the genotype for the
insulin gene
Evidence that the insulin gene was a predictor for susceptibility to T1D was reported in
the early 1980s in a case control association study that used RFLP analysis. This asso-
ciation was subsequently confirmed using the transmission disequilibrium test (TDT)
in families with at least one affected sibling pair. The 1994 genome-wide study that was
based on the use of variable number tandem repeats (VNTRs or microsatellites as they
are known) in affected sibling pairs also reported a signal in the insulin gene region,
though it was not as significant as that seen for HLA. The signal was given the label
IDDM2. The insulin gene codes for an insulin precursor known as ­pre-pro-insulin.
This precursor is converted in the endoplasmic reticulum to pro-insulin and pro-
insulin is converted to insulin by the enzymatic removal of a segment that connects
the amino end of the α-chain to the carboxyl end of the β-chain. This produces the
bipeptide chain of mature insulin. The segment that is removed is referred to as the
connecting (C) peptide. Despite ongoing studies, the underlying mechanism for
the association with the insulin gene remains unclear. However, it appears to involve
variation in the 5′-non-coding sequence that results in decreased expression of the
insulin precursor pre-pro-insulin. It is likely that this occurs in the thymus, but not
in the pancreas, and it has been proposed that the decreased expression in the thymus
leads to decreased immune tolerance to insulin resulting in an inappropriate immune
response in the pancreas.

Candidate gene studies have identified a number of other non-MHC

associations with T1D
Candidate gene studies have resulted in the detection of a number of associations for
T1D with non-MHC immune-related genes. As with most complex diseases, not all of
these associations have been confirmed. The best replicated associations include those with
genes encoding protein tyrosine phosphatase, non-receptor type 22 (PTPN22), cytotoxic
T lymphocyte associated protein 4 (CTLA4 ), and interleukin (IL)-2 receptor subunit α
(IL2RA) (Table 10.2). These associations are all plausible because of the important roles of
these gene products in T cell immunity. However, we need to be cautious about assigning
causality on the basis of plausibility. Plausible should be read as good potential candidate
and not as confirmed disease-causing mutation or allele.

PTPN22
PTPN22 encodes a lymphoid-specific intracellular phosphatase that is a negative
regulator of TCR signaling. This occurs by direct de-phosphorylation of various
cell signaling proteins including the Src family kinases LCK and FYN. Associations
with this gene have been described for a number of different autoimmune diseases
including rheumatoid arthritis.

2967_Ch10.indd 300 07/07/2015 10:55

 Early Genetic Studies in T1D 301

Table 10.2 Pre-genome era regions identified as potential areas for candidate genes in T1D.

Gene Location Potential candidate (and estimated impact if available)

IDDM1 6p21.3 MHC (λ = 3.1: 30–40% of overall genetic risk)
IDDM2 11p15.5 INS (λ = 1.3: 10% of overall genetic risk)
IDDM3 15q26 β-2-Microgobulin
IDDM4 11q13 LRP
IDDM5/8/15 6q21–q27 No candidate
IDDM6 18q21 Kidd blood group
IDDM7/12/13 2q13–q33 CTLA4
IDDM10 10p11–q11 IL2RA
IDDM11 14q24.3 No candidate
IDDM17 10q25 No candidate identified; in large Bedouin family only

The table provides a summary of potential areas for investigation or known to be associated with T1D prior to
the publication of the Human Genome Mapping Project. Among the candidates, IDDM1, IDDM2, and IDDM7
have all been confirmed.

CTLA4
CTLA4 encodes a co-stimulatory molecule expressed by activated T cells. It binds to
CD80 and CD86 on antigen-presenting cells (APCs) (previously known as B7) and
transmits an inhibitory signal, thereby deactivating the T cells and turning off the
immune response. This process is competitive. In early T cell activation, T cells express
CD28 which binds with CD80 and CD86 on the APCs, sending a positive signal
that promotes the T cell immune response. As the immune response progresses, more
CTLA4 protein is expressed on T cells and the immune response is downregulated
(Figure 10.5). There are many polymorphisms in the CTLA4 gene. Prior to the use of
GWAS, Ueda et al. investigated 108 single nucleotide polymorphisms (SNPs) in and
around the CTLA4 gene in a large cohort of T1D cases and identified a specific SNP
that they labeled CT60 as the strongest associated T1D CTLA4 SNP. Recent studies
indicate that CTLA4 is associated with T1D, but have not pinpointed the specific risk
allele. Associations with this gene have been described for several autoimmune dis-
eases, most notably with the autoimmune thyroid disease Graves disease. Interestingly,
Ueda et al. included Graves disease in their study and also found the strongest asso-
ciation with the so-called CT60 SNP. Finding associations with the same gene in
several diseases is not surprising, especially when the gene in question has a broad (or
non-specific) function, such as CTLA4 that encodes a non-specific immunoregulatory
protein.
IL2RA
IL2RA codes for a subunit of the IL-2 receptor. Like many receptors, the IL-2 receptor
is a dimer composed of an α and a β subunit. IL-2 is a major cytokine involved in T cell
activation. IL2RA is constitutively expressed at high levels in CD4+ T cells, which are
also positive for the FoxP3 protein. These cells are believed to be important in immune
tolerance to self-proteins. The association with T1D is protective and appears to involve a

2967_Ch10.indd 301 07/07/2015 10:55

 302 CHAPTER 10 Genetic Studies on Susceptibility to Diabetes

CD25+ CD4+ CD4+

T cell T cell

CTLA4 TCR TCR CD28

CD80/CD86 CD80/CD86
APC

Off switch On switch

Figure 10.5: Suppression of auto-reactive T cells by CTLA4 signaling. The figure illustrates the process
of CTLA4 signaling by the CD25+ CD4+ T cell. The process is the same as that for T cell activation; however,
in this situation the T cells send out a negative response on activation by the APC. Antigenic peptides are
represented here as stars. Antigenic peptides are presented to the TCR by MHC class II proteins (shown in
gray as an inverted triangle supporting a curved structure), which is mounted on the APC (also shown in gray).
There is co-recognition of this molecular complex by CD4 (shown as a lozenge with a short leg). The first part
of the process is the same for both T cells. However, a second interaction occurs which is different. The T cell
on the right expresses CD28, which interacts with the CD80/CD86 molecule (B7 in some texts; shown here as
dark curves) and this causes T cells to induce immune activation. The T cell on the left expresses CTLA4, which
interacts with CD80/CD86, sending a signal that downregulates the immune response. CD28 and CTLA4 are both
illustrated as gray tubes on the surface of the T cells.

gain-of-function polymorphism. However, we should be cautious and not over-interpret

such data. Biological systems are complex and often encompass some degree of redun-
dancy. Increases and reductions in receptor expression can both influence the level of the
immune response leading to greater or lesser levels of organ damage. Inappropriately high
levels of receptor expression may lead to over-activation, whereas inappropriately low lev-
els may prevent activation. However, high receptor expression does not necessarily lead to
higher levels of activation because there may be low levels of the ligand protein, which in
this case is IL-2. This is considered in more detail in Figure 10.6.
IFIH1
The interferon (IFN)-induced helicase C domain-containing protein 1 gene (IFIH1)
codes for a protein that is believed to be able to interact with viral RNA and to induce the
expression of IFNs. IFNs are key agents in both innate and adaptive immunity, and play
an important role in immune regulation. Numerous genetic polymorphisms with IFN
genes have been associated with genetically complex disease and it is not surprising to find
genes from the interferon pathways associated with T1D.

2967_Ch10.indd 302 07/07/2015 10:55

 GWAS studies in T1D 303

(a) (b)
120
70
100 Genotype A Genotype B
60
80 50
Biological activity

Biological activity
60 40

30
40
20
20
10

0 0
Time Time

Figures 10.6: Illustration of the reason why both elements of a system need to be considered when
considering the impact of a polymorphism on a trait. The two graphs illustrate the interaction between
genetic polymorphisms in a biological pathway. If we consider a model where we have only two components,
e.g. IL-2 and its receptor, biological activity can increase to infinity as long as there is adequate production of
both the receptor and its ligand. If genotype A represents those with high ligand expression, activity will increase
for as long as there is sufficient receptor expression. Similarly, if genotype A is associated with high receptor
expression, activity will increase provided there is a continuous supply of ligand. Genotype B, however, indicates
a more reasonable proposition. In the second example activity begins to reach an upper limit because one or
other of the two components reaches saturation. In real biological systems high expression of a ligand or receptor
like IL-2 does not necessarily result in activity. Thus, we need to be careful about interpreting data that indicates
increased expression of a gene or protein associated with a specific polymorphism.

10.4 GWAS studies in T1D

There had already been some GWAS in T1D prior to the conclusion of the Human Genome
Project (HGP), but these were few and far between. One using VNTRs performed in 1994
on T1D has been discussed already (Table 10.2). Most of the early studies were limited to
a few hundred markers and though they represented a step forward in their time, they now
appear primitive compared with today’s technologies. Most importantly, with so few mark-
ers tested, the possibility of missing significant associations was very high in the pre-genome
studies. It is therefore reasonable to say that the application of GWAS as a means of detect-
ing risk alleles in complex disease in the post-genome era was a major step forward with
great potential. The transition from a hypothesis-constrained approach, which applied
to most case control studies, to a hypothesis-generating approach and the technology
that enabled large numbers of cases and controls to be genotyped for upwards of 500,000
tagged SNPs was quite amazing. Even though new technologies involving higher through-
put and higher resolution are currently being applied and developed, which will make cur-
rent GWAS seem primitive in the future, the impact of the first GWAS is still fresh.

The 2007 WTCCC1 study was one of the first GWAS in T1D
The WTCCC1 2007 GWAS included 2000 T1D cases. Several regions were identified as
major risk determinants including all of the genes or regions listed above: MHC, CTLA4,
PTPN22, IL2RA/CD25, and IFIH1/MDAS. It is important to note that of these previ-
ously identified regions and genes only two had P values above the high statistical thresh-
old P < × 10−7 (MHC and PTPN22) and two were in the lower statistical threshold region

2967_Ch10.indd 303 07/07/2015 10:55

 304 CHAPTER 10 Genetic Studies on Susceptibility to Diabetes

Table 10.3 Six major risk genes and regions for T1D in the post-genome era.

Alleles or Gene(s) Location Function P value and/or OR

DQB1*02:01 (DQ2) 6p21.3 Antigen Presentation WTCCC1 study
DQB1*03:02 (DQ8) P < 2.42 × 10−134
DQB1*01:02 (DQ1) OR = 5.49
8.1 haplotype 6p21.3 Antigen Presentation
7.2 haplotype
B*39 (independent)
INS 11p15.5 Insulin
PTPN22 1q13 Protein tyrosine phosphatase non- WTCCC1 study
receptor-type 22 P < 1.82 × 10−26
OR = 5.43
CTLA4 2q33 Cytotoxic T lymphocyte protein 4 WTCCC1 study
Downregulates immune response P < 1.8 × 10−5
IL2RA 10p15 IL-2 receptor subunit α. Cytokine WTCCC1 study
receptor subunit P < 4.3 × 10−5
IFIH1 2q24 IFN-induced helicase C domain- WTCCC1 study
containing protein 1 P < 7.6 × 10−3

Four of the listed genes or regions were identified long before the publication of the Human Genome Mapping
Project and use of GWAS: MHC, INS, CTLA4, and IL2RA. Apart from INS, all of these genes or regions (MHC) have
associations with other diseases (i.e. they are not specific to T1D).

around P < × 10−5 (IL2RA and CTLA4) (Table 10.3). In addition to the associations above
a number of new regions were identified as showing strong risk (P < 10−7); these include
12q13, 12q24, and 16p13 together with other regions with similar levels of significance,
including 4q27, 12p13, 18p11, and 10p15 (CD25). Of these, the associations with 12q13,
12q24, 16p13, and 18p11 have all been confirmed in other studies. A number of potential
functional signals can be identified in this group, e.g. 12q13 is close to the ERBB3 gene
that encodes the receptor tyrosine kinase erbB-3 precursor, and 12q24 is close to SH2B3/
LNK (SH2-B adaptor protein 3), TRAFD1 (TRAF-type zinc-phosphatase domain con-
taining 1), and PTPN11 (protein tyrosine phosphatase, non-receptor type 11).
PTPN11 is a particularly interesting candidate for T1D as it is a member of the same fam-
ily of regulatory phosphatases as PTPN22. As discussed earlier, PTPN22 is associated not
only with T1D but also with Crohn’s disease and rheumatoid arthritis, suggesting over-
lapping pathology. The 12q24 association reported above is associated with a combined
signal (measured as probability value P) for T1D, Crohn’s disease, and rheumatoid arthri-
tis of 9.3 × 10−10. Overlapping genetic associations are reported for a number of different
complex diseases (Figure 10.7).
The association with the 10q15 region that contains the CD25 gene is found in Graves
disease, rheumatoid arthritis, and T1D. Graves disease was not included in the WTCCC1
study, but rheumatoid arthritis and T1D were both included. The study reported separate
independent associations for both diseases. CD25 encodes a high-affinity receptor for
IL-2, and this association may highlight the importance of the IL-2 pathway in T1D and
other autoimmune diseases.

2967_Ch10.indd 304 07/07/2015 10:55

 GWAS studies in T1D 305

Shared
susceptibility
genes

T1D Rheumatoid arthritis Crohn’s disease

• MHC • MHC • Weak MHC
• CTLA4 • CTLA4 • PTPN22
• PTPN22 • PTPN22

Figure 10.7: Genetic overlap in complex disease—finding the same associations in different diseases.
The figure illustrates some of the shared susceptibility genes for three very common complex diseases: T1D,
rheumatoid arthritis, and Crohn’s disease. The shared susceptibility regions or loci are the MHC (a region not a
locus), the CTLA4 genes, and the PTPN22 gene. The list is not exclusive as many other genes could be included; it
is only intended to illustrate a point.

Following the introduction of GWAS in 2007, research has resulted in

the identification of at least 40 further potential T1D alleles
Subsequent GWAS analysis has resulted in the detection of more novel genetic associa-
tions. These tend to be weaker than those discussed so far. The current count suggests
there may be more than 40 risk alleles in all (a selection of the strongest of these is shown
in Table 10.4). However, some of these are synonymous polymorphisms in specific
genes and others are in non-coding sequences. Synonymous polymorphism in the exome
sequence does not lead to sequence variation in the encoded polypeptide and therefore is
not functionally significant. Synonymous polymorphism in the non-­coding sequence is
also unlikely to be functionally significant. However, the associations with SNP markers
should not be written off; in both cases the associations may act as surrogates for func-
tional associations elsewhere on the haplotype.

Table 10.4 Other risk alleles for T1D.

Potential candidate Location

CD25 (high-affinity receptor for IL-2) 10p15
ERBB3 (receptor tyrosine protein kinase erbB-3 12q13
precursor)
SH2B3/LNK (SH2-B adaptor protein 3) 12q24
TRAFD1 (TRAF-type zinc finger domain PTPN11 is the most likely candidate with
protein 3) associations with both rheumatoid arthritis
PTPN11 (protein tyrosine phosphate, non- and Crohn’s disease in addition to T1D
receptor type 11)
KIAA0350 (dexamethasone-induced transcript) 16p13
Function of these genes is unknown
Possibly PTPN2, a member of the same family 18p11
as PTPN11 and PTPN22 protein tyrosine
phosphatase, non-receptors

2967_Ch10.indd 305 07/07/2015 10:55

 306 CHAPTER 10 Genetic Studies on Susceptibility to Diabetes

MHC
30– 40%

Unidentified
INS
genes T1D
10%
30– 40%

Other
identified
genes
20%

Figure 10.8: Genetic accounts book for T1D. The figure shows the current estimate of the total genetic impact
of genes identified as risk markers for T1D. The MHC has the strongest impact, with the insulin (INS) gene second.
The other group includes numerous genes involved in immune regulation, but there is still a substantial portion
of the genetic risk to be identified if these figures are accurate.

The addition of replicated novel associations discovered in GWAS to those already well-
established associations from a mix of family studies and candidate gene analysis has further
increased the extent of the genetic account for T1D to approximately 70% (Figure 10.8).
This is much higher than for the majority of complex diseases and is a very different situ-
ation to that seen with T2D (below).

10.5 Early Genetics of T2D

The vast majority of diabetes cases are of T2D, once called non-insulin-dependent diabetes
mellitus (NDDM), and the WHO estimates that there are currently more than 340 mil-
lion people suffering from this disease worldwide, with numbers expected to rise further
over the next 20 years. The rise in T2D reflects differences in the environment and in our
lifestyles. In addition, there is thought to be a substantial genetic contribution to T2D.
As the name T2D implies, this is not the same as T1D. It is a disease with a later average age
of onset than T1D, the genetics of T2D is more complex than that for T1D, and this is a
disorder that is less easy to investigate than T1D. Despite all of this, there have been intensive
efforts to identify the genetic risk factors in T2D over the past 20–30 years. Initially, linkage
analysis and candidate gene case control studies were used. More recently, GWAS has been
employed. This has resulted in the identification of more than 60 genetic risk alleles for T2D.
Prior to the introduction of GWAS, it was simply not possible to identify risk alleles for
many complex diseases, such as some sporadic cancers. However, in T2D, some early link-
age and candidate gene case control association studies successfully identified risk alleles.
In particular, three specific genes were identified as carrying potential risk alleles that

2967_Ch10.indd 306 07/07/2015 10:55

 GWAS Studies in Type T2D 307

were subsequently confirmed by GWAS. These included significant associations with non-­
synonymous SNPs in the PPARG, KCNJ11, and TCF7L2 genes. A number of other asso-
ciations originally described in the same era were not confirmed by GWAS. Confirmation
in later studies is a mark of quality and reflects study design, and especially sample size.
Unlike T1D, there are no associations with HLA or the MHC in T2D. This may indicate
that the disease pathogenesis is very different from that for T1D and does not involve the
same immunological (or other biological) processes.

There have been different interpretations of the associations with

PPARG, KCNJ11, and TCF7L2
Studies have shown that carriage of the relatively rare alanine-encoding allele PPARG vari-
ant is associated with a reduced risk of disease. This association has been widely con-
firmed in a number of studies. The PPARG gene is a plausible candidate gene for T2D as
it encodes the peroxisome proliferator-activated receptor-γ—a nuclear receptor that is a
regulator of adipocyte differentiation and glucose homeostasis. The ligands for this recep-
tor include prostaglandin metabolites and glitazone drugs. The PPARG protein has been
used as a target for treatment by the thiazolidinedione drugs, such as rosiglitazone.
In contrast to PPARG, the situation with KCNJ11 seems to be less clear. The KCNJ11 gene
encodes the inwardly-rectifying KIR6.2 component of the pancreatic β cell ATP-sensitive
potassium channel (KATP), E23K. KIR6.2 is found in islet cells of the pancreas. These
KATP channels couple cell metabolism to membrane excitability. In the pancreatic β cells
the channel is an octameric complex of KCNJ11 subunits and SUR1 (encoded by the adja-
cent ABCC8 gene) subunits. Four subunits form the channel pore. Each subunit is associ-
ated with a SUR1 subunit that regulates gating. Initial reports suggested an association with
a non-synonymous variant of KCNJ11. Looking at the function of this gene’s product, it
appears to be a very plausible candidate for T2D. However, it has been reported that it is not
possible to distinguish between the risk associated with this SNP and another non-synon-
ymous SNP in the adjacent ABCC8 gene that codes for the sulfonylurea receptor (SUR1).
SUR1 has been used as a target for another group of widely used anti-diabetes drugs.
The third genetic risk factor for T2D identified prior to GWAS was TCF7L2. This gene
was identified from a signal on chromosome 10q in linkage analysis of families affected
by T2D. Fine mapping showed that the signal related to an intronic SNP in TCF7L2.
TCF7L2 codes for the high mobility group (HMG) box-containing transcription factor
TCF4, which modulates the Wnt signaling pathway and possibly the secretion of insulin
by the pancreas. In addition, this gene may have a role in glucose production in the liver.
TCF7L2 has a very strong association with T2D. The per allele OR of 1.4 remains the
strongest risk factor for T2D identified so far. However, the underlying mechanism for the
effect of this intronic SNP on TCF7L2 gene expression remains unclear. It is possible that
the location in an open chromatin site in pancreatic β cells is important and there is also
some evidence that the risk allele is associated with increased transcription.

10.6 GWAS Studies in Type T2D

The first GWAS in T2D were reported in 2007. These studies confirmed some previously
reported associations and detected a range of additional genetic risk factors. The previ-
ously reported risk factors PPARG, KCNJ11, and TCF7L2 were all confirmed, and a num-
ber of novel risk loci were identified (Table 10.5). These initial studies typically involved

2967_Ch10.indd 307 07/07/2015 10:55

 308 CHAPTER 10 Genetic Studies on Susceptibility to Diabetes

Table 10.5 Selected well-replicated risk loci for T2D.

Name Gene, location, OR, and P value

Peroxisome proliferator-activated receptor-γ PPARG (3p25)
WTCCC1 OR = 1.23, P = 5.4 × 10−3
Potassium channel, inwardly rectifying, KCNJ11 (11p15)
subfamily J, member 11 Close to ABCC8 gene
WTCCC1 OR = 1.15, P = 5.2 × 10−3
Transcription factor 7-like 2 TCL7L2 (10q25)
WTCCC1 OR = 1.36, P = 5.2 × 10−12
Solute carrier family 30 (zinc transporter), SLC30A8 (4 GWAS)
member A8 OR = 1.12–1.15
CDK5 regulatory subunit associated protein CDKAL1 (4 GWAS)
1-like 1 OR = 1.12–1.25
Hematopoietically expressed homeobox HHEX (3 GWAS)
OR = 1.13–1.18
Hepatocyte nuclear factor-1α HNF1A (2 GWAS)
HFN1 homeobox A OR = 1.07–1.14
HFN1 homeobox B HNF1B (3 GWAS)
OR = 1.1–1.17
Insulin-like growth factor 2 mRNA-binding IGF2BP2 (3 GWAS)
protein 2 OR = 1.14
Potassium voltage-gated channel subfamily, KCQN1 (3 GWAS)
member 1 OR = 1.08–1.23
Cyclin-dependent kinase inhibitor 2A CDKN2A/2B (3 GWAS)
OR = 1.2

a number of centers, including the WTCCC1 study, and the number of cases varied from
around 2000 to approximately 8000 cases. Most cases were of European origin. There was
good agreement between the various studies in terms of risk factors detected and the find-
ings were also replicated in additional cases in most studies.

Examples from the WTCCC1 study

SLC308A
The SLC30A8 (solute carrier family 30, member A8) gene encodes the zinc transporter
ZnT8—a protein with a role in the secretion of insulin by the pancreas. It contributes
to insulin maturation and/or storage in the pancreatic β cells. This association was only
weakly identified in the WTCCC1 study, but was also identified in a study of French T2D
cases. Several GWAS have now confirmed this association.
FTO
The FTO (fat-mass and obesity-associated) gene on chromosome 16q12 was found to be
very strongly associated with T2D in the WTCCC1 study (OR 1.34, P < 5.24 × 10−8),

2967_Ch10.indd 308 07/07/2015 10:55

 GWAS Studies in Type T2D 309

but has not been universally reported. Surprisingly, the FTO association is consider-
ably stronger than that for SLC30A8. However, a number of studies have replicated
the association with SLC30A8, which contrasts with the situation for the FTO gene.
These differences between studies are important. They may reflect genuine evidence of
false positives or they may reflect differences in study design, e.g. the use of different
commercial chips (Illumina or Affymetrix) with different tagged SNPs. One solution to
this would be to use specifically designed chips for studying specific diseases and disease
groups. However, while many companies used to be willing to produce specific chips at
a reasonably low price, this is no longer considered such a viable option and it may not
be possible in the future. Observed differences in associations reported can arise from
differences in the close proximity between tagged SNPs and actual disease alleles for
which they are markers.

CDKAL1
The third strong association signal in the WTCCC1 GWAS was with the CDKAL1
(CDK5 regulatory receptor subunit associated protein 1-like 1) gene. The product of this
gene shares homology with the protein domain-level CDK5 regulatory subunit associated
protein 1 (CDK5RAP1), which is known to inhibit the activation of CDK5. CDK5 is a
cyclin-dependent kinase that has been implicated in normal β cell function. The associa-
tion has been confirmed in several GWAS.

Other risk alleles for T2D from other studies

To date, the number of replicated genetic risk factors for T2D has increased to more
than 60. This has been achieved by larger studies and particularly by meta-analysis.
The list of risk loci includes CDKN2A, HHEX, HNF1A, HFN1B, IGF2BP2, and
KCNQ1, among many others. Despite this work, our understanding of the underlying
mechanism by which these genes increase or reduce the risk of T2D remains unclear.
Each of the identified risk alleles is associated with relatively small though significant
ORs. Such low ORs indicate that we have probably identified less than 10% of the
genetic risk for T2D so far. These figures also serve as a reminder that the environment
plays a major role in T2D (Figure 10.9). However, there are likely to be many more risk
alleles yet to be identified in T2D—a situation similar to that for a range of different
complex diseases.
The genes that have been identified in T2D are mostly involved in insulin secretion by the
pancreas rather than insulin action. This suggests that these are genuine candidate genes
for T2D. The absence of genes associated with autoimmunity in T2D helps to validate the
case cohorts used in these studies. In addition, the absence of immune response genes in
this list also helps to validate studies in T1D where such associations are found. This also
proves the point that such comparisons between diseases with quite different etiologies
can be useful.
Attempts have been made to score individual risk for the development of T2D based on
the findings of these studies. Overall, knowledge of genotype is likely to be of limited value
in diagnosis, adding little to the list of already well-established clinical criteria and tests.
This is not surprising when we consider the low level of disease risk that is associated
with each of these alleles. However, some of the genes that have been recently identified
could encode potentially useful drug targets. One way forward could be stratification on
the basis of genotype. By using selected genetic (and other) risk factors we could develop
personalizing treatment to suit the needs of the individual patients. There is currently
some, albeit limited, evidence that this may work in T2D.

2967_Ch10.indd 309 07/07/2015 10:55

 310 CHAPTER 10 Genetic Studies on Susceptibility to Diabetes

Genes

T2D Lifestyle

Diet Figure 10.9: Factors influencing risk of T2D. The figure

illustrates the main factors associated with an increased risk
of T2D.

10.7 The Future of Genetics in T2D

Unlike T1D, for which a very significant component of the genetic risk has now been
identified, studies indicate that the pattern is likely to be much more complicated and
challenging for T2D. As T2D represents a greater clinical case load than T1D, it is essen-
tial that these studies continue. Two developments are likely to have a large impact on
future studies in diabetes: the introduction of genome sequencing as a routine research
procedure and the expansion of the use of epigenetics.

Future prospects in T2D research involve genome sequencing

Rare genetic variants whose contribution cannot be assessed by GWAS are likely to be
detected in genome-sequencing studies. One of the first studies on detection of rarer vari-
ants that contribute to disease used exome sequencing on 1000 Danish cases and 1000
controls. Polymorphisms identified as significant risk markers were replicated in a series
of approximately 80,000 Europeans. Some novel associations were seen when comparing
the genotypes identified with the metabolic traits in T2D, but the overall conclusions were
slightly disappointing. The study found that coding region polymorphisms with frequen-
cies in the range 1–5%, which would not have been covered by the previous GWAS, did
not show large risk effects (OR values) for diabetes or related metabolic traits. Further
studies may need to focus on very rare sequence variants, where effect sizes could be larger,
but such studies will require very large numbers of cases.

Epigenetics may be important in diabetes

The missing 90% of disease risk is likely to involve a substantial contribution from non-
genetic factors, epigenetic factors which are currently more difficult to detect than genetic
factors, and epistasis involving gene–gene interactions. Epigenome-wide association
studies (EWAS) where DNA methylation patterns across the genome are compared
between cases and controls have recently been applied to T2D. Currently, such studies are
technically more difficult to perform than GWAS for several reasons. These include the fact

2967_Ch10.indd 310 07/07/2015 10:55

 Genetics of Monogenic Diabetes 311

that patterns of DNA methylation may vary considerably between tissues, so using DNA
from accessible tissues such as blood may not be representative of methylation patterns
in disease tissues, e.g. the pancreas. A second issue is that accurate assessment of methyla-
tion requires DNA sequencing following treatment with bisulfite, though it is possible to
narrow down the chromosomal regions differentially methylated prior to this sequencing.
Methylation chips that cover methylation sites across the genome are also now available.
Methylation patterns may also be influenced by a variety of environmental factors and so
can change over time. In addition, studying DNA methylation represents only one mea-
sure of epigenetic regulation, which also involves processes such as histone acetylation.
Despite these caveats, studies based on both methylation of candidate genes and EWAS
for T2D have now been reported. One EWAS used DNA from diabetic pancreatic islets
and detected differential methylation at promoter regions of 254 genes. These methylation
changes were not present in DNA from blood samples. Another recent EWAS reported
lower levels of methylation in the region of genes such as TCF7L2 in DNA from diabetic
blood samples, but did not study methylation patterns in other tissues. However, both
approaches may be valid. Further discussion of epigenomics is beyond the scope of this
chapter, but it seems likely that this area of research may provide novel insights into the
risk of T2D and other diseases in the near future (see chapter 12, section 12.5).

10.8 Genetics of Monogenic Diabetes

There are a number of monogenic forms of diabetes that have some features in common
with either T1D or T2D, but they represent distinct forms of the disease. In these cases
the familial component is stronger, with at least one affected parent, and the age of onset is
usually very young (typically ranging from infancy to early adulthood). The collective term
for many of these disorders is MODY. This term is not applied to all early-onset forms of
diabetes as there are some forms with more specific names, e.g. Donohue syndrome and
Rabson–Mendenhall syndrome. Based on the location of the disease-causing mutation,
there are at least 10 different forms of MODY currently listed in the OMIM (Online
Mendelian Inheritance in Man) database, accounting for up to 5% of all cases of diabetes
(Table 10.6). The phenotypes differ from those of classical T1D or T2D and will depend

Table 10.6 Chromosomal locations of genes causing MODY.

Gene Location
MODY1 20q13
MODY2 7p13
MODY3 12q24.3
MODY4 13q12.2
MODY6 2q31.3
MODY7 2p25.1
MODY8 9q34.2
MODY9 7q32.1
MODY10 11p15.5
MODY11 8p23.1

2967_Ch10.indd 311 07/07/2015 10:55

 312 CHAPTER 10 Genetic Studies on Susceptibility to Diabetes

on the specific gene defect. These are monogenic diseases. Two subtypes of monogenic
diabetes are of particular interest in relation to our understanding of genetic susceptibility
to T2D: neonatal diabetes, which usually has an onset during the first 6 months of life, and
young onset diabetes, where the disease is due to mutations in transcription factor genes
such as HNF1A, which codes for hepatocyte nuclear factor (HNF)-1α. This mutation has
also been found to be associated with an increased risk of T2D.
A significant proportion of neonatal diabetes cases are due to gain-of-function mutations
in either KCNJ11 or ABCC8, which encode separate subunits of the pancreatic β cell
potassium channel. These mutations typically increase the electrostatic current through
the channel, preventing depolarization in response to glucose metabolism and impaired
insulin secretion. Polymorphisms in these genes are also associated with increased sus-
ceptibility to T2D. The relatively recent finding that many cases of neonatal diabetes
were due to mutations in potassium channel genes was an important advance in terms of
treatment. These patients would have been previously treated with insulin on the grounds
that their disease was insulin-dependent, but they are now normally treated with a high
dose of an orally administered sulfonyl urea that targets potassium channels directly.
Interestingly, another potassium channel gene, KCNQ1, has been found to be associated
with an increased risk of T2D.
Mutations in HNF1A are a common cause of young-onset diabetes. HNF1A encodes the
HNF-1α, a transcriptional activator that regulates tissue-specific expression of a range
of genes, particularly in the liver and pancreatic islet cells. The mutations reported are
dominant mutations which result in deterioration of pancreatic β cell function including
the ability to secrete insulin. HNF1A is now well established as a genetic risk factor with a
modest effect in T2D. Mutations in a gene encoding a separate transcription factor with
homology to HNF1A, HNF1B, are also associated with young-onset diabetes and T2D.
There is considerable overlap between the genes associated with these monogenic forms
of diabetes and those associated with increased risk of T2D. The individual mutations
involved in the monogenic disease differ from the polymorphisms predicting suscepti-
bility to T2D and the overall phenotypes also differ in a number of respects. However,
this is different to the example of breast cancer considered in Chapter 9, where the genes
contributing to the familial disease do not appear to make a significant contribution to
susceptibility in sporadic disease.

Conclusions
In this chapter we have considered the results of genetic studies in one of the most com-
mon forms of disease in the developed world, i.e. diabetes mellitus (T1D, T2D, and
MODY). There are profound differences between the genes identified in T1D versus T2D,
corresponding to the different phenotypes for these two diseases. T1D is an autoimmune
disease (see Section 10.1) with all of the features that we would expect with autoimmu-
nity. There is a strong association with the MHC and especially, but not only, with HLA
class II DQB1. There is also a strong association with a number of other genes, especially
the insulin gene INS and a number of immune-regulatory genes, including CTLA4 and
IL2RA. Overall, the genetic studies support the hypothesis that this is an autoimmune
disease. There is significant overlap with many other similar autoimmune and immune-
mediated diseases. However, there are also some differences, e.g. the same HLA-DQB1
allele that is associated with protection from T1D is associated with an increased risk of

2967_Ch10.indd 312 07/07/2015 10:55

 Conclusions 313

multiple sclerosis and narcolepsy. The picture presented is not a simple one and yet a high
proportion of genetic heritability (up to 70%) may now be explained by a relatively small
to moderate group of genetic risk factors that have a mix of very strong to moderate effect
sizes.
Compare this to the situation in T2D where the genetic risk factors are very different from
those identified in T1D. There are no associations with HLA in T2D and there are no
major associations with other immune-regulatory genes. This almost certainly reflects the
different pathogenesis of T1D compared with T2D. Furthermore, most of the identified
and confirmed risk alleles have small effects. The risk genes are different from T1D, but
there is some overlap with some of the early-onset monogenic forms of the disease, for
example the potassium channel genes KCNJ11 and KCNQ1.
Compared with T1D, where current estimates indicate a very significant portion of the
total genetic risk has been identified, in T2D only 10% of the genetic risk has been identi-
fied. However, this is not because there has been a lack of effort to identify the risk genes
in T2D. On the contrary, over 60 different candidate genes have been identified so far.
In the absence of strong associations it is going to be difficult to use the current data from
T2D studies in disease diagnosis or in patient management. Studies will continue and it
is likely that at some point, through a better understanding of the genetic components of
this disease, it will be possible to use this knowledge in patient treatment and management,
and to develop new treatments for this disease. When considered on a global scale, the
ability to offer individualized risk assessment for T2D using genetic information remains
an important target. Future studies will use a combination of new and old technologies,
and there will be an increased use of epigenetics to understand diabetes, especially T2D.
Genotyping has led us this far in many diseases, but phenotyping will produce the next
piece in the jigsaw. It is OK to be selective in terms of deciding which genes to look at as
long as all of the potential candidates are considered at some time. Otherwise the genotyp-
ing will have been a waste of time and the phenotyping work will fall into the same trap
that applied to early genetic association studies, i.e. rejection of positive associations, but
not based on numbers or study design (as in the past), but this time based on absence of
knowledge or absence of an obvious (simple) link.

2967_Ch10.indd 313 07/07/2015 10:55

 314 CHAPTER 10 Genetic Studies on Susceptibility to Diabetes

Further Reading
Books Murphy R, Ellard S & Hattersley AT (2008)
Armstrong L (2014) Epigenetics. Garland Clinical implications of a molecular genetic
Science. This is a useful informative book for classification of monogenic beta-cell diabetes.
students wanting to understand the applica- Nat Clin Pract Endocrinol Metab 4:200–213.
tion of epigenetics to complex disease and epi- Pal A & McCarthy MI (2013) The genetics of
genetics in general. type 2 diabetes and its clinical relevance. Clin
Holt RIG, Cockram C, Flyvbjerg A & Genet 83:297–306.
Goldstein BJ (2010) Textbook of Diabetes, 4th Polychronakos C & Li Q (2011) Understanding
edn. Wiley-Blackwell. type 1 diabetes through genetics: advances and
Parham P (2009) The Immune System, 3rd prospects. Nat Rev Genet 12:781–792.
edn. Garland Science. This is an excellent basic Steck AK & Rewers MJ (2011) Genetics of
immunology book for the student of human type 1 diabetes. Clin Chem 57:176–185.
immunology including links to immunogenet- Todd JA, Bell JI & McDevitt HO (1987) HLA-
ics and some useful data on diabetes. DQβ gene contributes to susceptibility and
Wass JAH, Stewart PM, Amiel SA & resistance to insulin dependent diabetes mel-
Davies MC (2011) Oxford Textbook of litus. Nature 329:599–604. This is a landmark
Endocrinology and Diabetes, 2nd ed. Oxford paper in understanding T1D and the relation-
University Press. ship between HLA and disease.
Toperoff G, Aran D, Kark JD et  al. (2012)
Articles Genome-wide survey reveals predisposing dia-
betes type 2-related DNA methylation varia-
Ashcroft FM & Rorsman P (2012) Diabetes tions in human peripheral blood. Hum Mol
mellitus and the β cell: the last ten years. Cell Genet 21:371–383.
148:1160–1171.
Ueda H, Howson JMM, Esposito L et  al.
Billings LK & Florez JC (2010) The genetics (2003) Association of the T-cell regulatory
of type 2 diabetes: what have we learned from gene CTLA4 with susceptibility to autoim-
GWAS? Ann NY Acad Sci 1212:59–77. mune disease. Nature 423:506–511. This is a
Davies JL, Kawaguchi Y, Bennett ST et  al. very important study on CTLA4 highlighting
(1994) A genome-wide search for human the complexity of association studies even when
type 1 diabetes susceptibility genes. Nature targeted to a specific gene.
371:130–136. This describes one of the first van de Bunt M & Gloyn AL (2010) From
genome-wide genetic studies. This was a pre- genetic association to molecular mechanism.
genome map study and limited, but neverthe- Curr Diab Rep 10:452–466.
less inspirational.
Volkmar M, Dedeurwaerder S, Cunha DA
de Miguel-Yanes JM, Shrader P, Pencina MJ et al. (2012) DNA methylation profiling iden-
et  al. (2011) Genetic risk reclassification for tifies epigenetic dysregulation in pancreatic
type 2 diabetes by age below or above 50 years islets from type 2 diabetic patients. EMBO J
using 40 type 2 diabetes risk single nucleotide 31:1405–1426.
polymorphisms. Diabetes Care 34:121–125.
Drong AW, Lindgren CM & McCarthy MI
(2012) The genetic and epigenetic basis of type Online sources
2 diabetes and obesity. Clin Pharmacol Ther http://www.who.int/diabetes/facts/en
92:707–715. This is an excellent site for finding information
McCarthy MI, Rorsman P & Gloyn AL (2013) on disease prevalence, incidence, morbidity,
TCF7L2 and diabetes: a tale of two tissues, and and mortality, and especially projections for the
of two species. Cell Metab 17:157–159. future.

2967_Ch10.indd 314 07/07/2015 10:55

 CHAPTER
11
Ethical, Social, and Personal
Consequences

In the previous chapters we have considered how, why, and what we hope to achieve by
investigating the genetics of complex disease, and used different example diseases to illus-
trate these points. Each chapter contains several examples of how common genetic variation
increases the risk of common genetically complex (i.e. non-Mendelian) disease. The issue of
common genetic variation being linked to common disease is critical in modern society. One
of the key elements in this consideration is ethics. In this chapter, we will consider different
ethical aspects of enquiry into complex disease with a brief overview of the general philoso-
phy behind ethical enquiry. We will consider lessons from the past, looking at how genetics
has been misused and why it is important to safeguard ourselves against the genetic inquisi-
tion. We will consider the use, and potential misuse, of complex disease genetics in a modern
society and why ethical guidance is so important in genetics. We will also consider the use of
genetic data, what is personal and confidential, and where the boundary of confidentiality
fits into the picture? Finally, we will consider who owns the genome, and the relationship
between commerce and academia. This chapter is designed to open the doors to ideas and
concepts not yet dealt with in other areas of the book; therefore, unlike the other chapters
in this book, this chapter is not crammed full of examples of genetic studies, except where
they are appropriate. Instead, it is designed to stimulate debate and reading for those with
an interest in ethical issues and in the history of genetics. The chapter includes some refer-
ences to the history of genetics that may be sensitive issues for some readers. Nevertheless,
it is thought important to include them. The chapter does not discuss the ethics of work on
animals. This latter omission does not reflect the views of the authors, but simply the fact
that the majority of the book reflects recent genetic research based on human subjects.

2967_Ch11.indd 315 08/07/2015 14:43

 316 CHAPTER 11 Ethical, Social, and Personal Consequences

11.1 Defining Ethics

The Oxford English Dictionary defines ethics as relating to morals or a set of principles of
morals; the latter may be considered as a series of rules of conduct. Ethics is a philosophi-
cal concept and the word is derived from the Greek word ethos. Christopher Tollefsen, in
Biomedical Research and Beyond: Expanding the Ethics of Inquiry (2008), refers to the eth-
ics of enquiry as a term that covers the philosophy of ethics across a range of disciplines.
Ethics applies to many aspects of human life. We are concerned with the ethics of enquiry
as it applies to biomedical research from the bench to the bedside. Tollefsen states there is
no unified ethics of enquiry; however, he goes on to say that by applying a single system-
atic approach we can consider all moral action is action for the sake of human fulfillment
and well-being. At first this seems complicated, but what this statement allows us to do
is to focus on one ethical principle: fulfillment and well-being. From this we are able to
consider both the ethical aspects of the conduct of biomedical research, and the issues
pertaining to the value of enquiry for individuals and different cultures.

There are philosophical arguments for and against ethical constraint

in biomedical research
There are some who argue that to apply moral (ethical) judgment to science is wrong.
However, others argue that as enquiry is a fundamental aspect of human behavior and we
are born to question, it must be controlled through the application of ethical guidelines.
Others believe in the autonomy of science (Figure 11.1). Scientists are not alone in this
view; some other academics have also expressed a wish for pursuit of the truth at any price.
Those against ethical/moral constraints argue that:
• Research could be more effective and perhaps cheaper if we put aside ethical
considerations.
• Research should be judged on the results alone (ethical philosophers refer to this as
consequentialism).
• Knowledge alone is sufficient to justify inquiry (this is a type of Aristotelian ethical
philosophy).
Further possibilities involve discussions about the nature of the state’s relationship with
individuals and politics. Neither of these are strictly ethical issues, but they do impact on
biomedical research. However, a detailed discussion of ethical philosophy is beyond the
scope of this book. We are concerned with the practical implication of ethics in current
research in genetics of complex disease only and readers wishing to delve into ethics in
more detail are directed to Tollefsen (2008) in the first instance.

Figure 11.1: Ethical control versus autonomy in science.

The figure is designed to represent the balance between those
who argue that applying moral ethical judgment in science is
wrong, i.e. scientific autonomy (the need for truth at any price
– black bowl), versus those who say there must be control or
regulation (white bowl). The reader must make his/her own
mind up as to how the scales hang and whether to champion
the dark or the white bowl as this is a matter of debate and
personal moral judgment.

2967_Ch11.indd 316 08/07/2015 14:43

 Defining Ethics 317

What are the practical ethical implications in the study of genetics of

complex disease?
Science is public and, like medicine, law, and journalism, is a genuine profession. It is
public in three respects: the methods applied, the respect obtained, and the use of the
results in terms of products and power. Tollefsen tells us that for any practice to become
a profession it must provide benefits not just to the practitioners, but to a wider soci-
ety. There is no doubt that science provides abundant benefits in healthcare as well as in
agriculture and engineering. Therefore, there is no questioning that under this definition
science is a profession. The importance of this is that professions are regulated and have
rules that govern them. In biomedical sciences we are concerned with the investigation
and the application of the science of medicine, and very strict rules are applied. In prac-
tice, all research applications undergo scrutiny by ethical committees, all new practices are
thoroughly considered by ethical boards and committees, and even ethical committees are
reviewed to ensure appropriate procedures are followed. Scientific results are reviewed by
groups of peers prior to publication and mutual criticism has become an essential brick
in the wall for scientists. Furthermore, scientific information cannot be hidden, especially
when advances rely on collaboration and public funding. This openness assures a level of
ethical quality that could otherwise be missing.

Ethical approval and counseling: good practice

Many problems and issues arise in researching complex disease, but most of them can be
dealt with through good practice in setting up studies and in provision of high-quality
counseling services. When testing patients they must be made aware of the reasons for the
test, the potential outcomes from the test, and the personal impact of the knowledge of
the results on themselves and their families and on their social group. Most ethical com-
mittees will require this as standard practice before permitting the introduction of new
practices or clinical trials. Sponsors, including the National Institutes of Health (NIH) in
the USA, the Medical Research Council (MRC) and Wellcome Trust (both in the UK),
and multinational drug companies, will all expect ethical approval before they accept and
fund research and clinical trials. Seeking ethical approval can be quite a complex process.
Practice and procedure vary between different centers and countries. As this is not the
subject of this book, then for those needing help, the local ethical committee is the correct
starting place.
As with ethical approval, counseling practice varies from country to country and even
between different regions. However, it is becoming increasingly important that counsel-
ing is considered if genotype information is to be released to the tested individuals. In
most clinical research centers genotype information has not been released to the patients
and controls, and the samples are stored as anonymous samples. This means that back-
tracking for identity is not an option. If, however, the intention is to provide the tested
patient or individual with the test data, then counseling is essential both before and after
the test. It may also be appropriate to collect evidence that the individuals understand
the potential impact of this information in the form of signed papers registering their
consent.
The age of computers poses a new problem regarding genetic counseling. Today we can
consent to tests online. Can this be ethical? How can the tester know that the client truly
understands the results of any test performed? The answer is that they cannot. Some may
argue we can never tell whether a client understands the potential impact of a test and even
if they say they do, the results can take them by surprise. However, the matter of whether
the client is present with the counselor or the process occurs online may be irrelevant.

2967_Ch11.indd 317 08/07/2015 14:43

 318 CHAPTER 11 Ethical, Social, and Personal Consequences

Others may say it is a step too far. Opinions will vary and it may be easier for us to ride
this issue out than become entangled within it, but online testing and application is part
of the future.

11.2 Ethics In Genetics: What We Can Learn

from the Past?
Historically, genetics has had something of a bad reputation. This branch of biological sci-
ence has been misused. Genetics itself could be described as the result of twinning Gregor
Mendel’s observations with those of Charles Darwin. Mendel and Darwin were both very
rigorous in the pursuit of their science, but others were far less cautious, willing to create
or stretch hypotheses beyond reasonable limits. It was not long before the idea of survival
of the fittest developed from Darwin’s own observations (by others) was adopted as pure
fact and from this sprang the concept that within populations some were genetically fit,
while others were unfit. This idea led to the development of eugenics by Francis Galton
(incidentally Darwin’s cousin) and social Darwinism promoted by Herbert Spencer. The
Eugenics Movement was mostly based on the idea that the fit members of the population
were those who prospered (i.e. the upper classes), while the unfit were the lower classes.
In Victorian and Edwardian England the differences between the upper classes and the
lower classes were marked, and extreme poverty was widespread amongst the lower classes
(Figure 11.2). Spencer’s idea was based on the notion that poverty and wealth are inevi-
table as they reflect the biological rules that govern society—the biological rules in this
case being the genes. Many of the upper-class intelligentsia were of the opinion that pov-
erty was a natural byproduct of laziness and a lack of thrift among the poor. The Eugenics
Movement, in particular, fed off this idea. In the warped view of the Eugenics Movement,
being lower class and poor must be a genetic trait.

The consequence of the Eugenics Movement and the ideas it spread

were extremely bad news for the developing science of genetics
Eugenics is not a science, it is a misuse of science. There are many examples of the use
of eugenics to justify extremely inhumane activities. Steve Jones, in the introduction to
his book The Language of the Genes (2000), writes a very informative section on eugenics
with examples that include enforced (sometimes secret) sterilizations all the way to the

Upper–rich
Middle
Lower–poor Figure 11.2: Class division in the UK
around 1900. The figure illustrates the
division between the rich (upper classes)
and the poor (lower classes) in the UK
around 1900. The members of the Eugenic
Movement were convinced that poverty
was an inherited trait.

2967_Ch11.indd 318 08/07/2015 14:43

 Ethics In Genetics: What We Can Learn from the Past? 319

death camps of the holocaust (see below). He talks of how Galton supported the idea of
breeding from the best and sterilizing those individuals whose inheritance did not meet
with his approval and how the Eugenics Movement joined the gentle concern for the
unborn with a brutal rejection of the rights of the living. Eugenics was an idea shared by
many members of society from the political left to the far right. The list includes George
Bernard Shaw, Winston Churchill, and Charles Davenport (Professor of Evolutionary
Biology at Harvard).

New Germania
At the lowest level, isolated communities such New Germania in Paraguay (which was
set up by Bernhard Förster and his wife, Elisabeth, the sister of the philosopher Friedrich
Nietzsche), were founded based on selected groups of individuals who set up home in
isolation from their brethren to maintain and develop more pure communities and
strengthen the stock. A glance at the peoples of this area today indicates that the experi-
ment failed, and the descendents of the original settlers are a poor and sickly population.
Unfortunately, that was not the limit of the application of this pseudo-science. In North
America, 25,000 Americans were sterilized because they might pass feeble-mindedness or
criminality to a future generation.

The Monist League

The German embryologist Ernst Haeckel founded the Monist League, which promoted
the use of the biological rules of society for the survival of some races versus others. Of
course Haeckel was himself amongst the select. However, it would be wrong at this stage
to suggest that the idea that one race may be superior to another has only occurred once in
human history. Human history is peppered with examples where one race or subpopula-
tion has considered itself higher, better, or more worthy than others. This idea has been the
most common cause for war between nations since historical records were kept. However,
what the Monist League was promoting was the idea that eugenics provided a scientific
basis for this and it was not long before this idea fell into the wrong hands, i.e. those of
Hitler, who wrote that whoever is not bodily and spiritually healthy and worthy shall not
have the right to pass his suffering in the body of his children. Mass forced (mostly secret)
sterilizations followed. Worse was still to come. The German Society for Race Hygiene,
many members of which carried university doctorates, was formed. This science was used
to justify genocide on a hitherto unimaginable scale.
There is no doubt from the historical record that Hitler was aware of the Eugenics Movement
and used the concept of inherited racial purity to justify the selection processes behind
his most hideous activities. However, as Jones tells us, Hitler was not the only person to
think in this way. One quote, in particular, is surprising: “The unnatural and increasingly
rapid growth of the feeble-minded and insane classes, coupled as it is with steady restric-
tion among the thrifty, energetic and superior stocks constitutes a national and race danger
which is impossible to exaggerate. … I feel that the source from which the stream of mad-
ness is fed should be cut off and sealed before another year has passed.” These words of
Winston Churchill in 1910 would today cause a riot and his immediate dismissal from par-
liament, and rightly so, but this latter quote does put things into context. These ideas, which
seem strange now and are obviously flawed when we look at them in 2015, did not seem so
strange in 1910. However, this is in no way justifies the actions of the past (Figure 11.3).
One idea that eugenics helped spawn was the idea of a link between genetics and crimi-
nality. This began in the early years of the Eugenics Movement, but surprisingly lasted
into the 1960s when studies of males in penal institutions indicated a higher frequency

2967_Ch11.indd 319 08/07/2015 14:43

 320 CHAPTER 11 Ethical, Social, and Personal Consequences

Darwin Mendel

Genetics

Galton

Eugenics
Movement

Monist
Haeckel
League

New
Germania

Figure 11.3: From Darwin and Mendel to eugenics: how science can be misused. The figure illustrates the
period of eugenics, and the main characters and events involved in the misinterpretation and application of
genetics.

of male inmates with an extra Y chromosome (XYY). In fact, this condition occurs in
1:1000 males and is not associated with hyper-aggressiveness as originally suggested,
but is associated with a mildly reduced level of intelligence. The idea was seen once
again in 1993 with a report on the MAOA gene (chromosome X) which encodes mono-
amine oxidase A. Monoamine oxidase A is an enzyme that plays a key role in the catab-
olism of a range of neurotransmittors. The study reported a link between this gene and
criminal activity in a large Dutch family. Unlike the XYY story, this report has stood
the test of time with the caveat that aggressive behavior only occurs in those males car-
rying the polymorphism who have also been maltreated during childhood. Such was
public and scientific ethical concern about promoting the idea of a link between crimi-
nal behavior and genetics that a conference planned to discuss the idea was canceled to
avoid raising controversy that may have had public and societal consequences.
Eugenics has left a scar across the heart of genetics as a science—one that has been hard to
shake off. Given this, it is perhaps most surprising that some of these screwball ideas have
lasted into the new millennium, but even today these ideas occasionally surface as Lone
Frank reports in her book My Beautiful Genome (2011). The message from the past is clear.
We must not allow this ideology to raise its ugly head in the future.
We need to consider several other ethical issues as more and more samples undergo
genome-wide scanning for clinical or personal enquiries:

2967_Ch11.indd 320 08/07/2015 14:43

 Looking Into the Future Use of Genetic Data 321

• How are we going to use the data?

• Who does the data belong to?
• Who should be able to access the data?

11.3 Looking Into the Future Use of

Genetic Data
There are multiple uses for data from studies of complex disease and these can be divided
into two main subgroups: those that impact on the disease itself (i.e. in clinical practice)
and those that are of a more personal nature.

Genetic studies of complex disease will have a major impact on

clinical medicine
This subject has been discussed at length elsewhere in this book and therefore requires
only a brief discussion here. The Human Genome Project (HGP) stated at the outset that
one of the goals of the study was to enable disease genes to be identified so that the infor-
mation could be used in disease diagnosis and patient treatment, and to help us under-
stand disease pathology (Figure 11.4). Though the results so far are mixed, there are some
examples illustrating at least a degree of success in each of these areas. However, not all of
the genetic associations we have dealt with were reported after the HGP report. Some, like
that of HLA-B*27 in ankylosing spondylitis, were reported long before the project began.
Despite this, the association is a great illustration of what is possible. Though the HLA-
B*27 alleles are quite common, the genetic association is so strong that testing can be
helpful in cases where a differential diagnosis is required. The use of genetics in informing
treatment options is also starting to be applied in clinical practice, e.g. the use of abacavir
in the treatment of human immune deficiency virus (HIV-1) induced acquired immune
deficiency syndrome (AIDS). In addition studies in diseases such as Crohn’s disease have

Human
Genome
Project

Improve
Aid to patient
knowledge of Aid to
care and
disease diagnosis
management
pathology

All of these need expert

counseling

Figure 11.4: Promises of the HGP. The figure shows three of the major promises of the HGP with respect to
complex disease. In each case, there is a need for expert counseling as the impact of risk alleles and haplotypes is
very different from that seen in Mendelian disease.

2967_Ch11.indd 321 08/07/2015 14:43

 322 CHAPTER 11 Ethical, Social, and Personal Consequences

identified a large number of risk alleles and this information is helping to advance our
understanding of the pathology of this disease.
The ethical issues concerning these promises
This is all very well, but what are the ethical issues arising from this. One of the major
problems with using genetic polymorphisms in disease diagnosis in complex disease is that
the polymorphisms are often common in the healthy population and have weak associa-
tions with the disease. Therefore, the value of the polymorphisms as diagnostic tools is
limited. Where they have larger effects, the presence of an allele may be a useful adjunct
in diagnosis and particularly in making a differential diagnosis between two diseases, e.g.
HLA-B*27 in ankylosing spondylitis (see above). However, we are faced with a dilemma.
If the allele is common we have to be careful not to over-promote the concept that the
allele is important in disease pathogenesis, because in most complex diseases possession
of the risk allele is neither necessary nor sufficient for the disease to occur. In some cases
there are strong genetic associations with very common polymorphisms, e.g. some human
leukocyte antigens (HLA), where up to 20% or more of the healthy population can carry
the risk allele or one member of the allele family. For example, the DRB1*04 family is the
most common group of the HLA-DRB1 alleles in the UK. This group is associated with
susceptibility and resistance to a range of autoimmune diseases, including rheumatoid
arthritis and autoimmune hepatitis (increased risk), and primary sclerosing cholangitis
(reduced risk). We cannot announce to the population at large that they carry a four- to
five-fold greater risk of a disease without a proper explanation of what this means. The
outcome of simply broadcasting risk values could create global (and unnecessary) panic.
The statistics need to be put into context.
When it comes to therapy there are some clear links between outcome measures and genetic
polymorphism. It is very helpful in some cases to perform genetic testing before a patient
is treated with a specific drug. In the science of pharmacogenetics, where the associations
are strong and the differences (i.e. in response to treatment) are clear, genetic testing will
be increasingly used prior to treatment. Here, the reason for using genetic information and
testing is obvious, e.g. consider the non-response to codeine (for pain relief ) or to β-blockers
(for cardiovascular problems), or the adverse reaction to abacavir used to treat HIV-1 infec-
tion. However, where the strength of the association is weaker, but the outcome poten-
tially lethal (e.g. in the case of some adverse drug reactions), we find ourselves in an ethical
dilemma. To genotype or not to genotype, that is the question. If the incidence of a severe
reaction is 1:100,000 new cases per annum, do we genotype everyone before we prescribe
the drug or do we monitor all new cases for signs and symptoms of adverse reactions? It is a
societal question and one for the ethics committee rather than one that we can answer.
Considering ethical issues is easier when it comes to understanding the disease pathogen-
esis. It is our human instinct that makes us question the world around us and our human-
ity that makes us want to solve the problems of disease. There is unlikely to be anyone who
would feel that we should not perform genetic enquiries to inform the debate on disease
pathogenesis. The question that we must be concerned with is how we handle the data and
use the outcomes from such research. Early publication of data can cause fear amongst the
“at risk” public.

The potential personal impact of data from studies in complex disease

is considerable
Excluding all of the above, which all have personal implications; there are many other ways
in which these data may impact on all of us (Figure 11.5). Lone Frank (2011) describes

2967_Ch11.indd 322 08/07/2015 14:43

 Looking Into the Future Use of Genetic Data 323

Family

Confidence
and Social status
lifestyle
Knowledge
of
genome

Access Access
to to
insurance healthcare

Figure 11.5: The personal impact of knowing your own genotype. The figure illustrates five of the factors
that may be impacted by genetic testing. Individual responses to genetic testing can have negative effects on
self-confidence and on family members as well as those tested. In addition, there is potential for testing to have
a negative impact on insurance and healthcare costs depending on local rules and regulations. Social status can
also be impacted. However, there can also be positive impacts in all of these areas.

waiting patiently (or not so patiently) for the postman to deliver the results of genetic
tests for common polymorphisms identified in complex disease. Her book gives a personal
view of how we ourselves may feel waiting for such results. She describes joy when low-
risk alleles are reported, but deep concern when higher-risk alleles are reported. Even with
counseling, the author reports being nervous on some occasions. Of course it depends on
what is being tested; consider, for example, BRCA versus the interleukin (IL)-2 receptor
gene IL2R. It depends on the size of effect and on the disease or diseases that are associated
with the polymorphism (or occasionally the mutation). It depends on your knowledge of
your family medical history, and your understanding of medicine and science. For both
the academic elite and the general public, the level of concern depends on how much we
understand the subject. Do we have an adequate grasp of the concepts before us? Do we
understand genetic risk? Even if we do understand the science there is no promise that
we will be better equipped to cope with the news. Therefore, it is also important to note
that when doctors become patients they deserve the same level of consideration and should
receive the same level of counseling as the less informed member of the public receives.
Patients being tested for diagnostic reasons and those making independent enquiries are
one group. But, what do we do about volunteers? Here the ethical dilemma is simple—
when testing samples, should we release data to the volunteers? The question may be sim-
ple, but the solution is far from it. We get around this by making all samples anonymous,
so that back-tracking to the original ID is not possible (Figure 11.6). However, this has
limitations. What if we find a polymorphism in our studies that identifies an allele that is
very strongly associated with a severe disease, but a disease that can be easily treated? If we
find that 2% of the healthy controls are carriers for the risk allele, then 2% of our healthy
control volunteers are at increased risk of the disease. As the DNA samples will be handled
as anonymous samples there is no possibility of alerting these volunteers to the risk and giv-
ing them prophylactic treatment. The same individuals may suffer the disease in later life.

2967_Ch11.indd 323 08/07/2015 14:43

 324 CHAPTER 11 Ethical, Social, and Personal Consequences

Clinical cases Self-interest Volunteers

Consent and Consent and

counseling counseling

Genetic testing

Post-testing Anonymous
counseling no post-testing
and counseling or
information information

Figure 11.6: Common (good) practice counseling in different groups. The figure illustrates the importance of
consent and counseling in all groups undergoing genetic testing. Those being tested for clinical purposes need to
give consent for testing and receive appropriate counseling prior to consent being given. Those requesting tests
out of self-interest should be treated in the same way. Those who are being tested as part of a research program
(i.e. volunteers) must also give consent and should also be counseled. However, normal practice in research is
to make all samples anonymous and therefore volunteers should not expect to receive information in return
or counseling after testing. In contrast, both clinical cases and those being tested out of self-interest should be
counseled following testing as well as before.

In complex disease it is always difficult to know how to handle the data, because the size
of effect is small and some of the genetic relationships are very complicated. Nowhere is
this better illustrated than in the major histocompatibility complex (MHC). The majority
of the genetic associations with this area involve odds ratios (ORs) of less than 10, though
values of between 0.02 and 150 have been quoted for different diseases. The problem is
that most HLA polymorphisms and haplotypes associated with disease are common in the
healthy population (often as high as 10–20% for the major allele families, e.g. DRB1*04 ).
In contrast, the diseases are not as common. This means that the majority of those with
the risk allele, family of risk alleles or haplotype do not develop the disease. Therefore,
being positive for an allele or haplotype is not useful in a diagnostic setting. However, it is
easy for the public to get the impression that these associations are diagnostic.

The example of HLA-B*27

We can consider the example from the 1980s of a middle-aged healthy volunteer who
agreed to be involved in a study of HLA and disease. The control was found to have
the HLA-B*27 phenotype by standard serological assay. The close-knit community from
which the volunteers were recruited meant that after HLA typing they would frequently

2967_Ch11.indd 324 08/07/2015 14:43

 Looking Into the Future Use of Genetic Data 325

ask to know their HLA type. On this occasion time passed without any issue until one
day the volunteer rounded on the immunogenetics team declaring their anger that they
had not been told they carried an allele that promotes ankylosing spondylitis. The team’s
defense, that though the association between ankylosing spondylitis and HLA-B*27 is
strong it is not found in all cases, and more importantly only one in approximately 20
HLA-B*27-positives develop the diseases, was no comfort. The volunteer had recently
discovered that a younger member of their family, also an HLA-B*27 carrier, had devel-
oped ankylosing spondylitis. The team were not even able to defend their position with
the observation that the volunteer had not indicated a history of ankylosing spondylitis
and was over the normal age at which they it would develop. Classical geneticists would
use two words: anonymity and counseling. Anonymity may have helped here, but prob-
ably not. Rigorous refusal to disclose the original data would not have been easy in
these circumstances. Counseling would have helped, but then that means either all those
undergoing genotyping would need to be informed of every HLA association that had
been reported and the risk with diseases before testing, or that all those with any high
risk alleles and haplotypes would require counseling after being typed, or both. If we
restricted counseling to the second group, when we consider HLA genotypes this would
mean almost everyone as nearly all the common HLA allele families have some positive
risk associations with one or more diseases. Considering HLA-B*27-positives alone, out
of 20 cases given counseling, only one would really need it. The effect in terms of stress
for the remaining 19 would be enormous. This would be a huge burden and represents
an impossible task.
Interestingly, Francis Collins reports in his book The Language of Life: DNA and the
Revolution in Personalized Medicine (2010) of having had himself tested by three com-
panies for a number of well-known risk alleles. He said “there was one test result that I
thought about just not looking at … the one for Alzheimer’s disease risk.” (We should
note Collins does not specify which Alzheimer’s gene.) Given that we are talking about
one of the best-informed members of the scientific community with regard to genetics
and disease, this is perhaps something of a surprise and it sets the question for the com-
munity at large. However, in considering his statement we need to remember that as yet
there is no cure for Alzheimer’s disease and that may be driving the concern expressed
here. Certainly the issue of whether there is a cure for this disease, or any other disease,
is one that is likely to affect an individual’s choice of whether to be tested at all and if so
which genes to be tested for.

Do we really want to know?

The answer to this question is yes and no. Some people do want to know and they are find-
ing out through direct consumer testing (DCT) as Collins puts it. There is quite a debate
about this in the USA at present concerning whether DCT should be allowed; some say
yes, others (e.g. the American College of Medical Genetics) say no. It is interesting that
Collins himself is among the yes group. The American Society for Human Genetics con-
siders that DCT is OK provided adequate information is available about the limitations
of the tests. As Collins says, personal genomics is here, but caveat emptor.

Confidence and lifestyle

All of the above can influence our confidence. In a stressful world more stress is something
no one wishes for. Knowledge that we have inherited a polymorphism that increases the
risk of disease, no matter how small, can have a big effect on our personality. Some indi-
viduals are more susceptible than others. This trait could also be genetic. Confidence is
hard to measure, but it impacts in all our lives, everyday, at work and at home.

2967_Ch11.indd 325 08/07/2015 14:43

 326 CHAPTER 11 Ethical, Social, and Personal Consequences

The immediate impact of reduced confidence may be stress (as above), but it may also
lead to changes in lifestyle, not all of which may be beneficial. Some individuals when
diagnosed with chronic disease become determined to defeat the illness. They become
champions in the crusade against the disease, getting involved in fundraising events and
even starting new charities. Many will become more self-aware, and diet and exercise
may become higher priorities than previously. However, not everyone reacts positively
nor is increased activity and diet the right path for everyone. Though being told you
have an increased risk of a common disease is not the same as being diagnosed with the
illness, some members of society will react to this type of information as though it were.
In extreme cases, some become withdrawn—almost developing a Munchhausen’s-like
state. This is why and where we need to consider the ethical side of genetics of complex
disease. We need to reduce and avoid this problem, and avoid sending out a negative
message. Counseling is essential.

Family issues
When genetics is mentioned most people think of familial disease, but this is not always
the case in complex disease. Very few complex diseases have large numbers of affected
families and even though there is increased risk for family members, it is not on the scale
of risk associated with Mendelian disease. However, knowledge of genetic susceptibility
can have multiple effects in a family. Even though selected individuals may be aware of the
impact of the risk allele, and its biological and personal significance, there is certainly no
possibility that all family members will be aware. Suddenly we may find family members
selling up and cruising round the world, living every day as if it were their last, only to
run out of money and find themselves impoverished with another 30–40 years of life to
go. Others may take more radical action. However, not everyone will react in this way and
radical responses can be prevented with careful counseling.
One other family issue for more immediate consideration may be in the choice of a mate
or the decision to start a family. It is one thing to be aware of your own risk of a complex
disease; it is another thing to know you are likely to pass that risk on to your offspring.
If you are made aware that you carry an increased risk of a particular disease, you may
decide not have a family. Some people may decide not to marry into families with a high
risk of certain common diseases and others may decide to adopt rather than have chil-
dren themselves. With a growing world population, some might say this is a good thing.
However, when we say this we are ignoring the fact that the reason for the decision may
be completely wrong. The issue is risk and size of risk. In complex disease there is incom-
plete penetrance of the allele in the disease and having the risk allele is neither necessary
nor sufficient for the disease to develop. The size of the risk varies between diseases, and
between risk alleles and groups of risk alleles. For example, the Ueda et al. (2003) paper
on CTLA4 gene polymorphism suggested the maximum OR (risk) in type 1 diabetes was
1.2 and in Graves disease was closer to 1.6. Neither of these is particularly high. Would
you consider not marrying your bride-to-be on the basis of having the CTLA4 risk allele?
How high does the risk need to be? As our knowledge grows, we will be better equipped
to offer counseling and help individuals make informed decisions. These decisions should
not be made without careful counseling. It is also important to be aware that the impact
of this information may be different in different societies and ethnic groups.
There are of course broader issues when it comes to genetics and families. Though we are
concerned with complex disease, we cannot ignore the other uses of genetic testing. In
the third millennium tracking our kindred through multiple generations has become a
worldwide business, not just because of links with disease, but also because of links with

2967_Ch11.indd 326 08/07/2015 14:43

 Looking Into the Future Use of Genetic Data 327

our history. However, understanding our ancestry can be used in both positive and nega-
tive ways. Paternity testing can be used to test for fidelity. There are strong financial and
legal reasons why testing is often requested by the mother or father of a child. For example,
fathers who prove to be surrogates as a consequence of such testing may withdraw their
support for the mother’s child or children. Ultimately it is the child who suffers.
On a more positive note, historical searches can reveal interesting details about our ances-
tors. In the USA, more than 50 companies are engaged in genetic genealogy studies.
Companies such as African Ancestry look specifically for ethnic and geographic matches
between mostly American clients and those of racial and tribal groups from various parts
of Africa. However, not all reports are confirmed and in some cases initial findings have
turned out to be incorrect. At least one high-profile case came to light recently where the
wrong link was proposed. Cases such as this illustrate the need for counseling here too.
Overall, we cannot get away from the need to inform the patient, client, or individual
fully about the consequences of testing as well as the strengths, weaknesses, and limita-
tions of any testing.
One of the most difficult issues in genetic testing is that of prenatal testing. Different coun-
tries have different laws regarding prenatal testing and the consequences of testing. In addi-
tion, opinions vary between different social, ethnic, and religious groups within societies.
In complex disease, genetic testing is not currently used to predict the health of the unborn
and indeed this would be an error. Due to low penetrance, the risk of disease is most often
very low and there is no justification for using such testing. However, we need to make
sure that this type of idea does not sneak in through the back door. We have the example
of prenatal testing for male status as an example of bad practice highlighted in Steve Jones’s
book The Language of the Genes. Termination of pregnancy based on testing for the sex
chromosomes is now less likely than it was 10 years ago, but undoubtedly it still occurs in
some countries and there is much written about this on the web and in papers cited there.

Social status
The societal impact of genetics in complex disease cannot be underestimated. Knowledge
of our genomes and how they may determine our lifestyle, health, even our wealth and
happiness can have a major affect both on our own view of our social standing and on
how others view us. One recent report questioned the value of knowing our genome, with
a gloomy prediction that the cost to our collective mental health is incalculable. Others
disagree saying most people would not be burdened. Opinions are divided.
Just as the Victorian and Edwardian Eugenics Movement was set against the survival of
the unfit, so too can modern society suddenly turn upon those who are seen to be less able.
Anyone with a disability will tell us that for them life is made more difficult through the
actions and attitudes of some members of society. In the context of this book we are not
considering the illness per se. We are considering just the potential impact of carrying the
knowledge of an increased risk of a common trait.
Some people when given bad news will develop a coping strategy based on ignoring the
information (or denial) or on accepting the information and using it to their benefit.
Some court sympathy. Others are activated and become associated with those in similar
circumstances. Other people when given bad news collapse. They go into a mental
meltdown of variable proportions and find it hard to cope with the information. Some
become hypochondriacs. In society as a whole, it is easier to cope with those who have a
positive response to a situation. No one wants to spend time with negative individuals.
However, it is not easy to maintain a positive response.

2967_Ch11.indd 327 08/07/2015 14:43

 328 CHAPTER 11 Ethical, Social, and Personal Consequences

Of course with high risk genes, e.g. BRCA1 and BRCA2 that are linked to breast cancer
and also to ovarian cancer, many of the patients tested positive opt for preventative surgery
rather than risk the illness itself. At its most extreme this can mean having both breasts and
ovaries removed. The personal consequences of this are considerable, with prolonged sur-
gical and medical treatment, post-surgical medication, and potentially early menopause.
The personal and physical impact of all of this together can be quite considerable and the
individual’s societal status can change. There is a question about whether preventative
surgery is appropriate. This question is one that only the individual is equipped to answer.
It can be very positive or it can be very negative. On one hand, an individual may take the
view that they can deal with the problem in a positive way through surgery; on the other
hand, a different individual may take the view that surgery is the only option (this may be
seen by some as a negative response).
How we handle our genetic portfolio is going to be one of the most interesting challenges
of the post-genome era. We need to understand the implications of risk and accept our
situation. We cannot change our genome. Everyone will have some risk alleles that are
protective from and others that are predictive of common traits. The genome informa-
tion itself is not the same as having the trait. Society can relax and we can relax within
it. There is no great societal threat and no threat to our social status unless we allow it
to happen. We need to avoid creating a genetic underclass as in Huxley’s Brave New
World, whereby the human race is divided into subgroups, with the upper class being the
alphas and the lower classes being the betas, etc., all the way down to the epsilons at the
bottom of the (genetic) caste system. To those who scoff at this idea, look at the past. A
mathematician will tell us that by definition, “if anything is possible” then “anything”
includes this possibility.
Access to healthcare and health insurance
One of the downstream consequences of knowing our genomes is that we could face a
problem with healthcare planning and cost. In the UK, the majority of individuals cur-
rently rely on the National Health Service (NHS), which provides healthcare free at the
point of entry for all members of the population. The NHS is paid for through Tax and
National Insurance contributions. Though there is some debate about this provision and
who has rights to automatic NHS treatment, at present no one is suggesting that access
be based on genetic testing. In addition, no one is suggesting that genetic tests be used to
determine how much individuals each pay in Tax or National Insurance. However, that
does not mean this will always be the case, but such change is unlikely. Other countries
have different systems of healthcare, many based on a personal subscription and opt-in
insurance schemes. It is easy to see why a privately funded system is more likely to be
interested in an individual’s genetic portfolio.
Of more concern is the potential application of genetic testing for health and life insur-
ance. Currently in the UK, genetic information does not have to be given to insurance
providers and they are prohibited from asking customers to provide such information. The
situation in the USA is different as the health system is not based on the same plan as the
NHS in the UK. Instead, the vast majority of health provision is through private health
companies. Practice also varies elsewhere and as most insurance companies are global or at
least international, access to genetic data needs to be carefully monitored. In the future it
may be seen as reasonable (by some) for genetic data to be used to set insurance premiums.
Insurance companies understand risk and work with risk models. They are well equipped
to assess the likelihood of adverse circumstances on a personal basis given the appropriate
information. Information on drinking, smoking, and other health issues is routinely gath-
ered when setting up policies. Extending the list to include genes would be a simple step.

2967_Ch11.indd 328 08/07/2015 14:43

 Who Does the Data Belong to? Interacting with Commerce 329

Overall, this is an ethical issue. Currently there are constraints in the UK and in some
other countries, and genotype information does not need to be provided on request. In
the USA, for instance, legislation is in place to outlaw the discriminatory use of predictive
genetic information in health insurance and the workplace. However, that does not mean
that the elevated risk of disability or illness cannot be used against an individual when
setting insurance costs. Privacy is a major issue and defending our individual freedom is
important, i.e. we need to defend our rights. These include the right to equality and free-
dom in the matter of our genome. Allowing this very minor misuse of genetics could be
seen as the first tiny step on the descent into madness.
Other issues
Imagine there was a gene for addiction, fidelity, sexual orientation, intelligence, or
spirituality. In each of these cases there may well be genes that predispose to these
traits. Each gene may carry a bad allele and a good allele. The question is do we want
to know the answer about which alleles we carry. How would we use the data? The
answer lies in each example. These are difficult issues. With regard to sexual orienta-
tion, there are those who wish to be able to say their sexual orientation is all (or partly)
in their genes, and others who want to be able to say it is a matter of choice and genes
have nothing to do with it. Underlying this are often different political agendas and
there are implications whichever explanation one applies. Spirituality is another diffi-
cult issue; again, there are some for and against a genetic explanation for this. Personal
agendas apply. None of the above are illnesses but all of them can influence behavior
and for that reason alone all are worthy of consideration in this chapter. These are both
individual and societal issues.

11.4 Who Does the Data Belong to?

Interacting with Commerce
One of the biggest questions we are faced with is who does the genome belong to and
who owns our individual genomes? One of the big developments in the past 25 years
has been the increasing interaction between academia and industry. In some cases
this has taken on a gladiatorial perspective with the giants of industry and commerce
apparently hammering out their differences in a public arena. This was best illustrated
by the competitive nature of the publication of the Human Genome Mapping Project
(HGMP) in 2001: on one side, academia with the NIH (USA), the Wellcome Trust
and Sanger Centre (UK), and others; on the other side, commerce with J Craig Venter
and Celera. The project started with a proposal in 1990 and the NIH, Sanger Centre
and others began working on mapping the genome. J Craig Venter then offered a more
rapid solution. In the end it worked out fine with the parallel publications in Science and
Nature, but there were times of controversy.
The struggle is also well illustrated in the story of deCODEme. Kari Stefannson, backed
by a number of venture capitalists, set up the company to genotype the entire population
of Iceland. There was bitter local opposition to the idea that the genome could be owned
by private business. Interestingly, the idea not only offended the Icelanders, but anger
spread far and wide. In the end the company had great success identifying a large number
of risk alleles for complex disease. For a short period at the end of the 2010s, deCODEme
worked as a diagnostics company offering direct to consumer genetic testing focusing on
that part of healthcare associated with assessing and preventing individual disease risk.

2967_Ch11.indd 329 08/07/2015 14:43

 330 CHAPTER 11 Ethical, Social, and Personal Consequences

Commerce Academia Figure 11.7: Interaction between commerce

and academia. The figure illustrates the increasing
collaboration between commerce and academia.
Almost all genetic testing is now being partly
outsourced and commerce has long since played a
very significant role in genetic research.

However, more recently the parent company deCODE (the major research base for the
company) has been purchased by Amgen. The service offered by deCODEme may not be
offered in future. The problem for commercial companies like deCODE and deCODEme
is that they need to generate payback from their investments. From a pure academic point
of view this can be a negative relationship, though individuals have different opinions
(Figure 11.7).
Many academics would rather see genomes mapped by university research laboratories
than in commercial companies. Unfortunately, there is no longer any option to exclude
commerce. Commerce has provided the technology for the great advances we have made.
Commerce has been involved frequently at the center of all we have done. We do not
exist in ivory towers in isolation any more. Our materials, apart from the DNA samples
we have collected, are all purchased. The equipment we use has been invented and devel-
oped by commerce, and today it is cheaper to contract out genotyping for all but a few
genes than it is to genotype in-house. As a consequence, we are inevitably tied to com-
merce as commerce is tied to us. Outsourcing or contracting out is going to be a major
part of the future.
Once more there are ethical issues, but there is no problem provided the anonymity of
the material sent out is maintained. This needs to be very strictly governed and assessed to
ensure compliance with ethical standards.

Do I own my genome?
Accepting that commerce is part of the current package, we need to know if our genomes
are our own possession or belong to another. If we allow others to genotype or sequence
our genomes at no cost it is possible that the sequence/genotype produced will belong
to the group who did the work. This could be a commercial company or it could be a
research group. However, just as with some tumor cell lines that have been grown in
laboratories for many years, if the research group or commercial company failed to get
permission or explain the situation clearly and obtain written consent, then this could be
contestable. Different countries have different regulations on this and it could be costly to
contest such an issue.
It is certainly true that a number of the genetics companies would like to develop geno-
typing techniques for the detection of common risk genes with large effects and thereby
be able to offer a commercial genetic testing service. Some companies concentrate on
just a few genes, e.g. the company Myriad Genetics (Salt Lake City, USA) have a pat-
ent to test the BRCA genes, others such as deCODEme (Iceland) and 23andME (USA)
will assess 500,000 to 1 million single nucleotide polymorphisms (SNPs) and provide us
with a personal profile, but they have not yet obtained a patent for the genes. Myriad

2967_Ch11.indd 330 08/07/2015 14:43

 Who Does the Data Belong to? Interacting with Commerce 331

Genetics not only tests the genes, but they recently (February 2013) won a court case
enabling them to patent a number of cancer genes in Australia. Appeals against this rul-
ing in have been dismissed. Protesters in Australia say that this should not be allowed
and it could have a major affect on future research in cancer. The depth of debate about
patents is reflected in Luigi Palombi’s book Gene Cartels: Biotech Patents in the Age of Free
Trade (2009). The author states that “no matter how important it is to identify a gene
linked to a disease, it’s still not something that Myriad or anyone else has invented;” he
goes on, “Politicians must not change the law to prevent patenting of genetic materials”
(http://www.bloomberg.com/news/2013-02-15/).
The American Association for Molecular Pathology and the American College of Medical
Genetics have been quoted as saying they are worried that the company is trying to get
legal ownership of part of the human body. However, the situation in the USA is quite dif-
ferent. A Supreme Court ruling in June 2013, authored by Justice Clarence Thomas, states
that naturally occurring DNA segments are not patentable. This will apply to a vast range
of testable genes. However, the ruling permits edited forms of genes not found in nature to
be patented. In the USA there are two major implications from this ruling. Firstly, genetic
testing will become cheaper as companies will compete in the testing market. Second, the
cost of whole genome sequencing is likely to fall. This is because the restriction that is
imposed by patenting sequences, in the genome, is no longer an issue and therefore whole
genome sequence is not likely to infringe patents.
One of Myriad’s fact sheets states that under USA patent law:
“No-one can patent anyone’s genes. Genes consist of DNA that is naturally occur-
ring in a person’s body and as products of nature [they] are not patentable. In order
to unravel the mysteries of what genes do, researchers have had to separate them
from the rest of the DNA by producing man-made copies of only that portion of
the gene that provides instructions for making proteins (only about 2% of the total
DNA in your body). These man-made copies, called “isolated DNA,” are unique
chemical compositions not found in nature or the human body. The U.S. Patent
and Trademark Office has been granting patents on isolated DNA to universities,
hospitals, patient advocacy groups and companies for over 30 years.”
Currently this is correct. However, there have been efforts to appeal some of these laws in
the USA but so far they have failed.
Some scientists argue that their work has been affected because of the costs of pay-
ing royalties and in some cases they have even received letters demanding they stop
using patented inventions. In response, the biotech companies may reply arguing that
if they cannot protect their inventions, then they cannot compete with others in the
market place.

Counseling the customer

In terms of ethics, it may be ethical for commercial companies to hold patents and provide
services, but there is a problem with how the data are presented to the individuals they
test. Should individuals simply be provided with the data and be left to get on with it
without counseling? There is a very strong argument against this. In the UK, the Human
Genetics Commission is quite concerned that the lay public will simply not understand
the information and the effect, without counseling, could be similar to that described
above. The situation is more complex in some states in the USA, which have prohibited
testing unless prescribed by a medical practitioner.

2967_Ch11.indd 331 08/07/2015 14:43

 332 CHAPTER 11 Ethical, Social, and Personal Consequences

11.5 who should be able to access the data?

There is a clear overlap between sections in this chapter and some of the ideas in this sec-
tion have also been considered in previous sections. There are those who would like to
be able to use genetic tests to determine our general fitness. Where this information is to
be used for productive reasons, there is no problem (e.g. for health advice etc). However,
if the data were then to be used to determine our financial contribution to the NHS
or equivalent scheme and our health and life insurance, then there would be problems.
Currently, in the UK and most other European countries, insurers cannot ask for this
data, but in the future they may be able to. The UK Data Protection Act is supposed to act
as a foundation for any company to follow when requesting access to personal informa-
tion, protecting the individual, and conserving their personal rights. Most importantly, it
requires companies to obtain individual consent before obtaining personal information.
Once again, by playing with our genes we are potentially opening Pandora’s Box. There
is an echo from the past here. In the 1970s, there was a great deal of concern about the
use of recombinant DNA to develop sequences that are not produced in nature for use
in the agrochemical industry. In the past we have seen examples ranging from the genetic
engineering of strawberries, antifreeze genes to the development of pigs with genes for
Alzheimer’s disease. The concerns of the 1970s and 1980s have not gone away. At the
present time strict guidelines exist for both current and future research and practice. The
use of these guidelines goes far beyond humans.

Conclusions
Looking at the genetics of complex diseases from a social point of view allows us to con-
sider the ethical issues associated with it. Ethics is best defined as the fulfillment of well-
being. Some would suggest this is not part of science and argue for the pursuit of truth
at any price. Others would argue that ethics is part of science, and that we should apply
more rigid rules and laws. When it comes to the genetics of complex disease, ethics refers
to the issues around testing, counseling, data handling and storage, publication, and access
to data.
Genetics has been misused in the past and for that reason there is a higher level of sus-
picion regarding genetics than for many other branches of science. Through the dark
pseudo-science of eugenics, genetics became attached to some of the darkest chapters in
human history. Eugenics illustrates the misuse of science extremely well. It also reminds us
not to allow our science to be misused in the future.
Given the promises of the HGP, it is important to understand how ethical and soci-
etal issues are themselves composite parts of the puzzle. Using genetics in diagnosis is a
key example. Should we, or should we not, test. Using genetics in disease treatment and
patient care also raises the same issues. Testing can be very advantageous for a patient for
whom a new treatment is seen as having a high likelihood of a good response, but it can
be to the disadvantage to a patient who, after testing, finds out that they are genetically
less likely to respond to a newly developed potent drug. No one would argue that using
genetic data to understand disease pathogenesis is not worthwhile. However, if as part of
the testing risk alleles are identified in healthy individuals we are once again faced with an
ethical and societal dilemma. We need to be aware of personal and societal issues relating
to confidence, confidentiality, social status, access to healthcare, and health insurance in
relation to all of these promises.

2967_Ch11.indd 332 08/07/2015 14:43

 Conclusions 333

What is the relationship between the individual, commerce, and public bodies? Who owns
the genome, who owns an individual’s DNA sequence, and who should be able to access
an individual’s personal genetic data? Science and commerce are strongly interlinked in
genetics—more so now than ever before. Separating them is not possible, but protecting
one’s data is. Laws and regulations vary in different countries; however, most countries
currently do have some form of regulation to restrict access, especially the USA and UK.
All in all, the ethical issues are relatively straightforward in complex disease research at
present. There are rules and guidelines in the UK; rules on consent and the storage of
human tissue (e.g. the Human Tissue Act) are designed to prevent the misuse of materials
and enable the best practice to be followed. In the USA, it took 12 years to get a bill passed
that prohibited DNA being used to discriminate against individuals. The bill, a genetic
information non-discrimination act, was signed in the Oval Office on 4 May 2008.
Ethics is a philosophical activity, and therefore views and ideas change, and the debate
about the ethics of enquiry will continue. New challenges will arise with new procedures
and new opportunities. We need to maintain a balanced opinion. We have seen from the
past what happens when that balance is lost and the extremists have free reign. We cannot
allow that to happen again. Nor should we stifle scientific advancement. Simple measures,
such as informed consent, not only for invasive research but also for data collection, are
important. In many ways the ethics of scientific investigation is far more advanced than
other areas of enquiry and altogether this is a very positive sign for the future.

2967_Ch11.indd 333 08/07/2015 14:43

 334 CHAPTER 11 Ethical, Social, and Personal Consequences

Further Reading
Books Tollefsen CO (2008) Biomedical Research
Collins F (2010) The Language of Life: DNA and Beyond: Expanding the Ethics of Inquiry.
and the Revolution in Personalised Medicine. Routledge. This book provides insight into the
Harper Collins. This is excellent book to delve ethical basis of scientific enquiry from a philo-
into to look at the whys and wherefores of gene sophical perspective.
hunting in complex disease.
Frank L (2011) My Beautiful Genome: Articles
Exposing Our Genetic Future, One Quirk at a Ueda H, Howson JM, Esposito L et al. (2003)
Time. Oneworld. This is an excellent book for Association of the T-cell regulatory gene CTLA4
the lay public and scientist alike. It is interest- with susceptibility to autoimmune disease.
ing, controversial, and humorous, and perfect Nature 423:506–511. This is a very important
for the student of complex disease genetics. study on CTLA4 highlighting the complexity
Jones S (2000) The Language of the Genes, 3rd of association studies even when targeted to a
ed. Harper Collins. This book has a very good specific gene.
introduction for those with an interest in the
history of genetics.
Online sources
Maynard-Moody S (1995) The Dilemma of
the Fetus: Fetal Research, Medical Progress and http://23andme.com
Moral Politics. St Martins. This site offers genotype information about
health and ancestry for a relatively low cost with
Palombi L (2009) Gene Cartels: Biotech Patents a rapid response. The site enables the reader to see
in the Age of Free Trade. Edward Elgar. This what is offered commercially by such companies.
book by patent law academic Luigi Palombi is
a timely contribution to heated global debates http://www.decode.com
about the ownership of human genes. The correct current site for deCODEme is that
Parham P (2009) The Immune System, 3rd ed. shown above. deCODE is part of the Amgen
Garland Science. Though the general subject of company.
this book is immunology, it is a good source http://www.myriad.com
for those interested in immunogenetics and the Myriad Genetics is a company from Salt Lake
impact of MHC alleles/haplotypes in complex City, USA with interesting patents on a num-
disease. ber of cancer genes. They offer testing for a
Spector T (2012) Identically Different: Why variety of common cancers and are working on
You Can Change Your Genes. Weidenfeld & genetic diagnosis for a number of other condi-
Nicholson. This book champions the subject tions. Their patent rights are a matter of some
of epigenetics in complex traits and provides controversy at the time of writing, and serve as
a different perspective for the scientist. The an example of the division between academia
study of identical twins represents a gold stan- and commerce.
dard for geneticists, but it also presents a chal- http://www.gtglabs.com
lenge because they are rare and because they are The Genetic Technologies website offers a vari-
genetically identical at birth. Therefore, testing ety of services for genotyping common diseases
inherited polymorphisms is of little value, but in humans and animals. Another informative
identifying concordance levels between identi- site that illustrates the potential for commercial
cal versus non-identical twins is valuable. Novel exploitation of genetic research, whatever our
sequences (new mutations) are informative and own individual views are.
epigenetic changes are also informative. In the
era of sequencing, more use may be made of http://www.hta.gov.uk
identical twin sets. Spector also tackles a series This site contains details about the Human
of occasional controversial topics in this book, Tissue Act (or links to) and is updated regularly.
some of which are covered briefly in this chapter. http://www.patient.co.uk/doctor/Medical-
Strachan T, Goodship J & Chinnery P (2014) Ethics.htm
Genetics and Genomics in Medicine. Garland This site and the one above are two sites for the
Science. UK Human Genetics Commission: one deals

2967_Ch11.indd 334 08/07/2015 14:43

 Further Reading 335

with general issues, and the other deals spe- http://www.patientslikeme.com

cifically with ethics and good medical practice. This is a website for patients that allows them
These are excellent sites for medical students and to share their experience and knowledge with
trainees in medical research and medical sciences. others who suffer from the same and similar
http://www.familytreeDNA.com diseases. Some may question the use of this
This website offers to test the Y chromosome, and the openness of the site. There are certainly
mitochondrial genomes, and other chromo- ethical issues about this type of site concerning
somal markers, providing genealogy data using the presence or absence of counseling, and the
a databank from close to 500,000 individuals. accuracy and peer review of information added
http://www.oxfordancestors.com to it.
A UK-based company that offers a range of ser- http://www.snpedia.com/index.php/Promethease
vices from paternity and maternity testing to A web-based program that uses a tool to build
genealogy testing. a report on a system called SNPedia that can be
http://www.africanancestry.com applied to data from customers of 23andMe.
A US-based company that traces African ances- com, FamilyTreeDNA.com, Ancestry.com,
try aimed specifically at the African-American and others. It allows the customer to link SNPs
market. to diseases and traits.

2967_Ch11.indd 335 08/07/2015 14:43

2697_Prelims.indd 2 09/07/2015 09:35
 CHAPTER
12
Sequencing Technology
and the Future of Complex
Disease Genetics

In this chapter, we will discuss the present and future of complex disease genetics, look-
ing at what needs to be done to bring us closer to meeting the promises of the Human
Genome Project (HGP). Among the many promises of the HGP were the future use of
genetics in disease diagnosis, disease treatment, patient management, the development of
novel therapies (including new personalized therapies), and a better understanding of the
pathogenesis of the genetics of common complex non-Mendelian disease. Throughout
this book we have consistently focused on these points and the extent to which these out-
comes have or have not been achieved.
Many of the techniques and technologies discussed in this chapter are also discussed in
previous chapters of this book, for example genome-wide association studies (GWAS).
The purpose of this chapter is to look into the future, and consider how techniques and
technologies such as next-generation sequencing (NGS), genotyping the transcriptome,
the expanding field of epigenetics, the use of imputation analysis, the future for GWAS,
and the value of the HapMap, 1000 Genomes Project, and ENCODE may be applied to
further advance our understanding of complex disease. In addition, we will also consider
metagenomics as applied to the bacteria in our guts.

2967_Ch12.indd 337 08/07/2015 14:46

 338 CHAPTER 12 Sequencing Technology and the Future

12.1 DNA Sequencing: The Past, Present,

and Future
The key to understanding genetics in complex disease lies with our ability to sequence
DNA, either searching for allelic variation in individual genes or across the whole genome.
However, prior to 1977, sequencing was very difficult and considered as an impossible task
by some scientists. Two things changed this: the development of more robust sequenc-
ing methods by Frederick Sanger and colleagues in 1977, and the advent of the use of
the polymerase chain reaction (PCR) analysis (whereby a polymerase enzyme is used to
amplify short DNA sequences) by Mullis et al. in 1986. PCR amplification was rapidly
adopted as the method of choice for genotyping in both clinical and research laboratories,
and automated sequencing transformed DNA sequencing from a dream to a reality.
Since then the technological developments applied to genetic research in complex dis-
ease have been exceptional. Many of the studies discussed elsewhere in this book concern
genotyping single nucleotide polymorphisms (SNPs) as markers of disease risk, either at
individual or multiple candidate loci, for chromosomal regions (including whole chromo-
somes), or across the whole genome. However, while this work has been going on, other
researchers have been increasingly focusing their efforts on the possibilities of sequencing
whole genomes. This approach is completely different to that used in more conventional
studies of SNPs where only selected SNPs are genotyped. In whole-genome sequencing,
the entire genome is sequenced and nothing is left out. This technology has now become
a reality and the pace of development within this technology has increased year on year.
For example, it took almost 10 years and an uncountable number of man-hours to map
the first human genome. In 2008, complete human genomes belonging to the three main
population groups were sequenced: one European, one Asian, and one African. Two
years later, in 2010, the first genome of an ancient human was published in Nature by
Rasmussen et al.. In November 2012, the 1000 Genomes Project announced the publi-
cation of 1092 human genome sequences (http://www.1000genomes.org) and it is now
possible using NGS technology to produce millions of DNA sequence reads, providing a
full genome sequence in a much shorter period of time (weeks and months, not years). At
present, the study of genetics is evolving so rapidly that today’s technologies may be out
of date by tomorrow.
Due to the massive progress in genetics in the early part of the twenty-first century it is
difficult to accurately predict what technological trends there will be in the future. High-
throughput technologies are rapidly changing the landscape of genetics, providing the
ability to answer questions with exceptional speed. New DNA-sequencing technologies
are expected to be widely employed in research into the genetic basis of common disease.
It is quite clear reading from the past that progress has been made, but there is still much
to do. Will we reach our final goals? Or will we end up with more questions than answers?

The development of DNA sequencing using the Sanger sequencing

technique opened the way to sequencing the genome
The most direct method for determining the information contained in a gene is DNA
sequencing. Sequencing allows the genetic information contained in a DNA molecule to
be read, and can provide an enormous amount of information about protein sequence,
gene structure, and function.
In 1977, Sanger et al. described a method to chemically decode the sequence of DNA
based on DNA replication relying on the incorporation of dideoxynucleotide triphosphates

2967_Ch12.indd 338 08/07/2015 14:46

 DNA Sequencing: The Past, Present,and Future 339

(a) (b)
P P
P P
P P

5’ CH2 O Base 5’ CH2 O Base

3’ 3’
OH H No OH

Deoxynucleotide triphosphate Dideoxynucleotide triphosphate

Figure 12.1: Decoding the DNA sequence during DNA replication using ddNTPs. The figure shows the
structure of (a) the normal substrate dNTP, which carries a hydroxyl group, and (b) the abnormal substrate ddNTP,
which does not carry the hydroxyl group. The interaction with the dNTP allows the DNA strand to grow, whereas
the interaction with the ddNTP substrate terminates the growth.

(ddNTP) during the synthesis of a DNA strand. The structures of dNTP and ddNTP
are illustrated in Figure 12.1. Dideoxynucleotide triphosphate molecules are identical
to deoxyribonucleoside triphosphates (dNTPs), except that they lack a hydroxyl group
(-OH) at the 3′ end. Sanger reasoned that if ddNTP is added to a growing DNA strand,
the strand can no longer grow because the dideoxyribonucleotide is missing the 3′ hydroxyl
group. The absence of a hydroxyl group prevents the formation of a phosphodiester bond
between the DNA and the new nucleotide. Consequently, after ddNTP has been incorpo-
rated into the DNA strand, no more nucleotides can be added and thus the incorporation
of ddNTPs terminates DNA synthesis.
In 1986, automated DNA sequencing methods based on the Sanger method were devel-
oped. These machines used fluorescent dyes to label each ddNTP, allowing the sequencing
reactions to be quickly monitored. The mechanics of Sanger sequencing are simple and
they are illustrated in Figure 12.2. First, the target DNA molecule is denatured into two
single strands by heating. A primer or a short fragment of DNA is then used to trigger
the sequencing reaction. The primer anneals adjacent to the sequence of interest and is
extended by a DNA polymerase. During the extension reaction the sequence chain is ter-
minated by the random incorporation of fluorescently labeled dideoxynucleotides, which
have complementary identity to the base on the opposite strand. Each ddNTP is labeled
with a different color fluorescent dye corresponding to one of the four bases (A, T, C, and
G). Once the extension reaction has been terminated, the resulting mixture containing
fluorescently labeled DNA strands of varying length is separated by capillary electropho-
resis. The final results are read as fluorescent peaks and are used to determine the DNA
sequence.
Sanger sequencing was used to map the first human genome; it has a base accuracy of
approximately 99.9% and can sequence DNA fragments up to 1 kb in length. Sanger
sequencing is used clinically to identify mutations in selected Mendelian disease genes.
However, the level of sensitivity of testing using the Sanger technique (generally estimated
at 10–20%) may be insufficient for the detection of somatic gene mutations in solid
tumors and in acute leukemia, and for the characterization of complex microbiological
specimens. In addition, it is not able to achieve efficient high throughput for analyzing
complex diploid genomes at low cost. Since the late 1990s, researchers in both academia
and industry have revisited DNA sequencing methods, creating a new generation of DNA
sequencing methodologies—the so-called NGS technologies.

2967_Ch12.indd 339 08/07/2015 14:46

 340 CHAPTER 12 Sequencing Technology and the Future

Cycle sequencing reaction:

Primer Template DNA

dATP, dTTP, dCTP, dGTP
ddATP, ddTTP, ddCTP, ddGTP
+ + CGATTGGCTAGT

(a)
C
CG
CGA
CGAT
CGATT
CGATTG
CGATTGG
CGATTGGC
CGATTGGCT
CGATTGGCTA
CGATTGGCTAG
CGATTGGCTAGT

(b)

Figure 12.2: Sanger sequencing. The figure illustrates the principles of Sanger sequencing. (a) DNA is denatured
into two strands by heating and is treated with a primer to trigger sequencing. The primer anneals with the
DNA sequence of interest and is extended by a polymerase enzyme. The sequence chain may be terminated at
any time during this extension phase by the addition of random fluorescently labeled ddNTPs. Each ddNTP is
labeled with a different color corresponding to one of the four nucleotide bases. (b) Once the extension reaction
has been terminated, the DNA strands of varying length are separated by capillary electrophoresis and the
fluorescent peaks can be used to interpret the sequence. (From Thomas A [2015] Introducing Genetics, 2nd edn.
Garland Science.)

The new era: next-generation DNA sequencing

The sequencing revolution began in 2005 with the introduction of the
­sequencing-by-synthesis technology developed by 454 Life Sciences and the multiplex
polony sequencing protocol. Following on from this, a number of different commer-
cial next-generation DNA sequencing systems started to appear in the market place.

2967_Ch12.indd 340 08/07/2015 14:46

 DNA Sequencing: The Past, Present,and Future 341

These new sequencing machines all have the capacity to deliver significantly cheaper
and faster sequencing, and are having a significant impact in bioscience. Most NGS
technologies do sequencing in parallel, which means that hundreds of thousands or
even millions of DNA fragments are simultaneously sequenced, resulting in very high
throughput levels. It is likely that in the future NGS will become more widespread and
better established. A comprehensive technical overview of each platform is beyond the
scope of this book, but we will look briefly at the fundamentals of a number of them.
However, as this field is rapidly advancing, readers are advised to refer to the different
manufacturer’s websites and other sources referenced in this chapter for the most up-
to-date information on these technologies.
In each of the cases described below, the output of genome sequencing is not the whole
DNA sequence as a single item. Instead a series of sequences of short DNA fragments
called reads is created. These need to be assembled using bioinformatics software in order
to determine the complete sequence of the whole DNA sample. The bioinformatics step
is challenging and it needs to be performed with powerful computers able to handle the
very large amounts of data generated by the sequencing platforms. A comparison of some
of the different platforms can be found in Table 12.1.

Table 12.1 A comparison of some commercially available second-generation DNA sequencing

platforms.

Life Technologies
Roche/454 SOLiD Illumina HiSeq 2000
Library Emulsion PCR on bead Emulsion PCR on bead Enzymatic
amplification surface surface amplification on glass
method surface
Sequencing Polymerase-mediated Ligase-mediated Polymerase-mediated
method incorporation of addition of two-base- incorporation of end-
unlabeled nucleotides encoded fluorescent blocked fluorescent
oligonucleotides nucleotides
Detection Light emitted from Fluorescent emission Fluorescent emission
method secondary reactions from ligated dye-labeled from incorporated dye-
initiated by release of oligonucleotides labeled nucleotides
pyrophosphate
Post- Not applicable Chemical cleavage Chemical cleavage of
incorporation (unlabeled nucleotides removes fluorescent fluorescent dye and 3′
method are added in a base- dye and 3′ end of blocking group
specific fashion, oligonucleotide
followed by detection)
Error model Substitution errors rare, End of read End of read
insertion/deletion errors substitution errors substitution errors
at homopolymers
Read length 400 bp/variable length 75 bp/50 + 25 bp 150 bp/100 + 100 bp
(fragment/ mate pairs
paired end)

From Mardis ER (2011) Nature 470:198–203. With permission from Macmillan Publishers Ltd.

2967_Ch12.indd 341 08/07/2015 14:46

 342 CHAPTER 12 Sequencing Technology and the Future

(a)

(b) Break
Clonal microreactors,
amplification and enrich for
occurs inside DNA-positive
microreactors beads

Load beads into

PicoTiterPlateTM

Load enzyme
beads and
centrifuge

Figure 12.3: An overview of the Roche/454 sequencing technique: 454 GS FLX sequencer workflow.
(a) Isolated genomic DNA is fragmented, ligated to adapters, and denatured into single strands. (b) Each fragment
is bound to beads in a proportion of 1:1 and each fragment–bead complex is isolated in droplets of a water/oil
mixture. PCR amplification is then performed within each droplet, resulting in beads carrying millions of copies
of a unique DNA template. The emulsion is broken down and the DNA strands are denatured; beads carrying
single-stranded DNA templates are enriched and are randomly deposited into wells of a fiber-optic slide. Clonally
amplified beads generated by emulsion PCR serve as sequencing features where the pyrosequencing is carried
out. (Adapted from Margulies M, Egholm M, Altman WE et al. [2005] Nature 437:376–380. With permission from
Macmillan Publishers Ltd.)

The 454 technology (http://www.454.com) was launched in 2005 by 454 Life Sciences
when Margulies et al. published the entire genome of the bacterium Mycoplasma geni-
talia with 96% coverage and 99.96% accuracy in a single GS 20 run. In 2007, Roche
Applied Science acquired 454 Life Sciences and introduced the second version of the
454 sequencer, the GS FLX sequencer, which has the ability to sequence a longer
sequence per run with longer read length. The 454 GS FLX sequencer is illustrated in
Figures 12.3 and 12.4.
An alternative platform is the Applied Biosystems SOLiD (Supported Oligonucleotide
Ligation and Detection) platform (http://www.appliedbiosystems.com). This uses
short-read sequencing technology based on ligation. This approach was applied in 2005
to re-sequence an evolved strain of Escherichia coli and is similar to the 454 approach
(above). One of the advantages of the SOLiD technique is that it uses an offset sequenc-
ing primer strategy. Therefore, each nucleotide in the sequence is interrogated twice.
A given nucleotide in the template sequence will generate two different fluorescent

2967_Ch12.indd 342 08/07/2015 14:46

 DNA Sequencing: The Past, Present,and Future 343

Capture
bead

T
A
TG
ACGCTAGCC Single-stranded
A
TA DNA template
C
TG Polymerase

G dGTP
G
PPi
Apyrase
Sulfurylase + APS
ATP
Luciferase
Light

Signal CDD camera

detection
Pyrogram

A T GC A TG C A T G Nucleotide added

ATACTG Nucleotide sequence

Figure 12.4: The principles of pyrosequencing technology. The figure illustrates the basic principles
of pyrosequencing technology. DNA templates are linked to a capture bead and then exposed to multiple
sequencing rounds. Only one nucleotide is added in each round. As each nucleotide (dGTP in this example) is
incorporated into the DNA sequence, inorganic pyrophosphate (PPi) is released which reacts with adenosine
5′-phosphosulfate (APS) and sulfurylase to generate ATP. ATP is then used as a substrate for luciferase to
generate light, which can be detected and quantified. (From Armougom F & Raoult D [2009] J Comp Sci Syst
Biol 2:74–92.)

signals based on the identity of the neighboring base. As a result, the false-positive rate
for mutation detection is reduced, as a SNP will generate two color changes when com-
pared with the reference sequence. At the end of a 6-day run, the SOLiD instrument
is capable of generating 4 Gb of sequencing data. The SOLiD method is illustrated in
Figure 12.5.
The Illumina Genome Analyzer produced by the Illumina Corporation (http://www.
illu​mina.com) is a short-read sequencing platform. This sequencing-by-synthesis process
on an Illumina GA IIx appears to be the most widely used platform (at the time of writ-
ing). The method is said to generate read lengths of 35 bp with greater than 99% raw base
accuracy and an overall throughput of approximately 5 Gb over a 3-day run (Figure 12.6).

The upcoming era: third-generation sequencing

To date, most of these advances have been applied within the research setting; however,
these technologies will increasingly be used in the future to provide accurate, rapid, and
low-cost information in clinical practice.

2967_Ch12.indd 343 08/07/2015 14:46

 344 CHAPTER 12 Sequencing Technology and the Future

(a) Fragment library

Primers P1<<P2
P1
Mate-paired library OR + + Polymerase + coupled
beads

(b)

PCR
3’

Emulsion
modification

(c)
Deposition
Bead

(d)
1. Prime and Ligate POH 5. Repeat steps 1-4 to extend sequence
+ Ligase Ligation cycle 1 2 3 4 5 6 7 ... (n cycles)

AT TT GT GT TT CA CG
TA AA CA CA AA GT CG
Universal seq primer (n) AT 3’
1μm
bead 3’
P1 adapter TA Template sequence
6. Primer reset
2. Image Excite Fluorescence

Universal seq primer (n-1)

3’ 2. Primer reset 1. Melt off extended
sequence
3’
TA 1μm
bead 3’
3. Cap unextended strand
Phosphatase
PO4
7. Repeat steps 1-5 with new primer

3’
–1
4. Cleave off fluor
Cleavage agent Universal seq primer (n-1) AT AT AT TC AA TA CC
3’
1μm
HO bead 3’
T GT GC AG TT AT GG
AT
P
3’
TA
8. Repeat reset with, n-2, n-3, n-4 primer

Read position 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35
Primer round

Bridge probe

Bridge probe

Bridge probe

Indicates positions of interrogation Ligation cycle 1 2 3 4 5 6 7

2967_Ch12.indd 344 08/07/2015 14:46

 DNA Sequencing: The Past, Present,and Future 345

The third generation in the history of sequencing is now emerging and is expected to bring
new insights on DNA sequences. Third-generation sequencing has two main advantages
compared with second-generation technology (above). The first main advantage is that
the PCR step is no longer needed before sequencing. Though PCR amplification increases
the number of copies of DNA fragment, it can introduce errors into the sequence. The
ideal DNA sequencing platform, which is what the third-generation platform aims to be,
would combine the advantages of high-throughput and rapid-sequence technology with
the capability to sequence long stretches of DNA directly from a single DNA molecule,
without any pre-sequencing PCR step. This would prevent errors introduced by PCR
amplification being introduced into the sequence. The second advantage is that the signal
is captured in real-time.
Though this technology is still developing, several prototypes exist such as nanopore DNA
sequencing, single-molecule real-time (RNA polymerase) sequencing, and the HeliScope
sequencer, which has pioneered an innovative methodology able to perform high-through-
put DNA sequencing, starting from a single-molecule. The HeliScope sequencer relies
on the principle of true single-molecule sequencing that does not require a preclonal
amplification step. The absence of an amplification step during sample preparation cir-
cumvents the problems of sequencing errors attributable to PCR artifacts, as stated above.
This system is capable of sequencing up to 28 Gb in a single sequencing run in about 8
days, but it can also generate short reads with a maximal length of 55 bases. The Helicos
platform was first used in 2008 to sequence the 6407-base genome of the bacteriophage
M13 and a year later to sequence an individual human genome (Figure 12.7). Currently
there is major competition in the sequencing arena and in 2015 not all of the original
companies are still trading. Some have been subsumed into other companies and some
have closed down completely. Consequently the reader needs to be mindful that the avail-
ability of information from websites changes over time. Nevertheless websites when open
offer a good source of information.
There are a number of companies with an interest in nanopore (a tiny hole) sequenc-
ing and nanopore technologies, amongst which Oxford Nanopore Technologies
(https://www.nanoporetech.com) have declared an interest in DNA sequencing and
other activities. The US government recently released a number of grants for the fur-
ther development and use of this technology with five new grants in September 2013
alone. Though there is great interest, there have been a few problems. The speed at
which DNA is tracked through the nanopore makes the process of sequencing espe-
cially tricky. However, it appears that altering the wavelength of light that the system
employs can effectively slow down the flow and this may solve the problem. The poten-
tial for this technology in DNA sequencing is great.

Figure 12.5: The Applied Biosystems SOLiD sequencing-by-ligation method. Fragments of DNA are ligated
with oligonucleotide adaptors at each end and hybridized to complementary oligonucleotides attached to
magnetic beads (a). Beads are then placed in a water/oil emulsion where DNA amplification is performed (b).
Once amplification is finished, the beads are placed on a glass surface and entered into the sequencer (c). In
the sequencer a universal sequencing primer, complementary to the adaptor sequence, is used to trigger the
sequencing reaction where ligation cycles with fluorescently labeled degenerate probes is performed. Once the
probe anneals to the DNA template, a DNA ligase covalently binds the probe to the sequencing primer and the
fluorescence is recorded. The probe is then cleaved and another cycle starts. After seven rounds of sequencing,
the extended universal primer is removed and a new universal primer is added that is offset by one base (d). The
sequence of the read is inferred by interpreting the ligation results for the 16 possible dinucleotide interrogation
probes. (Courtesy of Applied Biosystems.) (See also color plate.)

2967_Ch12.indd 345 08/07/2015 14:46

 346 CHAPTER 12 Sequencing Technology and the Future

(a)

A
A

T
+
T

T A
A T

Adaptor-modified DNA strand hybridized to

(b)
oligonucleotide anchor

Denature,
cleave

Cluster generated by Sequencing of forward

bridge amplification strands

(c)
T
POL C
POL A
G
G
T T T T
A C
G
Fluor A C
C
Incorporation cleavage G
G
C
Block
removal

Template
strand

Figure 12.6: The principles of the Illumina sequencing technique. DNA fragments are first ligated onto
oligonucleotide adaptors to form double strands (a). Adapter-modified, single-stranded DNA is then added to
a flow cell and immobilized by hybridization. Bridge amplification generates clonally amplified clusters (b). The
cluster fragments are denatured, annealed with a sequencing primer, and subjected to sequencing-by-synthesis
employing DNA polymerase and four reversible dye terminators. Once incorporated, the terminator stops
the sequencing reaction, which restarts immediately by cleavage of the incorporated dye terminator. Post-
incorporation fluorescence is recorded (c). (See also color plate.)

2967_Ch12.indd 346 16/07/2015 10:57

 The Future of NGS in Clinical Practice and Research 347

F F
F C
F C
C
C
5’
A G T
F C T C
C T
G
C
A F
A
C 1. Synthesize
Next C T C G
base C
T
G
A
G
T
C
A
A
T
T
A
T A T A T A
T A T A T A
T A T A T A
T A T A T A
T A T A T A
T A T A T A
T T T
T T T
5’

5’ 5’
A G T A G T
C T C C T C
T C A T C A
G A C F G A F C
C T C G C T C G
C G G C A T C G G C A T
T A T A T A T A T A T A
T A T A T A T A T A T A
T A
A
T
T
A
A
T A
A
4. Cleave T A T A T A 2. Wash
T T T A T A T A
T A T A T A T A T A T A
T A T A T A T A T A T A
T A T A T A T A T A T A
T T T T T T
T T T T T T
5’ 5’

5’
A G T
C T C
T C A
G A F C
C T C G
C G G C A T
T A T A T A
T A T A T A
T A T A T A
T A T A T A
T A T A T A
T A T A T A
T A T A T A 3. Image
T T T
T T T
5’

Figure 12.7: The principles of the Helicos sequencing technique. DNA fragments are captured by poly-T
oligomers linked to an array. During each sequencing cycle, a DNA polymerase and a single fluorescent labeled
nucleotide is added to the array. The array is imaged and the fluorescent label is then removed, and the cycle
repeated.

12.2 The Future of Ngs in Clinical Practice

and Research
In the recent past, the size and complexity of the human genome (3 billion base pairs)
was seen as a burden for every research group that aimed to sequence the genome.
Consequently, any proposal to use genome sequencing in clinical practice was considered
impossible; however, the idea of genomic sequencing has become a reality since the publi-
cation of the first draft of the Human Genome Map (HGM) in 2001. To generate the first
draft of the HGM took more than 10 years and cost several millions of dollars. Since then,
a number of developments, including the advent and feasibility of NGS, have not only
dramatically reduced the cost and the time required to sequence genomes, but massively
increased the scale of data output (Figure 12.8). We may therefore be approaching the
time, in the near future, when sequencing whole genomes of patients will be practical in
the clinical setting. In this section, we will briefly discuss some of the possible applications
of NGS technology in both research and clinical practice.

2967_Ch12.indd 347 08/07/2015 14:46

 348 CHAPTER 12 Sequencing Technology and the Future

Output per instrument run

2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
1014
1012
1010
Output (kbp)

108
106
104
102
0
Platforms
ABI 3730xl 454 GS 20 Solexa/illumina ABI SOLiD Roche/454 Illumina Ca Ilx, Illumina HiSeq
capillary pyrosequencer sequence sequencer Titanium, SOLiD 3.0 2000
sequencer analyzer Illumina GA ll

2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
1000
1000 Genomes, Watson Genomes pilot
Draft human Human Microbiome genome and HapMap3
genome HapMap Project begins ENCODE Project begins projects begin publication publications
ENCODE Project First tumor: normal Human genetics
pilot publications genome publication syndromes publications

Projects and publications

Figure 12.8: Changes in instrument capacity over the past decade and the timing of major sequencing
projects. The figure provides a perspective on a decade of advances in technologies applied to complex
disease and a time-line of the major research projects during the same period. The figure illustrates how
technology and science are linked. (From Mardis ER [2011] Nature 470:198–203. With permission from
Macmillan Publishers Ltd.)

Using NGS will enable high-resolution genotyping for SNPs in

complex disease
Genotyping and association studies have long been employed to catalog single nucleotide
variation across the genome and to associate genetic variation with phenotype. In recent
studies, high-density oligonucleotide arrays have been the predominant methodology for
SNP genotyping in complex disease; however, the identification of specific risk alleles
remains challenging and the results of both genome-wide linkage analysis (GWLA) and
GWAS have been mixed. Some GWLA and GWAS studies have been very successful, but
others have not been so productive.
One problem with GWLA and GWAS is that the early studies were conducted mainly
on genotyping arrays that, though high throughput, test for a limited number of selected
SNPs. In addition, the SNPs to be tested were selected to identify polymorphisms where
the least frequent variant was present in 5% or more of the population. This was done
to maintain a reasonable level of statistical power. As a consequence, in many common
diseases, such as cancer, and some clinically important quantitative traits, which may have
very complex genetic and epigenetic (in the broadest sense) architectures, this standard
genotyping format may not have been sufficient to identify the risk alleles or causative
variants. Relevant marker SNPs may have been absent from the original (standard) geno-
typing array. More recently, with the larger-scale studies, it has become possible to inves-
tigate less common SNPs, i.e. those were the least common variant is 1% or less, with a
reasonable degree of statistical confidence.

2967_Ch12.indd 348 08/07/2015 14:46

 The Future of NGS in Clinical Practice and Research 349

Beware the fly in the ointment

Though the selection of which SNPs to investigate can depend on the size of the study
population, selection is currently more likely to be determined by the commercial testing
platform used. Most commercial platforms offer a pre-set selection and selection is not a
free-for-all process based on the most interesting SNPs identified by the researchers.
NGS promises more
The next-generation high-throughput sequencing approach promises more. NGS tech-
nologies provide higher resolution genotyping designed to detect and characterize SNPs,
rare variants, and also mosaic mutations. The power of high-throughput sequencing to
identify unknown causative mutations in human disease has recently been demonstrated
in a family with a recessive form of Charcot–Marie–Tooth disease and in a family with
both Miller syndrome and primary ciliary dyskinesia. Whether these examples, which
are based on Mendelian (i.e. non-complex) disease, prove the potential for this type of
analysis in some complex diseases, where the risk alleles may be present in a small per-
centage of the disease group, is a matter of debate. If we consider a genetic variation that
is not associated with reproductive fitness, we cannot expect this to be present at high
levels in a population. Therefore, any such genetic variant under the old GWAS model
design, which generally selected tagged SNPs where the least common allele is found at
5% or more in the population, would be unlikely to be successful. Francis Collins suggests
this may be particularly important in relation to diseases involving mental health, such
as autism, bipolar disorder, and schizophrenia, where reproductive fitness may be low.
Genome sequencing is more likely to be useful in these cases. As genome sequencing is not
restricted by the frequency of a specific marker and it also includes variations that are not
SNPs, such as copy number variations (CNVs), which may also be important in determin-
ing disease and disease traits (e.g. response to commonly used pharmaceutical agents such
as codeine), it will be capable of identifying a greater range of risk alleles and sequences.
The key application of NGS technologies therefore is studying genetic variation between
individuals using whole-genome or targeted re-sequencing in order to discover new asso-
ciations with disease and to improve the resolution for the identification of previously
detected associations so that the precise risk allele can be identified. Such comprehensive
SNP identification will undoubtedly improve the predictive power of GWAS and GWLA,
and impact on our understanding of complex trait loci in complex disease, but it will not
necessarily solve all of our problems—only increasing sample size can do that.
The greater the number of markers tested or sections of the genome sequenced, the finer
the map will become. Fine mapping refers to the level of resolution being achieved in
a study. Fine mapping is not a single technique—it is an idea that can be achieved by
applying one or more techniques. Fine mapping has not been the first step in most stud-
ies, as stated in Chapter 4. Instead, it is usually performed to confirm and extend a study.
Often it involves focusing on a particular area or series of chromosomes rather than the
whole genome. However, with direct sequencing and the possibility to sequence the whole
genome, fine mapping may increasingly be the first step in a study, not a second or later
step. There is also a link between fine mapping and imputation. Imputation can be used to
find missing data based on known patterns of linkage disequilibrium within haplotypes.
Imputation analysis is discussed in further detail below.
NGS can also be applied for the discovery of novel somatic mutations in cancer, leading
to a better understanding of the molecular pathogenesis of cancer. This will undoubtedly
result in novel diagnostic and therapeutic approaches.

2967_Ch12.indd 349 08/07/2015 14:46

 350 CHAPTER 12 Sequencing Technology and the Future

Using NGS will enable better identification of CNVs

CNVs include deletions, duplications, inversions, and rearrangements of pieces of
the genome. These range from small insertions and deletions to very large cytogenetic
changes that involve entire chromosomes. CNV has been associated with a number of
genetically complex diseases, for example Alzheimer’s disease and schizophrenia. Many
of these findings are based on the use of array technologies and comparative genomic
hybridization, which though powerful for the detection of large CNVs (with a resolution
of approximately 1 kb) is unable to detect small balanced structural variations such as
inversions.
High-throughput sequencing technology is able to detect any kind of CNV including
balanced and unbalanced variation by employing a strategy called paired-end mapping.
Using this approach, a single DNA fragment is sequenced from each end and thus the
same fragment produces two sequencing reads that could overlap at their extreme ends
provided they are long enough. Paired-end sequencing is emerging as a key technique for
assessing genome rearrangements and structural variation on a genome-wide scale. This
technique is particularly useful for detecting copy-neutral rearrangements, such as inver-
sions and translocations, which are common in cancer where novel fusion genes can also
be produced. Paired-end mapping was demonstrated by Korbel et al. in 2007 as a means
of detecting deletions, inversions, and insertions with an average resolution of 644 bp, and
the clinical potential for this technique has been demonstrated in both lung and breast
cancer. However, sequencing-based approaches to detect CNVs are at the moment still
too expensive and difficult for routine clinical diagnostics, but lower costs and the allure
of more rapid results in future may enable sequencing techniques to replace array-based
genomic hybridization in the clinical laboratory.

Sequencing the RNA transcript and the whole transcriptome is an

alternative way forward
High-throughput sequencing technology has also been adapted to enable the study of
DNA transcripts. This RNA sequencing technique is called whole-transcriptome shotgun
sequencing (WTSS), which aims at mapping and quantifying DNA transcripts in bio-
logical samples. This approach was first applied to report the human transcriptome from a
human embryonic kidney and a B cell line, and later to detect gene fusions in cancer from
chronic myelogenous leukemia cells. The goal of transcriptome sequencing is to sequence all
transcribed genes.
Transcriptome sequencing is similar to exome sequencing (see Section 12.3) as it does
not detect mutations in non-coding regions of the genome, but it has the main advantage
of obtaining quantitative information about gene expression levels as well as detecting
changes in gene expression (e.g. alternative splicing) and fusion transcripts produced by
chromosomal rearrangements. One limitation of transcriptome sequencing is that it is
biased toward abundantly expressed transcripts and therefore the detection of sequences
or genes expressed at lower levels can be low or absent.

12.3 WHOLE-GENOME VERSUS EXOME SEQUENCING

As noted above, whole-genome sequencing may soon be available in all clinical labo-
ratories where it may be used as a standard genetic test. However, in order to answer a
specific clinical question, targeted analysis of specific genomic regions may be preferable

2967_Ch12.indd 350 08/07/2015 14:46

 The Next Generations of Genome/Exome-Wide Association Studies 351

Construct
shotgun library Hybridization

Genomic DNA Fragments

Wash

Pulldown

AGGTCGTTACGTACGCTAC
GACCTACATCAGTACATAG
GCATGACAAAGCTAGGTGT

Mapping, alignment,
variant calling DNA sequencing Captured DNA

Figure 12.9: Diagram showing workflow for whole-exome sequencing. The figure shows the sequence of
events in whole-genome sequencing. The DNA is broken up into fragments that are hybridized, washed, and
captured on beads before undergoing sequencing. (From Bamshad MJ, Ng SB, Bigham AW et al. [2011] Nat Rev
Genet 12:745–755. With permission from Macmillan Publishers Ltd.)

to sequencing the whole genome. When applied to the whole genome, restricting analysis
to the protein-coding regions, that is exome sequencing, is more cost-effective compared
with sequencing the entire genome. Whole-exome sequencing has been applied in the
identification of germ-line mutations underlying some Mendelian disorders, some somatic
mutations in a variety of different cancers, and de novo mutations in some neuro-devel-
opmental disorders. In the future the greatest application may be in cancer therapy where
it may be useful to screen genes encoding protein targets before applying chemotherapy.
These so-called enrichment strategies that target specific genomic regions will save time
and money, and because smaller data bundles are formed than in whole genome analysis,
which includes non-coding regions, whole-exome analysis can be performed more rapidly.
Use of exome as opposed to genome sequencing may speed up the transfer and application
of sequencing to clinical practice.
Restricting analysis to the exome does introduce some limitations. This analysis does not
look at mutations and polymorphisms within non-coding regions, and it is not able to
detect most structural variants, such as chromosomal translocations happening within
intronic break points. In contrast, whole-genome sequencing will identify all of the varia-
tion in the genome not just that found in protein coding regions (Figure 12.9).

12.4 The Next Generations of Genome/Exome-

Wide Association Studies
The same basic principles apply to all investigations whether a study design is based on an
investigation of the whole genome or the whole exome. The limitations and the ultimate
goals have not changed, but the methods used have.

2967_Ch12.indd 351 08/07/2015 14:46

 352 CHAPTER 12 Sequencing Technology and the Future

Missing and non-genotyped SNP data can be imputed using large

databases and known patterns of linkage disequilibrium
Genotype imputation is a technique increasingly used in GWAS that allows geneticists
to evaluate the evidence for associations with genetic markers that are not directly geno-
typed. There are two uses for this method: to fill in missing data and to assign genotypes
for markers not typed in the study in order to increase the size and range of the study in
terms of the number of potential risk alleles investigated. For whatever reason imputation
is applied, the method is based on linkage disequilibrium. In statistical analysis, imputa-
tion is a method used for handling missing data (Figure 12.10). Missing data are simply

(b)

(a) (c)
–log10P

–log10P

1 2 3 4 5 6

After imputation

Figure 12.10: Imputation analysis. The two graphs illustrate a genotype sequence before (on the left) and
after (on the right) impute has been applied. The central section of the figure illustrates in detail six different
haplotypes numbered 1 to 6 for a combination of sixteen bi-allelic genes illustrated as double rings. Each gene
is polymorphic, eight of the genes have been successfully genotyped. There are two alleles for all of genes. The
alleles are illustrated by gray shading (A allele) or no shading (white—B allele) in those genes that have been
successfully genotyped. The eight genes illustrated in black have not been genotyped. Imputation analysis can
be applied to enable the genotypes of the black genes on this haplotype to be estimated based on the known
pattern of linkage disequilibrium for the surrounding genes which have been successfully genotyped. To do this,
data are extracted from large genotype databases to fill in the “missing” genotypes. For example if we know that
persons with haplotype 1 which runs top to bottom; white, unknown, unknown, unknown, white, unknown, gray,
unknown, white, gray, gray, unknown, unknown gray, unknown, white (or B-?-?-?-B-?-A-?-B-A-A-?-?-A-?-B)
carries the white (B) alleles at positions 1, 5, 9, and 16 and the gray (A) alleles at position 7, 10, 11, and 14 in the
sequence, then the full sequence for this haplotype can be assigned. Using the known genotypes from positions 1,
5, 7, 9, 10, 11, 14, and 16 we can identify the unknown genotypes for the other eight genes on the remaining five
haplotypes (haplotypes 2 to 6). This imputed data can be included in the analysis. Because linkage disequilibrium
is so strong in certain areas of the human genome, imputation analysis from initial GWAS data can massively
extend the studies. Thus an initial GWAS study of based 300,000 to 400,000 SNPs can generate data for up to one
million or more markers. (From Marchini J & Howie B [2010] Nat Rev Genet 11:499–511. With permission from
Macmillan Publishers Ltd.) (See also color plate.)

2967_Ch12.indd 352 08/07/2015 14:46

 The Next Generations of Genome/Exome-Wide Association Studies 353

substituted with predicted values based on the known levels of linkage disequilibrium.
When all missing values have been imputed, the resulting data set can then be analyzed
using standard techniques. The sort of databases used for this exercise include those pro-
duced by the 1000 Genomes Project and HapMap.

GWAS identifies both synthetic (false) associations and direct (real)

associations
To date, GWAS have identified hundreds of alleles at gene loci that may be involved in
determining susceptibility and resistance to common diseases. Most early GWAS restricted
analysis to common variants where the minor allele frequency (MAF) was greater than
5%. This is discussed above.
However, very few GWAS have identified causal variants and this has led many geneticists
to assume that the causal variants lie elsewhere in the genome, but in close proximity
to the identified associated alleles and their markers. In general, many feel that with a
few exceptions the genetic polymorphisms identified so far are unlikely to explain the
majority of phenotypic variation in disease susceptibility. If susceptibility is determined
by multiple rare or low-frequency variants, then these may not be identified in a GWAS
even with lower statistical thresholds. Consequently, some, perhaps even many, of the cur-
rent GWAS signals could reflect linkage with multiple rare variants. In this case, associa-
tions will be wrongly credited to a common allele, while in reality the effect that is seen
relates to inheritance of other alleles with which they are in linkage disequilibrium. These
associations with common markers resulting from multiple (unobserved) low-frequency
causal variants are referred to as synthetic associations—a form of indirect association.
Thus, in a typical GWAS with large sample sizes, rare variants may create multiple signifi-
cant associations, but some may be synthetic. The impact of synthetic associations can be
dramatically different from that which occurs when the causal variant is identified. For
example, the causal variants can be megabases away from the SNPs found to be significant
in GWAS and the real risk can be several-fold greater than that for a synthetic association.
More recent studies have suggested that the contribution of rare variants is likely to be
small and thus the importance of synthetic associations generated by multiple variants has
perhaps been overstated. It is not yet known how many GWAS signals are due to synthetic
associations; perhaps we will only know the answer when all of the associations have been
tested in functional studies and a link between genotype and phenotype established. In the
meantime, we should question some of the inferences from GWAS and consider the use of
refined approaches for follow-up studies. The possibility of false positives (synthetic asso-
ciations) provides strong support for the growing need to follow-up GWAS and GWLA
studies with sequencing. In cases where the sequencing is restricted to a specific chromo-
somal region, the study design should consider expanding the region of interest beyond
the target block. These follow-up studies will employ faster and better NGS technologies.

The importance of linking genotype to phenotype

When associated marker SNPs are located a long distance away from the nearest gene, it
is often assumed that a regulatory variant with a small effect is the cause of the associa-
tion. If we consider the case of SNPs on chromosome 8q24 that have been found to be
associated with colorectal cancer, these SNPs map within range of the MYC oncogene
and it is thought that the associated SNPs may play a role in influencing MYC function.
However, these SNPs are located hundreds of kilobases away from the MYC oncogene,
and no connection between the associated SNPs and the expression of this gene has yet
been demonstrated.

2967_Ch12.indd 353 08/07/2015 14:46

 354 CHAPTER 12 Sequencing Technology and the Future

Genotyping on new arrays provides a focus and higher level of

resolution for GWAS
As stated above, the basic criterion used in the first phase of GWAS was to catalog com-
mon variants (5% or greater) and genotype them either directly or indirectly (through
linkage disequilibrium) in cases and controls, and identify associations with particular
traits and diseases. However, if the frequency of a SNP contributing to a phenotype is less
common in the normal population, for example in the range of 1–5%, then we need to
broaden our SNP selection criteria and test for less common variants, otherwise significant
associations may be missed. This is one of many reasons for the 1000 Genomes Project,
which is extending the catalog of known human variants down to a frequency near to 1%.
With respect to the first phase of GWAS that scanned primarily for common variants
in complex disease, it seems reasonable in 2015 to suggest that even if we have not yet
reached the end of the first phase, we have at least reached the end of the beginning. A new
wave of higher-resolution GWAS is now being applied that will investigate less common
variants. At present, this new wave of GWAS is based on one of two approaches: GWAS
on new genotyping arrays and GWAS on NGS technologies.
Companies such as Illumina and Affymetrix have already designed chips containing both
common and low-frequency variants discovered in the 1000 Genomes Project. The new
analysis will identify millions of genotypes on these new chips and it is likely that they will
identify new associations. However, there is some concern that the causative variants may
be very rare variants. There are questions over how such data can be used. Though not
useful in disease diagnosis, identifying rare associations with complex disease does help us
to understand disease pathology, and also has the potential to help in patient management
and therapy. Thus, this is not wasted time, even though the diagnostic use of this data
may be limited. However, as stated above to accurately identify association signals for rare
variants we will need even larger, sample pools compared with previous studies in order to
achieve statistical significance at an appropriate significance threshold.
Sample size is dictated by the expected size of the effect, i.e. the expected genetic impact
of the allele set as an expected risk ratio [either odds ratio (OR) or relative risk (RR)]. For
rare alleles with large risks, sample sizes can be smaller. An example of this is seen with the
32-bp deletion in the C-chemokine receptor 5 gene (CCR5-Δ32), where approximately
1% of Europeans are homozygous for the mutation and as many as 10% may be heterozy-
gous. The homozygous individuals are virtually immune to early HIV-1 infection. As this
mutation has a large effect in HIV-1/AIDS, identifying this association with a reasonable
level of statistical confidence does not require large numbers.

Different forms of NGS technology will impact on how GWAS is used

NGS will be used to conduct whole-genome and whole-exome sequencing of individuals
rather than genotyping and cataloging specific variants. Such studies will be performed
in a manner similar to current GWAS sequencing, with both cases and healthy controls
being studied, and similar statistical analysis for associations applied.
Whole-exome sequencing
Whole-exome sequencing has already been used to identify causal alleles for several
Mendelian disorders, such as those shown in Table 12.2. These studies looked for rare
alleles or novel alleles that are more common in affected individuals compared with unaf-
fected individuals (controls). Using public databases, this method could be used to fil-
ter out potential causative variants by assuming that any allele found in both cases and

2967_Ch12.indd 354 08/07/2015 14:46

 The Next Generations of Genome/Exome-Wide Association Studies 355

Table 12.2 Some Mendelian and a few complex diseases identified up to 2011 via exome
sequencing.

PubMed Mode of
Disorder ID inheritance Na Strategy Gene(s)
Comparison of unrelated cases
Kabuki 20711175 AD 10 10 cases/10 kindred MLL2
Schinzel–Giedion 20436468 AD  4 4 cases/4 kindred SETBP1
Fowler 20518025 AR  2 2 cases/2 kindred FLVCR2
Sensenbrenner 20817137 AR  2 2 cases/2 kindred WDR35
Comparison of related cases
Miller 19915526 AR  4 4 cases/3 kindred DHODH
Retinitis pigmentosa 21295283 AR  3 3 cases/1 kindred DHDDS
Spinocerebellar ataxia 21106500 AD  4 Linkage + 4 cases/1 TGM6
kindred
Primary failure tooth eruption 21404329 AD  4 Linkage + 4 cases/1 PTH1R
kindred
TARP (talipes equinovarus, 20451169 XLR  2 Linkage + 2 cases/2 RBM10
atrial septal defect, Robin kindred
sequence, and persistent left
superior vena cava)
X-linked 21415082 XLR  2 Linkage + 1 case/1 MCT8
leukoencephalopathy kindred
Homozygosity mapping
Autoimmune 21109225 AR  1 1 case/1 kindred FADD
lymphoproliferative syndrome
Complex I deficiency 21057504 AR  1 1 case/1 kindred ACAD9
Non-syndromic mental 21212097 AR  2 2 obligate carrier TECR
retardation parents
Identification of de novo mutations
Sporadic mental retardation 21076407 Complex 30 10 parent–child Multiple
trios
Autism Complex 60 20 parent–child trios Multiple
The table lists a number of diseases, the traits associated with them, and the genes thought to carry the
responsible mutations.
Number of exome-sequenced individuals.
a 

AD, autosomal dominant; AR, autosomal recessive; XLR, X-linked recessive.

MLL2, myeloid/lymphoid or mixed-lineage leukemia 2; SETBP1, SET binding protein 1; FLVCR2, feline leukemia
virus subgroup C cellular receptor family, member 2; WDR35, WD repeat domain 35; DHODH, dihydroorotate
dehydrogenase; DHDDS, dehydrodolichyl diphosphate synthase; TGM6, transglutaminase 6; PTH1R, parathyroid
hormone 1 receptor; RBM10, RNA binding motif protein 10; MCT8, monocarboxylate transporter 8;
FADD, FAS-associated death domain (also called MORT1); ACAD9, acyl-CoA dehydrogenase family, member 9;
TECR, trans-2,3-enoyl-CoA reductase.
The numbers in the studies are small and therefore the statistical inference is not shown. These are conditions
from rare kindred and mostly they are Mendelian autosomal diseases; however, there are two complex diseases
included in the list.
From Bamshad MJ, Ng SB, Bigham AW et al. (2011) Nat Rev Genet 12:745–755. With permission from Macmillan
Publishers Ltd.

2967_Ch12.indd 355 08/07/2015 14:46

 356 CHAPTER 12 Sequencing Technology and the Future

healthy controls cannot be causative. Though this strategy can be exceptionally powerful
for rare Mendelian disorders, it is less useful in complex disease because the assumption
that a causative allele must be absent from the healthy population does not work in com-
plex disease, where many potentially causative alleles have been found to be expressed in
both diseased and healthy individuals. In addition, not all cases are positive for the caus-
ative allele in complex diseases.
Exome sequencing reduces the number of candidate genes tested. However, as with GWAS,
these studies still require large sample pools to provide strong statistical evidence for asso-
ciation and even though costs are currently falling, they are still too high. Therefore, two
different strategies have been proposed to reduce these costs: family-based sequencing and
basic extreme-trait design.
Family-based sequencing
Family-based studies are impractical for most complex diseases. Even when familial
cases occur we cannot assume that these rare cases represent the whole disease popula-
tion. Therefore, making assumptions based on testing families in complex disease is a rare
option. However, if families are tested, the strategy is to sequence the index case (i.e. the
first identified case) and any co-affected family member(s) to identify overlapping vari-
ants. The closer the family cases are to the index case, then the greater the number of can-
didate genes that will need to be tested as closer family members will share more genes. In
contrast, the more distant related cases are compared with the index case, then the lower
the level of genetic similarity is and therefore a smaller number of candidate alleles will
need to be genotyped.
Basic extreme-trait design
Basic extreme-trait design offers more potential. In this situation cases are carefully
selected from one or both ends of a phenotype distribution and then sequenced. This
strategy is based on the studies of Cohen et al., who in 2004 demonstrated the effec-
tiveness of sequencing candidate genes at the extreme ends of a phenotype distribution
to find rare alleles involved in the risk for a complex trait. An obvious example of this
extreme phenotype study is seen in individuals who are highly exposed to HIV-1, but
remain uninfected. As there is survival advantage in those who carry genetic variants that
contribute to the exposed/uninfected trait, these alleles will be enriched in the popula-
tion. This means that studies of small sample size can be performed with a higher degree
of confidence than if the alleles were less common. Identified candidate alleles can then
be genotyped for confirmation in much larger samples. In HIV-1, where genetic variants
contribute to protection against HIV-1 infection and HIV-1/AIDS, the high impact of
the protective variants/alleles means that whole-exome (or whole-genome) sequences of
a few individuals may be powerful enough to identify candidate variants/alleles associ-
ated with the disease trait. Once identified, the genotype of the identified candidate
can be re-investigated in a much larger group to confirm or refute the findings of the
sequencing report.
The use of NGS technology applied to the new wave of GWAS has the potential to dis-
cover the entire spectrum of sequence variations in a sample of well-phenotyped indi-
viduals. However, the use of NGS platforms in GWAS studies has raised some concerns.
The error rate of the NGS platforms is higher than the Sanger sequencing methods and,
because not all errors are random, they may obscure true associations or generate false-
positive associations if they are frequent enough. In addition, NGS technologies can also
produce data pools with high missing rates. In practice, however, high missing rates are
less of a problem because imputation methods may be used to recover (or back-fill) the

2967_Ch12.indd 356 08/07/2015 14:46

 Metagenomics and the Bacterial Genome 357

missing data. Authors using imputation should make sure that any data generated by this
means are clearly acknowledged in their publications.

12.5 EPIGENETICS: A COMPLIMENTARY STRATEGY IN

COMPLEX DISEASE STUDIES
Epigenetics is based on the word “epi-” meaning “above”, “outside”, or “in addition to” genet-
ics and is usually used to refer to heritable effects not produced by changes in the DNA
sequence. The term epigenetics is applied in a number of different research strands all
around genetics. It can have several meanings, as discussed earlier in this book. However,
the main one which concerns us is that where it refers to reversible biochemical modifica-
tion of DNA that does not involve any change in the nucleotide sequence. These modifica-
tions may persist through cell divisions and represent memories or molecular signatures
over multiple cellular generations. Examples of such modifications are DNA methylation
and histone modification, both of which serve to regulate gene expression without alter-
ing the underlying DNA sequence. Two alleles with the same sequence may have different
states of methylation and this may result in different phenotypes even though the sequence
is the same. Though these changes do not alter the DNA sequence, they may have major
effects on the expression of the gene. Some of these changes are heritable, though they do
not affect the DNA structure. Methylation establishes epigenetic inheritance as long as
methylase acts to restore the methylated state after each cycle of replication. Thus, a meth-
ylated state can be perpetuated through an indefinite series of somatic meiosis.
Determining how proteins interact with DNA to regulate gene expression is essential
for the full understanding of many biological processes and disease states. A prominent
genomics approach for detecting biochemical modifications to DNA is the chromatin
immunoprecipitation assay (ChIP-chip). This technique has been used primarily to
detect locations where a protein of interest is bound to DNA in vivo (Figure 12.11).
ChIP sequencing has been used to characterize histone and transcription factor binding sites
in human CD4+ T cells, in a cervical carcinoma cell line, and also in pluripotent murine
stem cells undergoing differentiation. While high-throughput sequencing has improved
our ability to detect and characterize DNA–protein interactions, further work will be
required to determine how these biological changes result in a clinical disease phenotype.

12.6 Metagenomics and the Bacterial Genome

Laboratory culture-based approaches have been traditionally used for characterizing microbe
genomes. However, microbes are difficult to culture in the laboratory because they normally
live in mixed communities, and different species interact both with each other and with their
habitats, including the host organism. Metagenomics describes genomics on a huge scale
and it is particularly important in relation to the study of microbial communities because it
can be used on mixed samples and does not require pure cultures for genotyping.
Metagenomics is based on the genomic analysis of microbial DNA directly extracted from
microbial communities present in samples such as soil, water, or feces. This technology
has had a tremendous impact on the study of microbial diversity in environmental and
clinical samples, and may lead to major advances in medicine, agriculture, and also energy

2967_Ch12.indd 357 08/07/2015 14:46

 358 CHAPTER 12 Sequencing Technology and the Future

Nucleus

Cross-link and
fractionate
chromatin

ChIP:
Enriched DNA
binding sites

Sequence

Figure 12.11: Illumina ChIP sequence experiment

Binding site
workflow. DNA fragments are first cross-linked in situ,
mapping
fractionated, and then immunoprecipitated. DNA is then
sequenced using an Illumina platform to identify genome-wide
sites associated with proteins of interest. (Courtesy of Illumina.)
(See also color plate.)

production. In a typical metagenomic study, the genomic content of all the microbes
in a sample is sequenced using NGS technology. This results in millions of sequences of
DNA fragments. Computational methods are then employed to predict the taxonomic
affiliation of these DNA sequence fragments, which are then compared against a ref-
erence database such as those of the National Center for Biotechnology Information
(http://www.ncbi.nlm.nih.gov) where the sequences of microorganisms are collected.
Sequences are then affiliated to the corresponding microorganism and compared with those
of others to identify the level of biodiversity present in the sample. This approach has been
used to identify novel viral pathogens, detect viral drug resistance mutations, and diagnose
bacterial infections. However, given the relatively high cost of high-throughput sequencing,
these techniques are unlikely to replace traditional microbiological techniques for routine
pathogen identification in the immediate future, but it may not be long before metage-
nomic screening is standard. Metagenomic technology is at present the main approach
employed by the Human Microbiome Project (http://commonfund.nih.gov/hmp/)—an

2967_Ch12.indd 358 08/07/2015 14:46

 Metagenomics and the Bacterial Genome 359

initiative launched in 2008 that aims to characterize the microbial communities associated
with both health and disease in different parts of the human body, such as nasal passages,
oral cavities, skin, and gastrointestinal and urogenital tracts.

To put metagenomics into context, we need to consider the

impact it may have
Tim Spector, in his book Identically Different: Why You Can Change Your Genes (2012),
discusses the impact of bacterial genes on human health. He states “... within us all there
lies a parallel universe of warring personalised bacterial colonies, each of them with genes
and proteins with a possible role in our health of which we remain largely ignorant.” He
continues “We are only just scratching the surface of this hidden universe ...” and suggests
“... the type of bacteria and extra genes we have in our guts could explain around 50 per
cent of obesity, compared with a paltry 1–2 per cent explained by the common human
genes we have discovered so far.” The evidence for this hypothesis is, as the author sug-
gests, currently limited, but it is a subject that is worthy of serious investigation. Bacteria
react closely with the gut wall, and modifications in the permeability of the gut wall and
host immune tolerance to certain bacteria have been shown to be particularly important
in common diseases of the bowel (Crohn’s disease and ulcerative colitis, in particular), but
also in some cancers (gastric and colon cancer). This interaction between host and bacteria
illustrates how we cannot consider the human genome in isolation. Not only is it essen-
tial to consider host systems and how they interact (including epigenetics), but we must
also consider the environment and how environmental factors interact (Figure 12.12).
This idea is not new. However, there has been a tendency to consider invading bacteria as
passive, almost inactive, in the context of predisposition to and protection from disease.
Of course pathogens can cause disease, but not all bacteria are malignant—some may
be beneficial. It is clear from the findings reported above that bacteria are not simply
the passive targets for the host immune response nor are they simply acting as malicious

Genetic
variation

Lifestyle Diet

Disease

Epigenetics Bacteria

Figure 12.12: Genetic and non-genetic factors impacting on disease risk. The figure illustrates the idea
that multiple factors influence susceptibility to complex disease and not all are genetic. The factors interact in
complex ways. In order to understand the genetics of complex disease we need to apply a holistic approach that
takes account of these other factors.

2967_Ch12.indd 359 08/07/2015 14:46

 360 CHAPTER 12 Sequencing Technology and the Future

Lactobacillus in Increased GABA

yogurt neurotransmitter

Happy mouse

No Lactobacillus in No increase in
yogurt GABA
neurotransmitter

Figure 12.13: Lactobacillus-rich yogurt and mice. The figure

Unhappy mouse shows the effect of nutritional factors on biosystems and reminds
us that our genes are influenced by our environment. In this
example, mice fed a diet with Lactobacillus appear to be happier
and have higher measurable levels of GABA neurotransmitters
than those not on the same diet.

pathogens. The close interaction suggests there may be a level of symbiosis. Once again,
this is not a new concept. For example: millions of people consume food supplemented
with Lactobacillus which is thought to be beneficial in maintaining homeostasis between
the host and the bacterial flora in the gut. Looking at this from a genetic perspective pro-
vides a different focus. Simply altering the diet of mice by giving Lactobacillus-enriched
yogurt can modify the levels of expression of the γ-aminobutyric acid (GABA) neurotrans-
mitter receptor. The mice with the yogurt supplement appear to become happier mice
(Figure  12.13). The Lactobacillus must gain something from this relationship, which
appears to be a true symbiotic relationship. The balance of such relationships will deter-
mine whether a bacterium is beneficial to the host or not and may lead to some traits being
promoted, while others may be restrained. Once again we are talking about susceptibility
and resistance, i.e. risk—the one word that defines complex disease.

12.7 Major Ongoing International Genome

Projects
As discussed throughout this book, the worldwide web provides a wealth of resources for
the student of complex disease genetics. The databases developed from major international
research projects, such as the Human Genome Mapping Project, the Haplotype Map
(HapMap), the SNP Map, the 1000 Genomes Project, and ENCODE, are especially
useful. These are not the only sources that have been used, but all of them have impor-
tant functions for the future. These have all been discussed elsewhere in the book and
therefore in this section we will only consider three recently complete or ongoing projects:
HapMap, 1000 Genomes Project, and ENCODE.

HapMap is a project with major significance in current research,

especially GWAS
HapMap was launched in 2002 by a consortium of academic centers and private com-
panies of six nations (Canada, China, Japan, Nigeria, the UK, and the USA) aiming to

2967_Ch12.indd 360 08/07/2015 14:46

 Major Ongoing International Genome Projects 361

pinpoint the genes behind common diseases. Phase 1 typed 270 individuals from four
geographically diverse populations (Table 12.3), identified approximately 1.3 million
SNPs, and was published in 2005 (http://www.hapmap.org). Phase 2 of the project aimed
to increase the resolution of the haplotype map identified in phase I by adding a further
2.1 million SNPs to the test pool.
The third phase of the HapMap project (HapMap3) published findings in 2010 based
on 1.6 million common SNP genotypes in 1184 individuals from 11 different popula-
tions (Table 12.3). In addition, 10 regions of 100 kb were sequenced in 692 genomes.
HapMap3 includes both SNPs and copy number polymorphisms, and pinpoints pop-
ulation-specific differences for low-frequency variants. The idea behind phase 3 was to
produce high-resolution haplotype maps (including both the most common and some of
the less common variations from different populations) aimed at helping us to examine
patterns of linkage disequilibrium within different populations. Haplotypes are an essen-
tial piece in the jigsaw for imputation analysis and the design of association studies.
The map produced by the HapMap project describes common patterns of human genetic
variation and thus it has been extremely valuable in the design of association studies, accel-
erating the search for genetic factors in human disease. The value of the HapMap should
not be underestimated and it is likely to guide research for years to come. With approxi-
mately 38 million SNPs existing in the human population (in a biallelic locus there are in
principle 2n haplotypes, where n is the number of SNPs), there is a tremendous potential
for diversity; however, fewer haplotypes than expected are observed in practice due to the
extensive linkage disequilibrium across the genome. This fact has an important conse-
quence in association studies whereby if a causal variant is not directly tested it could still
be identified indirectly. The haplotype map also confirms the idea that the human genome
contains recombination hot spots. This means that linkage disequilibrium is not continu-
ous (or equal) across the genome, but there are regions where the recombination rate is

Table 12.3 Populations genotyped in HapMap phases 1–3.

Phase
Population Code 1 2 3
Utah residents with ancestry from Northern and CEU   
Western Europe
Yoruba in Ibadan, Nigeria YRI   
Japanese in Tokyo, Japan JPT   
Han Chinese in Beijing, China CHB   
African ancestry in South-Western USA ASW 
Chinese in metropolitan Denver, CO, USA CHD 
Gujarati Indians in Houston, TX, USA GIH 
Luhya in Webuye, Kenya LWK 
Maasai in Kinyawa, Kenya MKK 
Mexican ancestry in Los Angeles, CA, USA MXL 
Tuscans in Italy TSI 

2967_Ch12.indd 361 08/07/2015 14:46

 362 CHAPTER 12 Sequencing Technology and the Future

higher than expected and others where it is lower than expected (cold spots). The human
major histocompatibility complex (MHC) is a noted area for cold spots with higher than
expected levels of conservation of haplotypes.
All of the data produced by the HapMap project are freely available for unrestricted public
use at the HapMap website (http://www.hapmap.org). This site offers bulk downloads of the
data set, as well as interactive data browsing and analysis tools for data mining and visualiza-
tion. In 2015, the HapMap is a key resource for researchers designing any genetic association
studies aimed at identifying genetic variants associated with common disease or investigating
responses to common therapeutic drugs as well as other environmental factors.

The 1000 Genomes Project has major potential in studies

of complex disease
The 1000 Genomes Project is another challenging project aimed at producing an extremely
detailed catalog of human DNA variation that can be used in future studies. Specifically,
the goal of the project is to employ high-throughput sequencing technologies to char-
acterize over 95% of human genomic variants with allele frequency of 1% or higher in
each of five major population groups (populations in, or with ancestry from, Europe,
East Asia, South Asia, West Africa, and the Americas) using a sample of at least 1000
individuals (Figure 12.14). The project was launched in January 2008 by a collaborative
agreement of three world-leading genome Institutes: the Beijing Genomics Institute in
Beijing (China), the Wellcome Trust Sanger Institute (Cambridge, UK), and the National
Human Genome Research Institute (Bethesda, MD, USA). Early phase 1 and phase 2
studies have now been published, and the study is now in phase 3, aimed at extending the
pool to 1500 and eventually 2500.
In November 2012, whole-genome sequences for 1092 individuals from 14 populations
were published in Nature. The results provided a validated map of 38 million SNPs,
1.4  million insertion/deletions (indels), and 14,000 larger deletions. In addition, the
results showed that there is considerable variation in genetic structure between popula-
tions and this can best be seen by looking at the low-frequency variants. Individuals from
different populations carry different profiles of both rare and common variants. The num-
ber of the rare alleles within each individual genome varies substantially and many of these
are rare non-coding variants localized at conserved sites, such as motif-disrupting changes
in transcription factor binding sites.
The 1000 Genomes Project is not yet complete, but the impact of the project on medical
genetics is expected to be massive. Data from the studies so far are available at the 1000
Genomes public database (http://www.1000genomes.org), and can be freely used for
comparisons to screen variants in exome data from individuals with genetic disorders and/
or to screen cancer genomes for mutations that might suggest new therapeutic approaches,
and/or to speed diagnoses of diseases. Another major use of the 1000 Genomes Project is
imputing genotypes in existing GWAS; the large data set provided by the 1000 Genomes
Project will allow more accurate imputation of variants in GWAS and thus better localiza-
tion of disease-associated polymorphisms. The information contained within the 1000
Genomes database will be employed by worldwide researchers to analyze and interpret
DNA variations between people in order to understand which SNPs are implicated with
a disease. In addition, the 1000 Genomes Project has stored cell samples from all of the
people sequenced and this will allow future functional studies to determine the effects that
DNA variations may have on a phenotype. The 1000 Genomes Project represents a step
toward a complete description of all human DNA polymorphism and though the project

2967_Ch12.indd 362 08/07/2015 14:46

 Major Ongoing International Genome Projects 363

EUROPE
CEU IBS GBR FIN TSI
A. 85 A. 14 A. 89 A. 93 A. 98
B. All LCL B. All LCL B. All LCL B. All LCL B. All LCL
C. 45m/40f C. 7m/7f C. 41m/48f C. 35m/58f C. 50m/48f
D. 78t/3d/4s D. 14t D. 3d/86s D. 93s D. 98s

Finland
AMERICAS Great Britain
EAST ASIA
Utah,
MXL JPT
A. 66 USA Beijing, China A. 89
Southwest, Italy
B. All LCL Los Angeles, Spain Tokyo, B. All LCL
C. 31m/35f USA Hu Nan and Fu Jian C. 50m/39f
USA Japan
D. 59t/3d/4s Provinces, China D. 89s

PUR Puerto Rico CHB

A. 55 Ibadan, A. 97
B. 35 blood/20 LCL Nigeria B. All LCL
C. 28m/27f Webuye, C. 44m/53f
D. 47t/8d Medellín, D. 97s
Colombia Kenya
CLM CHS
A. 60 A. 100
B. All LCL B. All LCL
C. 29m/31f C. 50m/50f
D. 55t/5d D. 100t
ASW YRI LWK
A. 61 A. 88 A. 97
B. All LCL B. All LCL B. All LCL
C. 24m/37f C. 43m/45f C. 48m/49f
New 1000 Genomes D. 28t/22d/11s D. 65t/21d/2s D. 4d/93s

HapMap 3 AFRICA

Figure 12.14: Populations collected within the HapMap and 1000 Genomes Projects. Populations collected
as part of the HapMap Project are shown in blue and for the 1000 Genomes Project in green. The populations
involved include European: IBS (Iberian Populations in Spain), GBR (British from England and Scotland), CEU
(Utah residents with ancestry from Northern and Western Europe), FIN (Finnish in Finland), TSI (Tuscans in
Italy); East Asian: JPT ( Japanese from Tokyo), CHB (Han Chinese in Beijing, China), CHS (Han Chinese South);
African: ASW (African ancestry in Southwestern USA), YRI (Yoruba in Ibadan, Nigeria), LWK (Luhya in Webuye,
Kenya); and Americans: MXL (Mexican ancestry in Los Angeles, CA, USA), PUR (Puerto Ricans in Puerto Rico),
CLM (Colombians in Medellín, Colombia). (A) Total number of samples sequenced. (B) Source of DNA [blood
or lymphoblastoid cell lines (LCL)]. (C) Gender composition (male/female). (D) Number that are part of mother–
father–child trios (t), parent–child duos (d), or singletons (s); for trios and duos, only parent samples were
sequenced. (From The 1000 Genomes Project Consortium [2012] Nature 491:56–65. With permission from
Macmillan Publishers Ltd.)

has cataloged a large number of human genomes, it is not yet complete. The extended
project will only be complete when the scientists have sequenced 2500 individuals.

ENCODE will help to link genotype to phenotype in complex disease

The second phase of the ENCODE (Encyclopedia of DNA Elements) project was pub-
lished in Nature in 2012. The ENCODE project is designed to identify and catalog
functional elements across the genome. Only 3% of the total human genome encodes
functional proteins. The majority of the genome has been said to be made of “junk.” Of
course, this is a gross over-simplification. However, with so little of the genome encoding
functional sequences it is essential that we map and identify all of the functional elements
across the genome. The ENCODE project was designed to do this. We need to under-
stand how our genes are organized, and how they interact and regulate each other.
The first phase of ENCODE looked at 1% of the genome; the second stage (completed
in 2012) looked at 5% of the genome. The project has gone well beyond sequencing,
and is assessing the chromatin state and structure, level of DNA methylation, and pro-
tein expression, among other things, in multiple cell types from frequently used cell
lines, a human embryonic cell line, and some primary cells. The ultimate aim of this

2967_Ch12.indd 363 08/07/2015 14:46

 364 CHAPTER 12 Sequencing Technology and the Future

project is to bring together genotype and phenotype across the human genome. The proj-
ect is likely to ­continue for several years. The data generated so far can be viewed at
https://www.encodeproject.org/publications. Some scientists are critical of the ENCODE
project and not all believe that it will achieve all of its aims. However, time will tell.

12.8 Systems Biology

Considering systems biology allows us to look into the future
Biological systems are complex. Proteins interact. Protein agonists interact with specific
receptors and ligands, but many systems also include regulatory proteins in the forms
of receptor antagonists and decoy receptors. Systems also include a degree of redun-
dancy allowing our biology to make up for functional deficiencies in a specific product.
Altogether, we are complex beings. On top of this our genetics is complex, as we have seen,
and we produce variable levels of proteins depending (to a greater or lesser extent) on our
inherited genotype. We could say we each inherit different genotypes therefore we each
inherit different phenotypes.
We can illustrate the potential of genetic variation to impact on a biological system with
the following simplified example. Consider the interleukin (IL)-1 gene family that maps
to chromosome 2q. IL-1 is an important pro-inflammatory cytokine and genetic variation
in the IL1 genes has been associated with a variety of diseases, most notably with gastric
cancer. IL-1β is important in gastric acid production and gastric acid is important in
determining resistance to infection with the common bacteria Helicobacter pylori. There is
quite a large family of IL1 genes (Figure 12.15). To simplify the scenario, we will concen-
trate on one of the two major agonist (functional) forms of IL-1, IL-1β (encoded by the
IL1B gene), the active receptor gene IL1R1, a decoy receptor gene IL1R2, and a receptor
antagonist IL1RN. We can exclude the IL1A gene because the protein product, IL-1 α, is
mostly (if not exclusively) active within the cell.
All of the genes above are known to be polymorphic with multiple possible polymor-
phisms. Once again, for the sake of simplicity, let us consider each gene to have only two
alleles, which we can label allele 1 and 2. Let us also assume in this scenario that the two
alleles are associated with different levels of protein expression, with in every case high
levels of expression being encoded by allele 1 (high-producer allele) and low levels with
allele 2 (low-producer allele). In this scenario there are three genotypes for each gene: 1/1,
1/2, and 2/2. Consider first IL1B: those with the 1/1 genotype are high producers, those
with the 1/2 genotype are moderate producers (a combination of the effect of high- and
low-producer alleles), and those with the 2/2 genotype are low producers. The impact of
different levels of IL-1β production is, however, dependent on the other components of
the system. Thus, an individual with the IL1B 1/1 genotype and IL1R1 1/1 genotype can
have very high levels of IL-1β activity compared with someone with IL1B 2/2 and IL1R1 2/2.
This is because these individuals have high levels of the agonist form of IL-1 (IL-1β) and
high levels of the active receptor IL-1R1. In those with the IL1B 1/1 genotype (high pro-
ducers) and the IL1R1 1/2 and IL1R1 2/2 genotypes, which are associated with lower levels
of IL1R1 expression, overall IL-1β activity may be restricted. This is because even though
there is surplus IL-1β produced, the active receptor is less abundant and may become satu-
rated more rapidly. In this situation the effect of producing high levels of the agonist IL-1β
is reduced.
Add to this the further complications of a decoy receptor encoded by the IL1R2 gene and
a receptor antagonist encoded by the IL1RN gene. If the IL1R2 and IL1RN genotypes are

2967_Ch12.indd 364 08/07/2015 14:46

 Systems Biology 365

IL-1 map in detail

Chromosome 2q12–q22
Other genes in
IL-1 cluster IL1F

IL1R1 IL1A IL1B 7 9 6 8 5 10 IL1RN

IL1R2

10 kb 10 Mb 450 kb

Receptor IL-1 Antagonist

Decoy receptor

Figure 12.15: A map of some of the members of the IL-1 family located on chromosome 2q22. Only a
selected few of these genes are discussed in the example of a complex genetic system. The figure illustrates that
there are many more family members to consider, though not all are likely to be functional.

both low-level expression genotypes (2/2 in each case), then the potential activity of IL-1β
will remain high. However, where the genotypes for IL1B and IL1R1 are both high expres-
sion genotypes (1/1 and 1/1). IL-1β activity may be lower even when the genotype is that
associated with high IL1-B production (IL1B 1/1). In this model, high levels of the decoy
receptor will effectively reduce IL-1β binding to the active receptor and higher levels of
the receptor antagonist will interfere with the activity of the functional receptor. Overall,
IL-1β activity will be determined by the complex interaction of these four elements. This
interactive genetic minefield is illustrated in Figure 12.16.
Of course this is an oversimplified model for illustration only (though the genes do exist).
There are further levels of complexity in this system, e.g. IL-1β is produced in the cell in
the form of pre-IL-1β and needs to be converted by the enzyme caspase 1 [also known as
IL-1-converting enzyme (ICE)] before it can be exported from the cell. ICE may also be
polymorphic and so conversion of the pre-IL-1β is likely to be affected by this.

Agonist Functional
IL-1β receptor
IL-1R1

Receptor Decoy
antagonist receptor
IL-1RN IL-1R2

Figure 12.16: Simplified illustration of interaction in the IL-1 system. The figure illustrates the activity of the
products of four IL-1 genes: two genes encoding the agonist IL-1β and IL-1R, the receptor antagonist IL-1RN,
and the decoy receptor IL-1R2. IL-1β interacts with the receptor agonist IL-1R1 but this can be blocked by either
IL-1RN (the receptor antagonist) or by the decoy receptor (IL-1R2). This simple figure indicates how systems may
interact and coupled with the knowledge that the genes that encode these proteins may be polymorphic we can
produce a simple mind model to consider the potential complexity of this system. The gene family as illustrated in
Figure 12.15 is more complex.

2967_Ch12.indd 365 08/07/2015 14:46

 366 CHAPTER 12 Sequencing Technology and the Future

Even in this grossly simplified model it is possible to understand how systems biology
is important in helping us to understand the genetics of complex disease. It helps us to
understand why not all polymorphisms have functional significance and why in many
cases the risk associated with specific polymorphisms is small. Multiple interacting poly-
morphisms and extended haplotypes are likely to be important. Systems often also include
a level of redundancy whereby the effects of a dysfunctional gene or genotype can be com-
pensated for by another gene operating in the same, or in some cases another, pathway.
When either of these situations exist it is possible that effects of the susceptibility alleles are
only seen when the system is stressed by other factors, such as diet, illness, and infection.
Thus, most individuals carrying the risk alleles can cope with normal life and only when
stressed does the system collapse, leading to an increased risk of illness.

Conclusions
The past is part of the future, and the present is also part of the past and the future. Science
goes through phases that illustrate this concept well. Understanding the genetics of com-
plex disease began with old-fashioned technologies testing single genes and haplotypes for
case control association studies in the 1970s, 1980s, and early 1990s. Then came RFLP
and PCR analysis, which were faster and more efficient, but equally challenging. Just as we
were settling down, the new millennium was upon us with the HGM, HapMap, and SNP
Map. From HapMap and SNP Map, rapid genome wide case control and linkage studies
evolved (GWAS and GWLS). We are now entering a new phase. This is the beginning of
the third age in complex disease genetics. One in which techniques like whole genome
sequencing will be commonplace. These new technologies will produce data that will
increasingly be applied in clinical practice. However, as always with developments, there
are some pros and some cons associated with new technologies. The choice in the future
will be whether to use whole-genome or whole-exome sequencing. Choices over low- and
high-resolution genotyping will be applied in initial studies, but are unlikely to be on the
menu for large-scale studies. Missing data will be handled by systems like impute, which
will fill in the blanks.
There is clearly more to do in the study of the genetics of complex disease. The past 10
years have revolutionized the way we investigate the genetics of complex disease, but not
the reasons why we do it. We are beginning to understand just how complex the genetics
of common non-Mendelian disease may be. It is clear in 2015 that scientific isolationism
cannot be tolerated in genetics. If we are to understand how our genome contributes to
disease and use this information to the patient’s advantage, we must also understand how
factors other than our genes impact on human disease. Future strategies will also need to
include epigenetics and the environment, especially diet (discussed earlier in this book),
bacteria, and metagenomics.
We need to think of systems in biology and we need to interact more with our fellow bio-
scientists in order to identify the complex relationships that make the human body work.
A systems approach to genetics is required so that we can understand how genetic varia-
tions can predispose to, or protect from, complex traits. It is not going to be simple, but
we have made a start, and with hard work and perseverance great advances in biomedicine
are possible.
Studies will be able to take advantage of ongoing projects such as the 1000 Genomes
Project and ENCODE (and modENCODE), which were designed to inform studies
of human genetics, including those involving disease. Of course we cannot ignore the

2967_Ch12.indd 366 08/07/2015 14:46

 Conclusions 367

cornucopia of information provided by the HGP and the disease studies performed so far.
Better, faster, and more precise technologies will also contribute.
Ultimately, progress will be measured in terms of the extent to which our research fulfils
the three major promises of the Human Genome Mapping Project, i.e. that genetics will
be used in disease diagnosis, that genetics will be used in patient management (especially
in the development of individualized and novel therapies), and that genetics will be used
to inform the debate on disease pathology.
These three promises are linked. Understanding disease pathogenesis goes cap in hand
with better patient management and development of individualized and novel therapies.
The use of complex disease genetics in diagnosis is perhaps questionable in all but a few
cases, but there is certainly evidence that studies are helping us to understand disease
pathology and there are some examples where allelic variation may determine the selection
of specific therapeutic options. Some would call this translational medicine, but it is not
a new idea; scientists in medicine have always sought to relate basic findings to medical
practice. “Bench-to-bedside” is a community goal and not just the dream of a few.
Overall, great advances have been made in understanding complex disease. This is espe-
cially true in some areas, such as pharmacogenetics and in immune inflammatory diseases
such as Crohn’s disease. However, we should not be misled by wishful science, and we must
remember the words of satirist and philosopher Hugo Mencken who stated in 1917 “there
is always an easy solution for every human problem (that is) neat, plausible and wrong”.
The future lies with unpicking and reassembling these complex human problems and in
doing so reaching our final goal, i.e. to advance medical practice through patient diagnosis
and treatment, with a hope that some diseases (at least) may one day be eradicated.

2967_Ch12.indd 367 08/07/2015 14:46

 368 CHAPTER 12 Sequencing Technology and the Future

Further Reading
Books the first complete genome sequence from an
Armstrong L (2014) Epigenetics. Garland African DNA sample; see also Rasmussen et al.
Sciences. This is an excellent book on epi- (2010), Wang et al. (2008), and Wheeler et al.
genetics for the graduate and final-year under- (2008).
graduate scientist with an interest or need to Blencowe BJ, Ahmad S & Lee LJ (2009)
study this subject. Current-generation high-throughput sequenc-
Collins F (2010) The Language of Life: DNA ing: deepening insights into mammalian tran-
and the Revolution in Personalised Medicine. scriptomes. Genes Dev 23:1379–1386. This
Harper Collins. This is an excellent book to delve paper reports on sequencing the transcriptome;
into to look at the whys and wherefores of gene see also Wang et al. (2009).
hunting in complex disease. Branton D, Deamer DW, Marziali A et  al.
Frank L (2011) My Beautiful Genome: (2008) The potential and challenges of nano-
Exposing Our Genetic Future, One Quirk at a pore sequencing. Nat Biotechnol 26:1146–1153.
Time. Oneworld. This is an excellent book for Braslavsky I, Hebert B, Kartalov E & Quake SR
the lay public and scientist alike. It is interest- (2003) Sequence information can be obtained
ing, controversial, and humorous, and perfect from single DNA molecules. Proc Natl Acad Sci
for the student of complex disease genetics. USA 100:3960–3964.
Maitland-van der Zee A-H & Daly AK (2012) Campbell PJ, Stephens PJ, Pleasance ED et al.
Pharmacogenetics and Individualized Therapy. (2008) Identification of somatically acquired
Wiley. rearrangements in cancer using genome-wide
Spector T (2012) Identically Different: Why massively parallel paired-end sequencing. Nat
You Can Change Your Genes. Weidenfeld & Genet 40:722–729. This is one of three papers
Nicholson. on pair-ended mapping that demonstrate the
Strachan T & Read AP (2011) Human use and application of this technique; see also
Molecular Genetics, 4th edn. Garland Science. Korbel et al. (2007) and Stephens et al. (2009).
Strachan T, Goodship J & Chinnery P (2014) Cohen JC, Kiss RS. Pertsemlidis A et al. (2004)
Genetics and Genomics in Medicine. Garland Multiple rare alleles contribute to low plasma
Science. levels of HDL cholesterol. Science 305:869–
872. This study demonstrates the effectiveness
of sequencing candidate genes at the extreme
Articles ends of the phenotype distribution to find rare
Anderson CA, Soranzo N, Zeggini E & alleles.
Barrett JC (2011) Synthetic associations are Dean M, Carrington M, Winkler C et  al.
unlikely to account for many common disease (1996) Genetic restriction of HIV-1 infection
genome-wide association signals. PLoS Biol and progression to AIDS by a deletion allele of
9 (1):e1000580. This paper suggests that the the CKR5 structural gene. Hemophilia Growth
importance of synthetic associations generated and Development Study, Multicenter AIDS
by multiple variants has perhaps been over- Cohort Study, Multicenter Hemophilia Cohort
stated; see also Wray et al. (2011). Study, San Francisco City Cohort, ALIVE
Barski A, Cuddapah S, Cui K et  al. (2007) Study. Science 273:1856–1862. This report is
High-resolution profiling of histone methyl- of great importance in understanding the inter-
ations in the human genome. Cell 129:823– play between host genetics and the pathogen.
837. This paper describes the use of the ChIP Here the host is protected from infection and
sequencing method to characterize histone development of the disease by the presence of
and transcription factor binding sites on CD4 the 32-bp deletion CCR5-Δ32; see also Liu
T cells; see also Mikkelsenet et al. (2007) and et al. (1996) and Samson et al. (1996).
Robertson et al. (2007). de Souza N (2012) The ENCODE Project. Nat
Bentley DR, Balasubramanian S, Swerdlow HP Methods 9:1046.
et  al. (2008) Accurate whole human genome Dickson SP, Wang K, Krantz I et  al. (2010)
sequencing using reversible terminator chem- Rare variants create synthetic genome-wide
istry. Nature 456:53–59. This paper describes associations. PLoS Biol 8 (1):e1000294. This

2967_Ch12.indd 368 08/07/2015 14:46

 Further Reading 369

paper, along with Maher (2008) and Schork use of genotype imputation, which is increas-
et  al. (2009), poses the question of whether ingly used in studies of complex disease either
studies so far have identified the true suscep- to fill in the gaps or to extend the study; see also
tibility alleles for complex disease or are they Marchini et al. (2007).
synthetic associations? Liu R, Paxton WA, Choe S et  al. (1996)
Eid J, Fehr A, Gray J et  al. (2009) Real-time Homozygous defect in HIV-1 co-receptor
DNA sequencing from single polymerase mol- accounts for resistance of some multiply-
ecules. Science 323:133–138. exposed individuals to HIV-1 infection. Cell
Freeman JL, Perry GH, Feuk L et  al. (2006) 86:367–377. This report is of great importance
Copy number variation: new insights in in understanding the interplay between host
genome diversity. Genome Res 16:949–961. genetics and the pathogen. Here the host is
This paper illustrates how second-generation protected from infection and development of
sequencing will be used to identify CNV in the disease by the presence of the 32-bp dele-
the genome; see also Redon et  al. (2006) and tion CCR5-Δ32; see also Dean et al. (1996) and
Zhang et al. (2009). Samson et al. (1996).
Harris TD, Buzby PR, Babcock H et al. (2008) Lupski JR, Reid JG, Gonzaga-Jauregui C et al.
Single-molecule DNA sequencing of a viral (2010) Whole-genome sequencing in a patient
genome. Science 320:106–109. This paper with Charcot–Marie–Tooth neuropathy. N
describes the use of the Helicos platform (third- Engl J Med 362:1181–1191.
generation sequencing). Maher B (2008). The case of the missing heri-
Hayden E (2008). International genome proj- tability. Nature 456:18–21. This news feature
ect launched. Nature 451:378–379. This paper along with papers by Dickson et  al. (2010)
announces the launch of the 1000 Genomes and Schork et al. (2009), poses the question of
Project. whether studies so far have identified the true
susceptibility alleles for complex disease or are
Howie BN, Donnelly P & Marchini J (2009) they synthetic associations?
A flexible and accurate genotype imputation
method for the next generation of genome-wide Maher CA, Kumar-Sinha C, Cao X et  al.
association studies. PLoS Genetics 5:e1000529. (2009) Transcriptome sequencing to detect
gene fusions in cancer. Nature 458:97–101.
International Human Genome Sequencing This study demonstrates the potential use of
Consortium (2001) Initial sequencing and transcriptome sequencing in clinical practice.
analysis of the human genome. Nature
409:860–921. This paper, together with that of Marchini J, Howie B, Myers S et al. (2007) A
Venter et al. (2001), describes the preliminary new multipoint method for genome-wide asso-
sequence of the first whole human genome. ciation studies by imputation of genotypes. Nat
Genet 39:906–913. This paper discusses the use
Johnson PL & Slatkin M (2008). Accounting for of genotype imputation, which is increasingly
bias from sequencing error in population genetic used in studies of complex disease either to fill
estimates. Mol Biol Evol 25:199–206. This report in the gaps or to extend the study; see also Li
discusses some of the concerns that have been et al. (2009).
raised about the use of NGS in GWAS.
Margeridon-Thermet S, Shulman NS, Ahmed
Korbel JO, Urban AE, Affourtit JP et al. (2007) A et  al. (2009) Ultra-deep pyrosequencing of
Paired-end mapping reveals extensive struc- hepatitis B virus quasispecies from nucleoside
tural variation in the human genome. Science and nucleotide reverse-transcriptase inhibi-
318:420–426. This is one of three papers on pair- tor (NRTI)-treated patients and NRTI-naive
ended mapping that demonstrate the use and patients. J Infect Dis 199:1275–1285. This
application of this technique; see also Campbell paper describes a study of a viral drug resistance
et al. (2008) and Stephens et al. (2009). mutation using a metagenomics approach; see
Le T, Chiarella J, Simen BB et al. (2009) Low- also Nakamura et al. (2008) and Palacios et al.
abundance HIV drug-resistant viral variants in (2008).
treatment-experienced persons correlate with Margulies M, Egholm M, Altman WE et  al.
historical antiretroviral use. PLoS One 4:e6079. (2005) Genome sequencing in microfabri-
Li Y, Willer C, Sanna S & Abecasis G (2009) cated high-density picolitre reactors. Nature
Genotype imputation. Ann Rev Genomics Hum 437:376–380. The paper describes the use
Genet 10:387–406. This paper discusses the of a novel system for rapid “shotgun” gene

2967_Ch12.indd 369 08/07/2015 14:46

 370 CHAPTER 12 Sequencing Technology and the Future

sequencing of the genome of the bacterium a particularly interesting report of a human

Mycoplasma genitalium. genome sequence from an ancient sample.
Mikkelsen TS, Ku M, Jaffe DB et  al. (2007) This study proves the concept of rapid genome
Genome-wide maps of chromatin state in plu- sequencing and the ability to sequence not
ripotent and lineage-committed cells. Nature only fresh samples, but also ancient samples.
448:553–560. This paper describes the use of This paper describes the first complete genome
the ChIP sequencing method to characterize sequence from a European DNA sample; see
histone and transcription factor binding sites also Bentley et al. (2008), Wang et al. (2008),
in a cervical carcinoma cell line; see also Barski and Wheeler (2008).
et al. (2007) and Robertson et al. (2007). Redon R, Ishikawa S, Fitch KR et  al. (2006)
Mullis K, Faloona E, Scharf S et  al. (1986) Global variation in copy number in the human
Specific enzymatic amplification of DNA in genome. Nature 444:444–454. This paper
vitro: the polymerase chain reaction. Cold Spring illustrates how second-generation sequencing
Harbor Symp Quant Biol 51:263–273. This rep- will be used to identify CNV in the genome;
resents a major development in genetics. see also Freeman et al. (2006) and Zhang et al.
(2009).
Nakamura S, Maeda N, Miron IM et al. (2008)
Metagenomic diagnosis of bacterial infections. Ren B, Robert F, Wyrick JJ et  al. (2000)
Emerg Infect Dis 14:1784–1786. This paper Genome-wide location and function of DNA
describes a study of bacterial pathogens using a binding proteins. Science 290:2306–2309.
metagenomics approach; see also Margeridon- This paper describes the chromatin immuno-
Thermet et al. (2009) and Palacios et al. (2008). precipitation assay (ChIP-chip) used to detect
locations where the target protein is bound to
Palacios G, Druce J, Du L et al. (2008) A new DNA.
Arenavirus in a cluster of fatal transplant-asso-
Roach JC, Glusman G, Smit AF et al. (2010)
ciated diseases. N Engl J Med 358:991–998.
Analysis of genetic inheritance in a family
This paper describes a study of viral patho-
quartet by whole-genome sequencing. Science
gens using a metagenomics approach; see
328:636–639.
also Margeridon-Thermet et  al. (2009) and
Nakamura et al. (2008). Robertson G, Hirst M, Bainbridge M et  al.
(2007) Genome-wide profiles of STAT1 DNA
Pool JE, Hellmann I, Jensen JD & Nielsen association using chromatin immunoprecipi-
R (2010) Population genetic inference from tation and massively parallel sequencing. Nat
genomic sequence variation. Genome Res Methods 4:651–657. This paper describes the
20:291–300. This paper discusses population use of the ChIP sequencing method to charac-
variation in relation to whole-genome sequenc- terize histone and transcription factor binding
ing and the use of inference (imputation) to sites in pluripotent murine stem cells undergo-
assign missing data. ing differentiation; see also Barski et al. (2007)
Prokunina-Olsson L & Hall JL (2009) No and Mikkelsenet et al. (2007).
effect of cancer-associated SNP rs6983267 in Samson M, Libert F, Doranz BJ et  al. (1996)
the 8q24 region on co-expression of MYC and Resistance to HIV-1 infection in Caucasian
TCF7L2 in normal colon tissue. Mole Cancer individuals bearing mutant alleles of the
8:96. This report indicates that tagged SNPs can CCR-5 chemokine receptor gene. Nature
be located a long way from putative causative 382:722–725. This report is of great impor-
mutations in complex disease, but establishing tance in understanding the interplay between
a relationship with function can be difficult; see host genetics and the pathogen. Here the host
also Tomlinson et al. (2007). is protected from infection and development of
Pushkarev D, Neff NF & Quake SR (2009) the disease by the presence of the 32-bp dele-
Single-molecule sequencing of an individual tion CCR5-Δ32; see also Dean et al. (1996) and
human genome. Nat Biotechnol 27:847–852. Liu et al. (2006).
This paper describes the use of the Helicos plat- Sanger F, Nicklen S & Coulson AR (1977)
form (third-generation sequencing); see also DNA sequencing with chain-terminating
Harris et al. (2008). inhibitors. Proc Natl Acad Sci USA 74:5463–
Rasmussen M, Li Y, Lindgreen S et al. (2010) 5467. This represents a major development in
Ancient human genome sequence of an extinct genetics and, in combination with the devel-
Palaeo-Eskimo. Nature 463:757–762. This is opment of PCR by Mullis et  al. (1986), this

2967_Ch12.indd 370 08/07/2015 14:46

 Further Reading 371

enabled automated sequencing of DNA to disease. The impact of these maps and studies
become a reality. cannot be overestimated.
Schork NJ, Murray SS, Frazer KA & Topol EJ Tomlinson I, Webb E, Carvajal-Carmona L
(2009) Common versus rare allele hypotheses et  al. (2007) A genome-wide association scan
for complex diseases. Curr Opin Genet Dev of tag SNPs identifies a susceptibility variant
19:212–219. This paper, along with Maher for colorectal cancer at 8q24.21. Nat Genet
(2008) and Dickson et  al. (2010), poses the 39:984–988. This report indicates that tagged-
question of whether studies so far have identi- SNPs can be located a long way from putative
fied the true susceptibility alleles for complex causative mutations in complex disease, but
disease or are they synthetic associations? establishing a relationship with function can be
Stephens PJ, McBride DJ, Lin ML et al. (2009) difficult; see also Prokunina-Olsson L & Hall
Complex landscapes of somatic rearrange- JL (2009).
ment in human breast cancer genomes. Nature Venter JC, Adams MD, Myers EW et al. (2001)
462:1005–1010. This is one of three papers The sequence of the human genome. Science
on pair-ended mapping that demonstrate 291:1304–1351. This paper, together with
the use and application of this technique; see that of the International Genome Consortium
also Korbel et  al. (2007) and Campbell et  al. (2001), describes the preliminary sequence of
(2008). the first whole human genome.
Sultan M, Schulz MH, Richard H et al. (2008) Wang J, Wang W, Li R et al. (2008) The dip-
A global view of gene activity and alternative loid genome sequence of an Asian individual.
splicing by deep sequencing of the human tran- Nature 456:60–65. This paper describes the
scriptome. Science 321:956–960. This paper first complete genome sequence from an Asian
reports on the application of transcriptome DNA sample; see also Bentley et  al. (2008),
sequencing in human kidney and B cell lines; Rasmussen et  al. (2010), and Wheeler et  al.
see also Maher et al. (2009). (2008).
The 1000 Genomes Project Consortium Wang Z, Gerstein M & Snyder M. (2009)
(2010) A map of human genome variation RNA-Seq: a revolutionary tool for transcrip-
from population-scale sequencing. Nature tomics. Nat Rev Genet 10:57–63. This paper
467:1061–1073. reports on sequencing the transcriptome; see
The 1000 Genomes Project Consortium (2012) also Blencowe et al.
An integrated map of genetic variation from Wheeler DA, Srinivasan M, Egholm M et  al.
1,092 human genomes. Nature 491:56–65. (2008) The complete genome of an individual
This is the most recent report from the 1000 by massively parallel DNA sequencing. Nature
Genomes Project at the time of writing. More 452:872–876. This paper describes the first
up-to-date information will be available on the complete genome sequence from a European
website cited below. DNA sample; see also Bentley et  al. (2008),
The ENCODE Project Consortium (2012) An Rasmussen et  al. (2010), and Wang et  al.
integrated encyclopedia of DNA elements in (2008).
the human genome. Nature 489:57–74. This is
Wray NR, Purcell SM & Visscher PM (2011)
the original report from the ENCODE Project
Synthetic associations created by rare vari-
reporting from phase II, which started in 2007.
ants do not explain most GWAS results. PLoS
The International HapMap Consortium (2005) Biol 9:e1000579. This paper suggests that the
A haplotype map of the human genome. Nature importance of synthetic associations generated
437:1299–1320. by multiple variants has perhaps been over-
The International HapMap Consortium (2007) stated; see also Anderson et al. (2011).
A second generation human haplotype map of Zhang F, Gu W, Hurles ME & Lupski JR (2009)
over 3.1 million SNPs. Nature 449:851–861. Copy number variation in human health, dis-
The International HapMap Consortium (2010) ease and evolution. Annu Rev Genomics Hum
Integrating common and rare genetic variation Genet 10:451–481. This paper illustrates how
in diverse human populations. Nature 467:52– second generation sequencing will be used to
58. The HapMap data can be accessed through identify copy number variation in the genome;
the weblink in the Online sources and provides see also Redon et al. (2006) and Freeman et al.
a very useful database for all studies in complex (2006).

2967_Ch12.indd 371 08/07/2015 14:46

 372 CHAPTER 12 Sequencing Technology and the Future

Online sources http://www.encodeproject.org/publications/

http://www.hapmap.org This database shows the data produced by the
This is a most useful source for studies of the encode project so far.
genetics of complex disease and is essential in http://www.454.com
study design. This website provides information on the Life
http://www.1000genomes.org Sciences NGS technology owned by Roche.
Data from this public database can be used http://www.appliedbiosystems.com
freely for research purposes. The data can help This website provides information on the
with imputation studies and identifying com- Applied Biosystems SOLiD NGS system
mon haplotypes in different populations. owned by Applied Biosystems.
http://www.ncbi.nlm.nih.gov http://www.illumina.com
This website provides a reference database for This website provides information on the
metagenomic studies and bacterial genomes. Illumina Genome Analyzer—an NGS sequencer
http://commonfund.nih.gov/hmp produced by the Illumina corporation.
This website refers to the Human Microbiome
Project.

2967_Ch12.indd 372 08/07/2015 14:46

 Glossary

Acentric – term describing a chromosome that does not possess a centromere.

Adaptive Darwinian selection – selection of specific genotypes within a population as a
result of selection pressure.
Adaptive immunity – modified immune response following initial stimulation with an
antigen.
ADP receptor – adenosine diphosphate receptor. This is an important receptor for many
drugs.
Affymetrix gene chip – involves light-directed in situ synthesis of oligonucleotide probes
on a microarray that can produce approximately 6 million oligonucleotides on a chip array
of less than 2 cm2.
Allele(s) – alternative forms of the same gene.
Allele frequency – frequency of a gene variant at a specific locus.
Allelic heterogeneity/variation – existence of several different mutations or polymor-
phisms, all in the same gene, in unrelated individuals with the same phenotype.
Allo-reactive antibodies – antibodies that recognize antigens of the same species.
Allotropic expression – most often referred to regarding genes that are normally expressed
from the mitochondrial genome.
Alternative hypothesis (H1) – the opposite idea to the null hypothesis, which suggests
there is no significant difference in a study. The alternative hypothesis is incompatible with
the null hypothesis. The alternative hypothesis suggests there is a difference between the
two groups.
Amyloid plaques – caused by deposits of β-amyloid protein in the brain.
Ancestral haplotype (e.g. HLA 7.2 and 8.1) – a group of alleles located on a series of
different genes that are inherited as a single unit and from a common ancestor. Haplotypes
like this are preserved and their component alleles are found together more often than
expected by chance (i.e. they are in linkage disequilibrium).
Aneuploidy – an extra copy or missing copy of one or more chromosomes. Aneuploidies
include trisomy 21 (Down syndrome) and monosomy X (Turner’s syndrome).
Antibody-dependent cytotoxicity – the use of antibodies to kill cells as in an HLA sero-
typing assay.
Antigen-binding groove (HLA-binding groove) – the location on an HLA class I or II
molecule where antigenic peptides are bound and presented to the T cell receptor.
Antigen-presenting cells (APCs) – specific cells that present antigenic peptides to T lym-
phocytes and activate them. These include dendritic cells, macrophages, and B cells, and

2967_Glossary.indd 373 07/07/2015 11:09

 374 Glossary

are often referred to as professional antigen-presenting cells. All of these three cell types
express HLA class II molecules and a number of other co-stimulatory molecules on their
surface required for activation of naive T cells.
Antigenic peptides – refers to short antigen protein sequences that are presented to T
cells for the immune response. They can be derived from self-antigens or from non-self-
antigens (e.g. pathogens).
Appreciable frequency – can be any value suggested to be significant for a study. Generally
refers to numbers in studies, size of difference reported, and the level of expected probability
value. Also a term used in the definition of polymorphic genes as opposed to mutations.
Ascertainment bias – distorted recruitment (usually of individuals for a study, either as
controls, disease cases, or both).
Association(s) – tendency of a marker and a trait to occur together more often than
expected by chance. Association is technically a statistical observation and is not a genetic
phenomenon; however, association is used in genetic studies and can also indicate linkage
disequilibrium.
Association analysis (studies) – attempts to find genetic associations between genetic
markers i.e. alleles, SNPs, mutations, or other genetic variation and diseases by comparing
the frequencies of markers in two different groups.
Autophagy – a form of programmed cell death in which the cell components are broken
down as a means of coping with adverse circumstances or during infection.
Autosomal dominant disorder or disease – a disorder or disease (discord is also used)
showing a Mendelian pattern of inheritance whereby only one parental genome needs to
pass on the disease-causing mutation to their offspring for the trait to occur.
Autosomal recessive disorder or disease – a disorder or disease (discord is also used)
showing a Mendelian pattern of inheritance whereby both parental chromosomes need to
pass on the disease-causing mutation to their offspring for the trait to occur.
Autosome– any chromosome other than the sex chromosomes.
Balanced structural abnormalities – occur when there is no loss or gain of function
associated with an abnormality.
Balancing selection – occurs where mutant alleles confer increased fitness on the popula-
tion, but only in heterozygotes and not in homozygotes.
Basal cell carcinoma – common form of skin cancer; rarely fatal.
Bayesian – a form of mathematical calculation based on Bayes’ theorem.
Bonferroni’s correction – a mathematical correction applied by statisticians to correct for
multiple testing in different studies. It is mostly used in case control studies. It is much
criticized, but it has been widely used in the past.
Bottleneck effect – this occurs when a founder population is reduced in size as a result of
a catastrophic event and it leads to reduced genetic diversity in the population.
Candidate gene-based association study – a genetic study based on investigation of one
or more identified “candidates.” This is a hypothesis-constrained study.
Carcinogenesis – literally the creation of cancer. The process by which normal cells
become cancer cells.

2967_Glossary.indd 374 07/07/2015 11:09

 Glossary 375

Cardiotoxicity – dysfunction in the electrophysiological processes in the heart or heart

muscle.
Case control (association studies/population) studies – an investigation based on the
comparison of two or more groups where one (or more) group consists of patients (cases)
and the other consists of controls.
Cataplectic narcolepsy – severe form of a chronic neurological disorder (narcolepsy)
caused by failure to regulate the sleep/wake cycle. Cataplexy is caused by a sudden loss of
voluntary muscle tone during which time an individual is unable to move.
CD4+ T cells – specific T lymphocytes that express CD4 on their cell surface interacting
with HLA class II.
CD8+ T cells – specific T lymphocytes that express CD8 on their cell surface interacting
with HLA class I.
Centimorgan (cM) – measure of genetic distance on a chromosome based on the recom-
bination fraction (named after geneticist TH Morgan); 0.01 or 1% recombination frac-
tion is equal to 1 cM.
Centromere – point at which the two strands of the chromosome (p-short and q-long)
merge. It is important in the construction of the chromosome and in providing an anchor
point at which spindle fibers attach to pull chromatids apart during cell division.
Chemokines – a group of small proteins that guide white blood cells to the sites where
they are needed.
Chimeric – a functional mix of genetic or other biological material from one or other
members of the same or different species.
Chromatin – packaged DNA in a cell. Double coils of approximately 30 nm and histones.
Chromatin immunoprecipitation assay (ChIP-chip) – used to detect biochemical mod-
ifications to DNA and locations where proteins of interest are bound to DNA, such as
histone and transcription-binding sites on human CD4+ T cells.
Chromosomes – linear structures comprised of long DNA molecules coupled with a
variety of structural and regulatory proteins.
Chromosomal abnormalities – changes in chromosomal structure or number, e.g. the
Philadelphia chromosome or trisomy 21.
Chronic myelogenous leukemia – one form of cancer of the white blood cells.
Cis expression – the expression of two different alleles at different loci on the same pater-
nal chromosome. Resulting in the formation of a heterodimer in cis as opposed to a het-
erodimer in trans. See Trans expression.
Clinical heterogeneity – variation in the clinical conditions in a test population.
Clones – copies of a cell or molecule (e.g. a short DNA or longer chromosomal sequence)
produced under laboratory conditions.
Cochran–Armitage test – a statistical method used to test for significance when compar-
ing data for a variable with two possibilities (e.g. a gene with two alleles) against a variable
with multiple variables. The test is very similar to the χ2 test.
Codon – a nucleotide tri-repeat sequence that encodes an amino acid.

2967_Glossary.indd 375 07/07/2015 11:09

 376 Glossary

Common disease common variant hypothesis (CDCVH) – hypothesis that predicts

that common disease-causing alleles will be found in all human populations that express
specific traits and diseases.
Common disease rare variant hypothesis (CDRVH) – hypothesis that predicts that if a
disease is common in the population the genetic risk may not be common in the popula-
tion but may be due to the effect of multiple interacting risk alleles.
Compare and contrast model – a simple model where the details of two sides are com-
pared to seek similarity and differences between them.
Complement (complement activation) – a group of plasma proteins that interact to
neutralize pathogens and activate pathogen destruction and eradication.
Complement-mediated cell lysis – one of the major functions of the complement pro-
teins; the destruction of bacteria and also cells infected by viruses.
Complex disease/trait – a disease/trait or character where one or more alleles acting alone
or in concert increase the risk.
Concordance – the degree of agreement (harmony) between two observations, ideas, or
arguments.
Confidence interval (CI) – a statistical measure of the range of probability within which
any quoted value falls.
Consequentialism – an ethical theory that states the consequences of an individual’s con-
duct are the ultimate basis for any judgment of the right of that conduct.
Constitutional chromosomal abnormalities – chromosomal abnormalities present in
every cell in the body.
Copy number variation (CNV) – inherited variation in the number of copies of a specific
gene between individuals. Not all CNVs have functional consequences.
Co-segregation – the transmission of two or more genes on the same chromosome to the
daughter cell.
Crossover – occurs during meiosis whereby fragments of parental chromosomes interact
and combine to form new combinations (see Recombination).
Cyto-adherence – a process that can be used to block adherence of virus to the cell. The
process involves antibodies.
Cytochrome P450 – a family of phase I drug-metabolizing enzymes (see CYP2D6,
CYP2C9, and CYP2C19).
Cytokine – proteins made by cells that influence the behavior of other cells. If made by
lymphocytes, they are called lymphokines or interleukins.
Degrees of Freedom (d.f.) – a statistical measure of the number of tests performed in an
analysis. Probability values for some commonly used tests are routinely adjusted taking
into account the number of degrees of freedom and values of probability can be found in
published sources for tests such as the χ2 test, etc..
Dicentric – a chromosome with two centromeres.
Diploid – having two copies of the whole chromosome set.
Direct sequencing – early use of the term “direct sequencing” referred to DNA sequenc-
ing based on use of PCR to amplify the gene or region of interest. More recently, it has

2967_Glossary.indd 376 07/07/2015 11:09

 Glossary 377

been used to describe sequencing without the need for cloning of the DNA fragment
under investigation.
Directional selection – a situation in natural selection where a particular phenotype is
favored and the allele frequency moves in one direction.
Disease cassette – the concept that a number of risk alleles carried on the same haplotype
may each contribute to disease susceptibility.
Disease-causing mutation – a mutation that shows a direct causal link with disease (see
Mutation).
Disease phenotype – the pattern of a disease, including signs, symptoms, progression,
and response to treatment.
Disease profile – idea that a disease may arise when we inherit series of different alleles
with different patients having different personal profiles of risk alleles in complex disease
(see also personal portfolio).
Dispermy – two sperm fertilizing a single egg. A common cause of triploidy.
Dizygotic twins – twins who arise as a result of independent fertilization of two different
eggs by two different sperm at the same time. Dizygotic twins are as genetically identi-
cal as other siblings in the same family; they do not carry an identical genome. Also see
Monozygotic twins.
DNA – abbreviation for deoxyribonucleic acid.
DNA methylation – conversion of cytosine in DNA to 5-methylcytosine. This is an
important process in the regulation of gene expression.
Duplications – increase in the number of copies in a gene or a DNA sequence. Where there
are duplications in specific genes, individuals may carry more than one copy of a gene. This
does not always have any influence on the phenotype such as in circumstances where the dupli-
cated copy is not expressed. However, duplications can have interesting effects on phenotype.
Electrostatic charge – the charge on an amino acid: positive, negative, or neutral. Charge
is important in considering protein structures, formation, and function.
ENCODE – Encyclopedia of DNA Elements. An ongoing project targeting functional
elements in selected model organisms (new name modENCODE).
Endocytosis – uptake of extracellular materials into cells. Endosomes are formed from
segments of the cell membrane and capture extracellular materials. These are then trans-
ported through the cell. This process is particularly important in immunity.
Endoplasmic reticulum (ER) – an intercellular organelle that forms an interconnecting
network with the cell membrane. There are two forms of the ER: smooth (SER) and rough
(RER). The RER is involved with protein synthesis, while the SER is involved with lipids
and detoxification.
Enrichment strategies – targeting of specific selected genomic regions in the treatment of
disease by using data from studies of somatic and de novo mutations identified in whole-
exome sequencing.
Epidemiological – in the context of this book this is the study of health and disease in
populations – considering causes, response, and outcomes.
Epigenetic(s) – means above, outside or in addition to genetics and is usually used to refer
to heritable effects not produced by changes in the DNA sequence. DNA methylation is

2967_Glossary.indd 377 07/07/2015 11:09

 378 Glossary

the most often quoted form of epigenetics. The wider context for this growing sub-branch
of genetics is to be found in all things outside of “genetics.”
Epigenetic regulation – DNA methylation, in particular.
Epigenome-wide association studies (EWAS) – epigenetic studies performed on a
genome-wide basis have started and more will follow in the future.
Epigenomics – the growing interest in the wider consideration of genetics, especially
epigenetics.
Epistasis – means standing above. In this case, we are concerned with the interaction
of genes in pathways. Where there is epistasis, there is a sequence of activation of genes.
Failure in the early stages due to genetic variation would be expected to cause failure fur-
ther down the pathway. This is an important concept in systems biology.
Eugenics (Eugenics Movement) – an idea conceived by Francis Galton (the cousin of
Charles Darwin) based on the principle that social status is an inherited trait and therefore
the upper classes should do everything possible to conserve themselves, whereas the lower
classes are simply the unfortunate consequence of their genes. This misconception had
major historical consequences.
Exome – the region of the genome that encodes the genes (protein sequences).
Exome sequencing – restriction of genome sequencing to these coding regions only; this
reduces cost and time.
Exon(s) – DNA sequences in genes that encode the protein/peptide sequences. Also see
intron(s).
Extended major histocompatibility complex (xMHC) – the major histocompatibility
complex (MHC) after 2004, at which point the MHC was extended to include the hemo-
chromatosis gene HFE (telomeric of HLA-A). The extension added approximately 3.5 Mb
to the existing MHC.
Factorial (!) – mathematical value whereby the value is the sum of that value multiplied
in sequence with lower values, e.g. 5! = 5 × 4 × 3 × 2 × 1 = 120. Used in Fisher’s exact
probability test. The symbol “!” is the mathematical sign for factorial.
False discovery rate – the number of false positives discovered in a study compared with
the total number of tests performed and the total number of true positives.
False negative – acceptance of the null hypothesis for an allele that later turns out to be
incorrect. In GWAS, this means dismissing a genetic association that later turns out to be
true.
False positive – the rejection of the null hypothesis for an allele that later turns out to
be true. In GWAS, this means accepting the presence of a genetic association that is later
proven to be incorrect.
Familial (family studies/familial disease) – studies based on the collection of informa-
tive families.
Family-based association test (FBAT) – a form of transmission disequilibrium testing
(TDT).
Fisher’s exact probability test – a statistical method for testing for associations usually
applied when any individual value is less than 5 though not all studies follow this rule.
Fisher worked with Yates to set up the original χ2 test.

2967_Glossary.indd 378 07/07/2015 11:09

 Glossary 379

Fixation – the situation that occurs when through natural selection in a population there
is a change in the gene pool, such that where there were once two alleles, there is now only
one allele. In this situation the gene is said to be fixed, i.e. invariant.
Fluorescence in situ hybridization (FISH) – in situ hybridization using fluorescently
labeled DNA or RNA.
Founder effect – a high frequency of a particular allele due to small numbers in the origi-
nal population (founder population).
Frameshift – a shift in the remaining DNA when part of a DNA sequence is deleted. This
can have consequences for the gene product as, for example, with the CCR5-Δ32 mutation.
Gametocytes – precursors to the male and female gametes.
Gene – a basic unit of inheritance. A functional DNA unit that determines the phenotype
of an individual and segregates in pedigrees according to Mendel’s laws.
Gene duplications – see Duplication.
Gene flow – transfer of alleles or genes between populations through migration.
Gene pool – all the genes (or a selection) in a specified population.
Genetic anticipation – the tendency of the severity of a condition to increase over genera-
tions, often due to the increase in copy number from generation to generation.
Genetic divergence – accumulation of genetic variation from two or more ancestral pop-
ulations over time giving rise to even more genetic variation.
Genetic drift – random changes in gene frequencies over generations.
Genetic imprinting – when the expression of a gene is determined by the parental ori-
gin of that gene. Thus, expression can be different if the same allele is inherited from the
paternal or maternal route.
Genetic portfolio – the sum of an individual’s genotype, including the high-, low-, and
medium-risk alleles for all of the genes in the genome or in research reports for all of the
alleles tested.
Genetic protection – alleles that reduce the risk of a disease.
Genetic selection – the concept that survival is determined in part by genetic variation in
the population. This is a central pillar of Darwinism.
Genetic underclass – the distinction of a subgroup as inferior based on their genetic
makeup. Also see Eugenics Movement and Chapter 11.
Genome – the complete set of human DNA stored in our chromosomes (22 pairs plus X
and Y) and our mitochondria.
Genome sequencing – sequencing the entire genome as opposed to sequencing only
selected sections.
Genome project – an international project to sequence the genome of humans and other
species, and to catalogue variation within the human genome.
Genome-wide association analysis/studies (GWAA/GWAS) – analysis of genetic
markers across the whole genome to identify significant differences in allele distribution
between groups (usually healthy controls and disease cases, though not always). GWAS
has been highly successful in recent years.

2967_Glossary.indd 379 07/07/2015 11:09

 380 Glossary

Genome-wide linkage analysis/studies (GWLA/GWLS) – same as above, except uses

linkage analysis based on families or trios.
Genomic imprinting – see Genetic imprinting.
Genotoxic stress – damage to the genome of an organism attributable to a toxin.
Genotype – a combination of two or more alleles in an individual at a single locus or
multiple loci (e.g. a haplotype). Individuals are either heterozygous (carrying both alleles)
or homozygous (carrying two copies of one allele) for an individual gene.
Genotype portfolio – the sum of genetic data held on or by an individual (may also be
considered under Risk portfolio).
Genotyping – identification of genes or alleles individually or in groups using a variety
of methods.
Glial cells – small cells in the brain involved in the repair of neurons.
Golgi stacks – the Golgi apparatus is a membranous organelle where proteins and lipids
are sorted for transport within the cell and where newly synthesized HLA molecules are
held prior to their expression at the cell surface.
Gram-positive bacteria – bacteria that stain dark blue, purple, or violet when stained
with Gram stain.
Haploid – a cell with one single copy of each chromosome (usually a gamete).
Haplogroups – formation of a combination of different alleles into haplotypes and hap-
lotype families; this is a useful tool in analysis. Haplogroups and haplotypes are not always
exactly the same – haplogroups is a more relaxed broader term.
Haplotype(s) – a series of alleles at linked loci that are inherited on the same chromo-
some, usually in strong linkage disequilibrium.
Haploview – commonly used bioinformatics software designed to identify patterns of link-
age disequilibrium, estimate haplotype frequencies, and help select tagged SNPs for analysis.
HapMap – map of common human haplotypes created as part of the aftermath of the
Human Genome Project.
Hardy–Weinberg equilibrium (HWE) – the equation most often used to determine
whether observed values depart from the expected norm.
Hardy–Weinberg principle (HWP) – the simple principle based on the relationship
between allele frequencies and genotype frequencies that is found in many populations,
and is broadly used to verify studies in complex disease.
HeliScope sequencer – pioneered as an innovative rapid sequencing technology, it relies
on true single-molecule sequencing without a preclonal amplification step.
Heritability (heritable component) – the proportion of a trait that is inherited it is not always
genetically determined, but many heritable traits have a considerable genetic component.
Heritable trait – any inherited characteristic determined by genes or other heritable
factors.
Heterogeneous – a mixed genotype where both alleles are present–one from each parent.
If there is a single variation (i.e. two alleles at a locus), one may be the original (wild-type)
and the other may be a mutant. Where there are multiple alleles, this type of naming is
less easy to apply.

2967_Glossary.indd 380 07/07/2015 11:09

 Glossary 381

Heteroplasmy – mosaicism, usually in a single cell.

Heterozygote advantage – a situation that can occur when an individual or individuals in
a population who are heterozygous for a specific genotype have a reproductive or survival
advantage over those who are homozygous.
Heterozygotes or heterozygous – individuals who carry different alleles for the same
gene on each of the parental chromosomes (see also Genotype).
Histone(s) – a family of small basic molecules that complex with DNA to form
nucleosomes.
Histone acetylation – histones are acetylated on the core of the nucleosome as part of the
normal process of gene regulation.
Histone modification – histones can be modified by a number of biochemical reactions
usually involving methylation or acetylation (see Histone acetylation).
HIV escape epitopes – immunodominant epitope in HLA-B*27 that is associated with
changes in HIV viral loading.
HLA – abbreviation for human leukocyte antigen; named “leukocyte antigens” because
they were first described expressed on leukocytes, but not found on red blood cells.
HLA 7.2 (ancestral haplotype) – see Haplotypes(s), also Linkage Disequilibrium (LD).
HLA 8.1 (ancestral haplotype) – see Haplotypes(s), also Linkage Disequilibrium (LD).
HLA association – genetic associations with HLA. See Association(s).
HLA antigen-binding groove – the site at which HLA molecules bind antigenic peptides
for presentation to the T cell receptor in the formation of the immune synapse.
HLA class I (loci) – a distinct group of HLA genes encoding a series of molecules with a
similar structure and function. These molecules present short antigenic peptides to the T
cell receptor of CD8+ T cells in the formation of the T cell synapse. The use of the term
loci indicates that there are several HLA class I genes.
HLA class II (loci) – a distinct group of HLA genes encoding a series of molecules with
a similar structure and function. These molecules present short antigenic peptides to the
T cell receptor of CD4+ T cells in the formation of the T cell synapse. The use of the term
loci indicates that there are several HLA class II genes.
Homoplasmic – a cell where all the copies of the mitochondrial DNA are identical.
Homozygotes or homozygous – individuals where the gene or genes of interest carry the
same allele on both chromosomes (see also Genotype).
Hot spot positions – recombination hot spots are areas where the frequency of recombi-
nation is greater than expected. Recombination occurs naturally during prophase in the
first phase of meiosis when homologous chromosomes bind and exchange DNA segments.
Human Genome Map (HGM) – the first map of the positions of the majority of the
human genome was published in 2001 in Nature and Science.
Human Genome Project (HGP) – the term describing the entire genome project, includ-
ing mapping the genome, SNP identification (SNP Map), HapMap, and, more recently,
the development of next-generation sequencing (NGS) and the 1000 Genomes Project.
Human leukocyte antigen (HLA, HLA typing) – these are the protein products of the
HLA class I and class II genes in the MHC. HLA typing refers to the use of serum in a

2967_Glossary.indd 381 07/07/2015 11:09

 382 Glossary

microcytotoxicity assay to identify the antigens on the surface of lymphocytes prepared

from whole blood. This was normal during the 1970s to the late 1990s; in the late 1980s
and through the 1990s, genotyping with molecular-based methods started to take over.
Most laboratories now use molecular genotyping.
Humanized antibodies – modification of antibodies produced by non-human species to
increase their similarity to natural human variants (see Chimeric).
Hypothesis constrained (hypothesis driven) – research that is based on testing a specific
idea.
Hypothesis generating – research that is likely to generate ideas for testing (see also
Hypothesis non-constrained).
Hypothesis non-constrained (hypothesis free) – research that is open to all possibilities
and is not set up to test a predetermined idea.
iCOGS – a custom-designed Illumina array designed to test genetic variation in three
major cancers: breast, ovarian, and prostate cancer.
Identical by descent (IBD) – alleles in an individual are identical because they are
inherited from a common ancestor. In this case, the exact same allele and haplotype is
inherited.
Identical by state (IBS) – alleles in an individual appear to be identical to those of another
individual but there is no common ancestor. In this case, the allele is likely to be passed
through the family from different parents.
Illumina genome analyzer – produced by the Illumina Corporation. A short-read
sequencing platform like SOLiD. It is currently very widely used.
Immune regulatory pathway (genes) – biological pathways involved in turning the
immune response on and off (regulation). A pathway will involve the interaction of a
group of different proteins with different functions encoded by a range of different genes.
Immune stress – situation when the immune system is placed under stress such that
systems begin to be overwhelmed by demand and breakdown can occur with potentially
severe consequences.
Immune synapse – region of contact between receptor and ligand – in this book most
often used in relation to HLA peptide–T cell receptor interaction.
Immune tolerance – a situation in which the immune system does not respond to an
antigen (i.e. tolerate the antigen). This can apply to self-antigens and non-self-antigens.
Immunogenetics – branch of immunology that is concerned with the study of genes that
encode proteins involved in immunity and immune regulation.
Immunoglobulin domain – part of a protein structure composed of approximately 100
amino acids folded into β-pleated sheets and α-helices. IgG-like domains are found in
HLA class I and class II molecules.
Immunohistochemistry – the use of chemical staining by antibodies (often monoclonal
antibodies) to identify antigens in tissues (see also Fluorescence in situ hybridization).
Imprinting – the determination of the expression of a gene by its parental origin.
Imprinting control elements – a short sequence within an imprinted gene cluster where
methylation controls the imprinting status of the genes.

2967_Glossary.indd 382 07/07/2015 11:09

 Glossary 383

Impute (imputed, imputation analysis) – literally meaning to attribute or ascribe. In the

context of the genetics of complex disease, it means to assign genotypes based on a known
pattern of linkage disequilibrium. This method will back-fill missing data for which geno-
typing was not performed using the expected genotype at each locus as suggested by the
data for the genes that were fully genotyped, and the known and expected genotype com-
binations in other databases.
Inbreeding depression – when inbreeding occurs there is a higher frequency of identity
by descent (IBD) compared with identity by state (IBS).
Incidence – the number of new cases of a disease in a study population over a set period
of time (see also Prevalence rates).
Incomplete penetrance – the co-occurrence of an allele and a trait. When an indi-
vidual possesses the allele, but has not developed the trait, then there is “incomplete
penetrance.”
Incorrect segregation – a process that can lead to aneuploidy, altering the number of
chromosomes in the cell (see Aneuploidy).
Independently assorted – according to Mendel’s laws, traits are passed independently
from parents to their offspring, thereby being independently assorted.
Index case – first reported case within a family.
Indirect associations – see Synthetic associations.
Individual missingness – missing data in analysis. Usually due to poor SNP selection in
GWAS where a specific SNP has failed to give an adequate signal. This leaves a hole in the
data set – such holes can sometimes be back-filled using imputation. It can also refer to
failure of a sample from one of the test groups to work in a GWAS; the “missing” data in
this case cannot be back-filled (also relates to sample call rate).
Insulin – product of pre-pro-insulin and pro-insulin produced in the islets of Langerhans
in the pancreas; regulates glucose metabolism.
Intercellular neurofibrillary tangles (NFT) – aggregates of the hyperphosphorylated tau
proteins that occur in Alzheimer’s disease. Tau is a microtubule-associated protein and
there other diseases where this may occur.
Interferons – a family of cytokines with a variety of functions acting on a variety of dif-
ferent cell types depending on function. Interferons are key agents in both innate and
adaptive immunity, and also play an important role in immune regulation.
Intermediate metabolizers – a group who express the intermediate phenotype based on
their individual genotype for a gene that metabolizes a specific drug.
Intron(s) – the space between the protein-coding regions of the DNA sequences
(known as exons). Intronic sequences have a range of roles, not all of which are fully
understood.
Invariant chain – polypeptide that associates with HLA class II molecules in the endo-
plasmic reticulum where it acts as a molecular chaperone to prevent inappropriate peptide
binding until the HLA molecule can be transported to the endosomes where it is degraded
and peptides can compete for binding.
Inversions – a re-arrangement where a segment of a chromosome is reversed from end
to end.

2967_Glossary.indd 383 07/07/2015 11:09

 384 Glossary

Killer cell immunoglobulin-like receptor (KIR) – family of receptors on natural killer

cells that bind HLA class I on target cells and either activate or downregulate the natural
killer cell.
Kinship coefficients – a measure of consanguinity between individuals in complex dis-
ease studies; this can be used to estimate bias.
Least-squares criterion – statistical method of looking for errors in a study and cor-
responds with the maximum-likelihood ratios if the experimental error has a normal
distribution.
Leber hereditary optic neuropathy – disorder inherited from the mitochondrial DNA
causing degeneration of retinal ganglia cells and their axons.
Leukocyte (antigens, typing, antibodies) – white cells. Antigens refers to proteins on
white cells; typing refers to phenotyping using serological testing; antibodies refers to the
agents required to perform phenotyping.
Leukotriene – a family of inflammatory mediators produced by leukocytes and possibly
some other cells. They are involved in immune inflammation and vascular adhesion.
Linear regression model – mathematical model used in statistical analysis to define a
relationship between a dependent variable (on the y-axis) and one or more other explana-
tory variables (on the x-axis).
Linkage – a physical link between two or markers on a chromosome (see Linkage
­analysis). It can also refer to a link between a genetic marker and disease or disease trait.
Linkage analysis – one of two major strategies for identifying “disease alleles” based on the
likelihood of the marker alleles, mutations, copy number variants of a gene or genes being
inherited with the trait. Linkage analysis is mostly based on informative families, though
occasionally sibling pairs are used, and it works well for diseases that present in early life,
have high penetrance, and a near-Mendelian pattern of inheritance. It does not work in dis-
eases that are transmitted horizontally through the population, such as infectious diseases.
Linkage disequilibrium (LD) – the co-occurrence of alleles from two or more gene loci
more often than expected by chance [see also Ancestral haplotype (HLA 7.2 and 8.1)].
For example, two alleles each present at 10% in a population should be found together at
a frequency of 1% if there is equilibrium. However, when there is “dis”-equilibrium the
alleles are likely to be found together more often (or less often) than expected by chance.
In this latter case there is non-random segregation of the alleles.
Locus (plural: loci) – the location (or site) of a gene in the genome.
Locus heterogeneity – variation at a locus that may exceed the normal biallelic variation.
LOD score – log of odds (a statistical value usually to base 10). This is a measure of the
degree of linkage between a marker gene and a trait. It is used in linkage analysis.
Logistic regression (model) – statistical method used for determining the outcome of
categorically dependent variables.
Logit – a mathematical test. There are various forms of this test.
Lymphatic system – a circulatory system comprised of a network of vessels through
which lymph (a fluid derived from blood plasma is transported). The lymphatic system is
not a closed system, and there is free movement of plasma into the system from the blood
as and when required.

2967_Glossary.indd 384 07/07/2015 11:09

 Glossary 385

Macrophage(s) – large mononuclear phagocytic cells important in innate and adaptive

immunity.
Major histocompatibility complex (MHC) – region of approximately 7.2 Mb of DNA
on chromosome 6p21.3 encoding over 252 expressed genes, including the HLA genes.
HLA matching can have a major effect on the survival of transplanted tissues, though
other factors are also important. The term MHC should be used only to identify the
region and sub-regions within it.
Manhattan plots – a plot of probability values for SNPs tested across the genome in a GWAS
or GWLA study. In these plots, data for each SNP is entered chromosome by chromosome 1
to 22, and then X and Y. The individual dots for the low P value SNPs merge to give colored
squares at the foot of each chromosome column, but the SNPs above the significance thresh-
old are shown in a bright rising column. It is not unusual to see P ­values of less than 10−30.
Marker locus – a locus that is used as a trait marker in a linkage or association study. The
locus may or may not encode a gene of potential interest or may be in close proximity to
another more likely candidate. However, some candidates are more difficult to genotype
than others and so surrogate markers are selected that are known to have tight linkage with
alleles on the gene of interest.
Mature insulin – the product of pro-insulin produced in the Golgi apparatus of the β
cells of the islets of Langerhans.
Maximum-likelihood estimation – a statistical method used to determine the param-
eters of a model from a data set.
Meiosis – a special form of cell division that gives rise to gametes.
Mendelian autosomal dominant – a single copy of the disease-causing mutation causes
the disease.
Mendelian autosomal recessive – two copies of the disease-causing mutation are required
to cause the disease (one from each parent).
Mendelian diseases/disorders – see Mendelian.
Mendelian monogenic – Mendelian diseases by definition involve a single gene, though
different individuals with the same disease may have different disease-causing genes or
gene variants (alleles or mutations) (disease-causing mutations). There are several patterns
of Mendelian disease, referred to above.
Meta-analysis – the analysis of combined studies. This is performed to increase the statis-
tical power of genetic studies and identify variables not identified in initial testing.
Metagenomics – genomic studies on a large scale; used particularly in the study of the
bacterial genome and in the Microbiome project.
Metastasis – the ability of cancer or a tumor to migrate from one region of the body to
another.
Methylation chips – bead-chip technology currently being used to identify all of the
methylation sites in the human genome.
MHC class I chain (MIC)-related (MICA, MICB, MICC, MICD, and MICE) – a group
of stress-induced molecules that are expressed on the surface of epithelial cells in response to
infection or damage – expression is recognized by the NKG2D receptor on natural killer
cells. MICA and MICB are functional MICC, MICD and MICE are essentially pseudogenes.

2967_Glossary.indd 385 07/07/2015 11:09

 386 Glossary

Microbiome Project – launched in 2008 to characterize the genetics of bacterial com-

munities associated with health and disease in the human body.
Microcytotoxicity or microlymphocytotoxicity test – assay to identify different HLA
phenotypes using lymphocytes from whole blood and an array of different polyclonal anti-
bodies from human serum. The assay used complement to induce lysis where the lympho-
cytes had interacted with the antibodies. Toxicity was assessed by the use of a fluorescent
dye. The early assays were refined to allow micro-scale testing in 1964 and again in 1978.
Following the 1978 modification, the suffix NIH was added to the name. The assays were
performed on microtiter trays of 60, 72, or 96 wells each approximately 10 μl deep.
Microsatellites – short number tandem repeat sequences. These variable number
­tandem repeats (VNTRs) are short DNA sequences that are repeated in the DNA
sequence. Individuals inherit different numbers of repeat sequences and these can be used
to genotype regions of the genome for genetic associations.
Minor allele frequency (MAF) – alleles with low frequencies. A minor frequency can be
set at any limit provided the numbers in the study are large enough. Early GWAS based on
studies of 2000 cases and controls tended to use a 5% cutoff, but with studies using case
and control series sometimes 25 times larger (i.e. as many as 50,000 cases for example)
than the 2000 previously used, limits are now being set at 1% or even less. It is all a matter
of setting appropriate statistical thresholds.
Missing – genotypes that are not determined in analysis or are not identified in the origi-
nal test. The reason for these failures is often unknown, but when there are large numbers
it can impact on the study. However, if there are not too many they can be assigned
through imputation in some cases.
Mitochondrial diseases – diseases and traits that have been mapped to mutations and
polymorphism in the mitochondrial genome.
Mitochondrial genome (mtDNA) – the DNA encapsulated in the mitochondria (see
Nuclear genome).
Mitosis – process during embryonic development in which there is constant production
of new cells. These new cells are made by rounds of cell division called mitosis and the
daughter cells are genetically identical to the parental cell.
Mixed lymphocyte culture (MLC) – early method for HLA class II typing.
Molecular clock – this term refers to a specific hypothesis used in studies of evolution.
Essentially if evolution is consistent and there is a constant rate of divergence between two
sequences then this hypothesis can be applied to identify the likely time at which species
diverge from a common ancestor.
Molecular signatures – this term has a number of uses. It refers to G-coupled proteins
and is also used in discussions of natural selection. Both are relevant here.
Monocytes – blood cells with a single nucleus.
Monoclonal antibodies – antibodies that are mono-specific; made from cloned cells that
are identical.
Monogenic – involving a single gene.
Monosomy – having one copy.
Monozygotic twins – identical twins created from the same gametes. Also see Dizygotic
twins.

2967_Glossary.indd 386 07/07/2015 11:09

 Glossary 387

Morbidity – something affecting the quality of life, usually associated with some tissue
damage or loss of function.
Mosaic mutations – a mosaic mutation is one that does not occur in every cell (see
Mosaicism).
Mosaicism – situation when only some of the cells in the body possess the mutant gene. It
can also apply to a situation when only a proportion of the mitochondria in a cell possess
the mutant gene. This is especially relevant in cancers.
Multicellular eukaryotic organisms – an organism that has both nucleated cells (i.e. each
cell has a nucleus – eukaryote) and is composed of more than one cell (multicellular).
Multiple linear regression model – attempts to model the relationship between two or
more explanatory variables and a response variable by fitting a linear equation to observed
data.
Multihit hypothesis – the idea that a haplotype may carry more than one risk allele–thus
common haplotypes may have several “hits” perhaps accounting for the strong association
with certain haplotypes (e.g. HLA 8.1).
Multipoint mapping – this may involve actual genotyping on a large scale or imputation
(see Impute) on a large scale (i.e. multipoint imputation).
Multiple sclerosis – an inflammatory autoimmune disease in which the myelin sheath
that insulates the nerves is degraded, leading to a variety of symptoms.
Mutation – technically, the majority of genetic variation arises as a result of mutation
irrespective of frequency. However, in clinical genetics some use the term mutation when
genetic variation gives rise to a disease (i.e. it is a disease-causing mutation). Others pre-
fer to use the term for genetic variations that are found at a low frequency in the popula-
tion (usually less than 1%). Genetic variations at a frequency of greater than 1% tend to
be called polymorphisms. Both terms are used interchangeably in places in this book, but
polymorphism is preferred in complex disease studies.
Myelin sheath – a material that insulates the electrically charged neurons and is broken
down in multiple sclerosis.
Myopathy – general term used to describe muscle pain or adverse reactions involving the
muscle.
Myositis – general term for the inflammation of the muscles which can be common in
some autoimmune diseases.
Natural killer (NK) cell – large granular cytotoxic lymphocytes, particularly important
in innate immunity.
Natural selection – Darwin’s major work; the idea (or hypothesis) of evolution driven by
natural selection.
New Germania – a region in Paraguay where a group of German settlers set up a com-
munity, with a view to maintaining race purity and thereby improving the breeding stock.
Originally suggested and supported by Wagner and Elisabeth Nietzsche, the sister of the
famous philosopher.
Next-generation sequencing (NGS) technologies – technologies used to sequence DNA
in ever-shorter periods of time. This does not refer to one technique, but to all of the
recent and some of the future developments to come.

2967_Glossary.indd 387 07/07/2015 11:09

 388 Glossary

Non-Mendelian complex diseases – diseases that do not have a known pattern of inheri-
tance, but clearly have a heritable (genetic) component.
Non-parametric linkage analysis – linkage analysis in which the pattern of inheritance
is not predetermined or preset. This can be used for linkage analysis in complex disease.
Non-random segregation – the distribution of allelic sequences between daughter cells at
meiosis. Segregation should be random, but it is not due to the presence of hot spots and
cold spots in the genome.
Non-synonymous SNPs – changes in the DNA sequence that result in amino acid dif-
ferences. Unlike synonymous (conservative mutations) these SNPs do have a biological
effect. See Single nucleotide polymorphism.
Novel somatic mutations – new mutations such as those that occur in cancer. It is pre-
dicted that next-generation sequencing (NGS) techniques will have a greater ability to
detect these.
Nuclear genome – the DNA encapsulated in the nucleus (i.e. the 22 pairs of chromo-
somes plus X and Y); not the mitochondrial DNA.
Nucleosome – see the Nuclear genome.
Null hypothesis (H0) – the null hypothesis states that there is no difference (null) between
the tested groups or populations (see Alternative hypothesis).
Neutrophils – granular white cells. They often have segmented nuclei and are the most
abundant of all white cells in the body. They are important in innate immunity, in particular.
Odds ratio (OR) – a crude measure of relative risk, though more commonly used than
relative risk (see Relative risk).
Offset sequencing primer strategy (OSPS) – part of the SOLiD short-read sequencing
method. This is illustrated in Figure 12.5.
Oligogenic – many forms (oligo) of a gene (genic). In complex disease this term refers to
a disease or subgroup of patients with a disease where there is more than one susceptibility
allele, but the number of susceptibility alleles is limited. There is no numerical value that
can be applied to the term “oligo” (see also Polygenic).
OMIM (Online Mendelian Inheritance in Man) – a comprehensive and free-to-access
database of human genes and phenotypes.
OncoArray – a new chip array that is designed to carry 500,000 tagged SNPs for genotyp-
ing patients with cancers, including breast, ovarian, prostate, colorectal, and lung cancers
(see also iCOGS Illumina chip).
Over-dominance – heterozygous advantage, where the heterozygote is at greater advan-
tage than the homozygote.
Pair-wise kinship estimates – a measure of the level of genetic similarity with a kinship
group and indicator of the level of inbreeding in ancestral populations.
Pair-wise probability (of identity by descent) – this test would be performed using
PLINK.
Paired-end mapping – pair-ended sequencing or mapping is increasingly used to access
genome rearrangements and structural variations on a global scale. Using this technique, a
single DNA fragment is sequenced from each end, producing two reads that can overlap.

2967_Glossary.indd 388 07/07/2015 11:09

 Glossary 389

Paneth cells – major cell type in the epithelium of the small intestine.
Paracentric inversions – a stable outcome after the incorrect repair of two breaks in a
single chromosome. Here, the fragments rejoin in reverse order and there is no loss of
genes, only a change in the sequence – so this is an inversion. Paracentric inversion occurs
when the repair does not involve the centromere.
Parametric linkage (analysis) – linkage analysis where the parameters are preset (or
known), i.e. the pattern of inheritance has been proposed.
Pathology – the science of seeking, observing, and recording normal and damaged tissues
in medicine. In practice, it is the detectable changes in and damage to the tissue caused
by a disease.
Pearson’s χ2 test – the most frequently used form of the χ2 test first applied in 1900, and
later developed by Yates and Fisher. The test is based on the normalization of the sum of
the squared deviations between observed and expected results.
Pedigree file – a collection of family data.
Penetrance – a measure of the occurrence of the trait in individuals with the risk allele or
vice versa the risk allele in those with the trait.
Pericentric inversions – occur when two breaks occur at different ends of the same chro-
mosome and the two fragments created rejoin the terminal fragments. The gene sequence
along the chromosome is altered, but the overall content remains the same.
Permutation procedure – permutation procedures provide a computationally intensive
approach to generating significance levels empirically.
Personal profile – idea that we each inherit a personal profile of risk alleles for complex
disease (see also Disease portfolio).
Phagocytosis – internalization of matter from outside the cell by specialized cells called
phagocytes.
Pharmacogenetics – the study of genetic variation and response to different pharma-
cological agents, including adverse drug reactions. This term was first used in 1959 by
Freidrich Vogel.
Pharmacogenomics – the study of the whole genome from a pharmacological perspective
(see Pharmacogenetics).
Phase (phasing) – a method that can be used to determine which SNPs are inherited
together.
Phenotype – the result of genetic variation; this can be seen in the expression of a trait,
character, or disease (disease phenotype).
Phenotyping – the observation and recording of the expression of a trait.
Philadelphia chromosome – this results from an unusual translocation between chromo-
some 9 and chromosome 22. It is found in 95% of patients with chronic myelogenous
leukemia and 25% of patients with acute lymphoblastic leukemia.
Phylogenic tree – a diagram showing variation between species and populations.
pi-hat – a statistical test that can be performed in PLINK.
PLINK software – open-source GWAS toolkit designed to perform basic large-scale anal-
ysis: http://pngu.mgh.harvard.edu/purcell/plink/.

2967_Glossary.indd 389 07/07/2015 11:09

 390 Glossary

Ploidy – number of copies of the autosome (i.e. the chromosome set).

Polar body – the byproduct of uneven division of egg cells. After cell division, the polar
body will apoptose or die.
Polygenic – involving several genes. (See also oligogenic.)
Poly(ADP) ribose polymerase inhibitors (PARPs) – a family of proteins associated with
DNA repair and programmed cell death.
Polymerase chain reaction (PCR) (analysis) – a commonly used method for genetic
studies involving the amplification of short DNA sequences.
Polymorphic – multiple (poly) forms (morph).
Polymorphism(s) – existence of many (poly) forms (morphisms). Strictly speaking, this
should only be used if the least common variant/allele is found in more than 1% of
the population under study. However, as with mutations, not all accept this definition
even though it is widely used in complex disease. A further complication is that some
genes have so many variants/alleles that this strict application of the term polymorphism
becomes meaningless in relation to some genes.
Polyploidy – many copies of the autosome (see Ploidy).
Poor metabolizer – genetic variation that encodes a specific phenotype for drug
metabolism.
Population – populations can be defined in geographical terms or in terms of study
groups. In the context of this book, a population is most often a group of individuals with
shared characteristics.
Population genetics – the study of genetics in populations.
Population risk – the risk within a population. Sometimes termed population
­attributable risk (PAR), which epidemiologists use to assess the contribution of a factor
to the overall incidence of a trait.
Population stratification – the subdivision of a population.
Population vigor – the health of a group or population.
Positional cloning – identifying a gene using knowledge of its chromosomal location.
This mechanism has been widely used to identify genes responsible for the vast majority
of Mendelian disorders.
Positive (Darwinian/natural) selection – selection that favors reproductive success, sur-
vival, or increase in advantageous traits within a population.
Power calculation (or power) – the power referred to here is “statistical power”. By ensur-
ing there is adequate statistical power we can reduce the likelihood of false-negative asso-
ciations in a case control association study or in a GWAS.
Pre-pro-insulin – primary translational product of the INS gene (insulin gene). It is pro-
duced in a biologically inactive form and converted to pro-insulin.
Prevalence rates – the total number of cases expressing a trait at a particular time point
(see also Incidence).
Primary immune deficiency – immune deficiency occurring as a result of inherited traits
manifest from birth.

2967_Glossary.indd 390 07/07/2015 11:09

 Glossary 391

Principal components analysis (PCA) – a mathematical procedure used to convert a set

of possible correlated observations into a set of linearly uncorrelated observations.
Probability values (P) – the statistical measure of the likelihood of an event occurring.
Pro-drug – a drug that is produced in a pre-active form and converts to the active form
after administration (e.g. codeine).
Product rule – a statistical rule based on calculus that is used to identify the derivative of
the products of two or more functions.
Pro-insulin – precursor to insulin made in the β cells of the islets of Langerhans. Pro-
insulin is converted to mature insulin in the Golgi apparatus. Also see Pre-pro-insulin.
Proteasome – a large multi-subunit protease found in the cell. The proteasome degrades
proteins in the cytoplasm, generating smaller peptide units that are suitable for binding
and presentation by HLA class I molecules.
Protein – an organic compound containing carbon, hydrogen, oxygen, nitrogen, and
other elements.
Pulmonary fibrosis – fibrosis and scarring of the connective tissue of the lungs, often
secondary to other diseases.
Punnett Square – a simple table of data used in the χ2 test.
Purifying (natural) selection – selection against unfavorable alleles or genotypes.
Quantile–quantile (Q–Q) plot – a probability plot used to compare two probability
distributions using their quantile values.
Quantitative trait – a universal character rather than a character that some individuals
have and others do not.
Quantitative trait locus – locus that contributes to the phenotype. Such a locus encodes
one or more genetic variants to allow for this phenotypic variation.
Random genetic drift – random changes in gene frequencies over generations.
Reactive arthritis – autoimmune response that follows from an infection episode; swell-
ing in the joints is the classical symptom.
Recessive trait – an inherited trait whereby both parental haplotypes carry the same geno-
type (mutation or polymorphism). In other words, two copies are required for the trait
(disease or characteristic) to be expressed – one of Mendel’s patterns of genetic inheritance.
Recombination – natural process that occurs during prophase in the first phase of meiosis
when homologous chromosomes bind and exchange DNA segments.
Recombination cold and hot spots – recombination occurs naturally during prophase
in the first phase of meiosis when homologous chromosomes bind and exchange DNA
segments. A single recombination event produces two recombinant and two non-recom-
binant chromatids. Double crossovers can occur, but the overall average is 50%. There are
said to be cold and hot spots at which the frequency of recombination is either higher than
expected (hot spots) or lower (cold spots).
Recombination fraction (values) (θ) – the measure of recombination, which never
exceeds 50% (or 0.5).
Recruitment bias – the potential introduction of bias into cohorts under study.

2967_Glossary.indd 391 07/07/2015 11:09

 392 Glossary

Reductionist – taking apart an idea–dividing a question into its constituent pieces–often

involves over-simplifying an idea or task.
Redundancy – a mechanism by which biological systems protect their host from disaster.
This occurs through the use of multiple pathways for the same function, gene duplication,
and a host of other mechanisms.
Regression (analysis) – a statistical process for estimating relationships between variables.
In complex disease this means genes and traits or populations.
Regulatory cytokines – cytokines that regulate the immune response. This term usually
refers to those cytokines that switch off the immune response rather than those which
promote inflammation, such as tumor necrosis factor-α.
Relative risk (RR) – a measure of the size of effect of an allele in terms of increased or
reduced susceptibility to a trait. This is often presented as an odds ratio (OR), but these
are not exactly the same.
Restriction enzyme – an enzyme that cuts at a specific DNA sequence. These enzymes
have been used extensively in genotyping studies.
Restriction fragment length polymorphism (RFLP) (analysis) – one of most com-
monly used methods to identify genetic variation and genetic associations. Alleles can be
assigned based on the pattern of fragments identified by electrophoresis after digestion
with an enzyme which cuts at a specific sequence (i.e. is restricted). Most (but not all)
methods require amplification of the DNA sequence of interest by polymerase chain
reaction (PCR) prior to digestion with the enzyme–this increases the level of target DNA
in the sample and gives a clearer signal.
Ribosomal RNA – there are two mitochondrial rRNA molecules (12S and 16S) and
four cytoplasmic molecules. rRNA was not included in the Human Genome Mapping
Project (HGMP) and therefore we are currently unsure of the total number of rRNA
genes in the genome.
Ring chromosomes – potential outcome following incorrect repair of a chromosome.
The ring chromosome is a stable chromosome and can propagate to other daughter cells.
Risk (λ) – see Sibling relative risk.
Risk allele(s) – a variation in a gene that increases; is associated or linked with an increased
likelihood of a trait occurring.
Risk portfolio – an individual’s personal genotype data for susceptibility and resistance
alleles.
“rs” number – location of a SNP; true name: refSNP cluster ID number.
Salt bridge – a biochemical structure or a molecular bridge between two locations with
different electrostatic properties.
Sample call rate – the number of samples working (i.e. giving results) in an assay. Used
in PLINK.
Sclerotic plaques – these occur in a number of diseases and involve the formation of
fibrous plaques on the walls of different tissues that disrupt normal function.
Segregate – the distribution of allelic sequences between daughter cells at meiosis.
Segregation analysis is used to determine the likelihood that a child will inherit a trait
from a parent.

2967_Glossary.indd 392 07/07/2015 11:09

 Glossary 393

Selective pressure – any factor that reduces reproductive success within a population.
Environmental factors include infectious agents, e.g. the malaria parasite.
Sequencing-by-synthesis – developed by 454 Life Sciences using what is called a pol-
ony sequencing protocol. This was an early form of next-generation sequencing (NGS)
­moving towards rapid methods for DNA sequencing. See also SOLiD.
Semi-dominant – same as co-dominant; heterozygotes have an intermediate phenotype
compared with homozygotes.
Serotyping (serological typing) – HLA typing using antibodies in human serum to iden-
tify specific HLA antigens. Multiple sets of microtiter trays with 60, 72, or 96 wells are
used to determine each phenotype at varying degrees of resolution.
Sex-linked – Mendelian disease involving the two sex chromosomes X or Y.
Shared epitope – said to occur when two or more alleles share a sequence that encodes
amino acids in a region of functional importance within a molecule. It is most frequently
applied to studies of HLA to explain why there are often several different HLA alleles
associated with susceptibility to the same disease.
Sibling relative risk (λ) – a crude calculation of the increased risk of disease that a sib-
ling of an affected case is likely to have. The calculation is based on a comparison of the
incidence or prevalence in cases versus the healthy population (controls). The quality of
the result depends on the size of the population surveyed and the frequency of the trait.
Significance threshold – the preset measure at which a result will be considered as signifi-
cant. For example, in the WTCCC1 study any numerical value lower than the significance
threshold is significant (i.e. less than 5 × 10−7) and any value higher is not significant.
Single nucleotide polymorphism (SNP) – self-defined units, variations in single nucleo-
tides commonly polymorphisms used extensively in genetic studies.
Slow acetylators – a group of individuals who inherit genetic variation that causes them
to express this phenotype.
Small call rate – the number of samples that are successfully genotyped in a study.
SNP – see Single nucleotide polymorphism.
SNP Map – a map of the positions of all SNPs in human genome.
Social Darwinism – Social Darwinists generally argue that the strong should see their
wealth and power increase, while the weak should see their wealth and power decrease.
The precise definition of who belongs in either group is a matter of debate amongst those
who hold these outdated beliefs.
SOLiD (Supported Oligonucleotide Ligation and Detection) – a next-generation
sequencing (NGS) platform developed by Life Technologies. Performs short-read
sequencing based on “sequencing-by-ligation” (see also Sequencing-by-synthesis). It also
uses the offset sequencing primer strategy.
Somatic chromosomal abnormalities – abnormalities in any chromosome found in a
somatic cell.
Statistical confidence – the level of acceptance of a probability value. Similar to the sta-
tistical threshold, but can be expressed as a range of confidence intervals.
Statistical error (type I and type II) – errors in statistical calculation caused by low
sample size causing false positives (type 1) and false negatives (type 2). These are not errors

2967_Glossary.indd 393 07/07/2015 11:09

 394 Glossary

in the application of the statistical test, but errors introduced in the planning of the study
that are picked up through statistical analysis (see Statistical power).
Statistical power – in the context of this book, the likelihood of finding true-positive
associations and reporting true-negative associations.
Stratification score analysis – measure of the degree of stratification within a population.
Structural abnormalities – a missing, extra, or irregular fragment of DNA.
Structured association – relates to population stratification.
Sum rule – a complex mathematical calculation used in genetic studies; best left to the
statistical geneticist.
Super-coiled – the multiple rounds of coiling of DNA strands.
Survival of the fittest – concept based on Darwin’s theory of evolution. The term was first
coined by Herbert Spencer.
Synthetic association – low-frequency causal variants. A form of indirect association.
Tagged SNPs – selected single nucleotide polymorphisms (SNPs) used in GWAS in
particular. Each SNP is selected based on the likelihood of amplification during testing
and proximity to a marker gene.
TAP – abbreviation for transporter associated with antigen processing.
T cells – white cells or leukocytes that mature and undergo selection in the thymus.
T cell receptor (TCR) – the receptor for peptide antigens on mature T cells. A critical
element in the formation of the T cell synapse.
Telomerase – an enzyme that adds DNA repeat sequences at the 3′ end of the DNA mole­
cule in the telomere region.
Telomeric/telomere – the end of the chromosome.
Tetraploidy – possession of four copies of one or more chromosomes. All cases are lethal.
Thalassemias – a group of inherited autosomal recessive blood disorders.
Thrifty gene hypothesis – hypothesis proposed by JV Neel in 1962 to explain the grow-
ing incidence of diabetes in the Western World.
Thymus – area of the body where T cells develop, mature, and go through selection.
Toll-like receptors (TLRs) – receptors important in innate immunity.
Trait – a term used to describe a character or phenotype, which can be a disease, response
to treatment, subgroup within a disease, or a behavioral character in a study population.
Transcriptome – that part of the genome that encodes transcribed genes.
Trans expression – the expression of two different alleles from different loci on the same
chromosome, but from different parents. Resulting in the formation of a heterodimer in
trans as opposed to a heterodimer in cis. This can create a different molecular structure
than expected if proteins were only able to interact in cis. See Cis expression. Particularly
relevant in creating diversity of variation in HLA DQ and DP molecules.
Transfer RNA – serves as a physical link between DNA and RNA and the amino acid
sequence by loading amino acids onto the messenger RNA in peptide generation.

2967_Glossary.indd 394 07/07/2015 11:09

 Glossary 395

Translocations – transfer of chromosomal regions between non-homologous chromosomes.

Transmission disequilibrium testing (TDT) – a statistical test that can be used to iden-
tify genetic associations with candidate genes.
Triploidy – possession of three copies of one or more chromosomes.
True single-molecule sequencing – the basis of the HeliScope sequencer. It does not
require a preclonal amplification step.
Tumor genome (tumor DNA) – the genetic makeup of the tumor cells that may develop
mutations during cycles of division.
Tumor necrosis factor (TNF) – pro-inflammatory cytokine produced by macrophages
that increases the permeability of the vascular epithelium.
Tumor suppressor gene – a gene that protects a cell from developing into a cancer cell.
Twin studies – method frequently used to determine the level of heritability for a trait or
disease.
Ultrarapid metabolizers – phenotype in some individuals treated with specific drugs.
Unbalanced chromosomal abnormalities – involve increased numbers of chromosomes
(numerical abnormalities) or when there are structural abnormalities (involving deletions
and insertion or ring chromosomes, etc.).
Unicellular eukaryotic parasites – a parasite consisting of a single cell with a nucleus.
Variable number tandem repeats (VNTRs; microsatellites or minisatellites) – short
repeat sequences of DNA of variable length. The terms mini or micro are used depend-
ing on the sequence length. They have been used since the 1990s as markers for genetic
analysis.
Wahlund’s principal – reduction in heterozygosity causes changes in population sub-
structure due to genetic drift or migration.
Whole-exome sequencing (WES) (or exome sequencing) – sequencing the expressed
genes in the genome and not the whole genome. The majority of the human genome
considered to be “genetic junk” perhaps acquired through evolution. Exome sequencing
ignores this “junk”.
Whole-genome sequencing (WGS) – sequencing of the entire genome, not just the cod-
ing regions.
Whole-transcriptome shotgun sequencing (WTSS) – an RNA sequencing technique
that aims to map DNA transcripts in whole-blood samples.
Wild-type – the product of the standard, “normal” allele at a locus, in contrast to the non-
standard, “mutant” allele. However, as some genes have a large number of alleles assigning
the wild-type can be difficult. Equally, the wild-type is not always the most common allele.
X-linked dominant or recessive – Mendelian traits that are linked with genes on the X
chromosome.
Y-linked – Mendelian traits that are linked with genes on the Y chromosome.
Z score – statistical measure used in linkage analysis.
Zygote – the fertilized egg cell from which all other cells in the body are derived.

2967_Glossary.indd 395 07/07/2015 11:09

2697_Prelims.indd 2 09/07/2015 09:35
 Index
Note: Page numbers followed by F refer to Figures, those followed by T refer to Tables and those followed by
B refer to Boxes.

3′ hydroxyl group 339F, 339 Allotropic expression 53

454 Life Sciences 340, 341, 342F, 342, 348F Alpha (α) chain 179T, 185, 188T, 190T, 191, 192,
6-mercaptopurine 261, 262, 271, 288 198, 299, 300
23andMe 330 Alternative splicing 350
1000 genomes (consortium) project 237, 248, Alzheimer’s disease 16, 47, 56, 57F, 57, 67, 163,
284T, 337, 348, 353, 354, 360, 362, 363F, 325, 332, 350
366 amyloid precursor protein (APP) 57
amyloid plaques 57
β-amyloid-42 protein 58
A EOAD 47, 56, 57, 88, 111
α-helices (helices) 191, 192F, 192 fibrils 57F
Abacavir 110, 245T, 247, 255T, 267, 268T, 269, LOAD 47, 56, 57, 58, 67, 88, 111
271, 321, 322 neurofibrillary tangles 57F, 57
Acquired immunodeficiency syndrome (AIDS) PSEN1 57F, 58, 88
266, 238, 276, 321 PSEN2 57F, 58, 88
CCR2 242, 232T, 243, 244F, 249 senile plaques 57
CCR5 39, 237–245, 249 American Association for Molecular Pathology 311
CCR5-Δ32 (CCR5-Δ32 mutation/deletion) 39, American College of Human Genetics 325, 331
240–245, 254 American College of Medical Genetics 325
CCR5 promoter 242 Amino acid 117, 118, 123, 165, 181B, 185, 186T,
CXCR4 238F, 238, 29, 242, 243T, 244 191F, 191, 192F, 192, 194, 197, 198, 199T,
human immunodeficiency virus (HIV-1) 39, 199, 202, 203, 204, 205, 206F, 207, 208,
276, 321 209F, 211, 212, 248, 297, 299
RANTES 239F, 243T, 244, 245 amino acid at positions 118, 199, 198, 205,
SDF-1 239F, 243T, 244 206F, 211, 212, 299
Adaptive immunity 191, 192F, 193 214, 219, 302 amino acid sequence 117, 186T, 194, 197, 202,
Addison’s disease 216 203, 208
Additive model 156F, 157, 158, 201, 202F charged polar amino acid 203, 204
Adenosine triphosphate (ATP) 27, 51, 307, non-polar uncharged amino acid 203
340F, 343F Ancestral haplotype 9, 183T, 200, 217, 208, 209,
ADRB2 (Beta-2 adrenergic receptor gene) 264, 210, 211, 213, 215, 216, 217, 299
263, 265 Angelman syndrome 31, 49, 50T, 50
Adrenergic receptor 263, 264, 265 Ankylosing spondylitis (AK) 46, 47F, 105,
Adverse drug reactions 39, 65, 110, 201, 253, 108–110, 115, 116F, 116, 137, 176, 196,
254T, 263, 266, 267, 268T, 269, 270, 320 213, 321, 322, 325
Aerodigestive cancer 280, 286, 287 Anti-coagulation 134
Affymetrix gene chip 309, 354 Antibiotic 269
Agglutination 176 augmentin 204F
Alcohol dehydrogenase 280, 286, 287T, 287 flucloxacillin 110, 203F, 204F, 269
Aldehyde dehydrogenase 286 Antibodies 115, 116, 176, 177F, 178F, 185T, 200,
Allelic variation 2, 117, 191, 338, 367 202, 210, 290, 234, 240, 288, 289, 290
disease-causing allele 2, 8F, 8, 27, 86, 143 Antigen presenting cells (APCs) 193, 301
disease promoting alleles 6, 55 Antigenic peptides 191F, 191, 193, 194, 195F, 199,
dominant allele 21, 35 206, 210, 245, 299, 302F
mutant allele 17, 18, 21, 67 Antigenic polypeptides 194
mutated allele 8F Antigenic side chains 117, 191, 194, 212
non-disease-causing allele 2 Apolipoprotein E (APOE) gene 16
null allele 182, 185, 188F, 189T, 215, 217F, APOE 16, 58, 67, 88, 129T, 129, 130, 132T,
287T, 287, 288, 299 134, 135F
recessive allele 21, 235 APOE4 16, 58, 67
Allergy 82, 94 Applied Biosystems 342, 345F
Allison 232F, 232, 233 Ascertainment bias 122

2697_Index.indd 397 08/07/2015 09:27

 398 Index

Asthma 82, 235 C

Ataxia 36, 52T, 263, 264, 355T
Carbamazepine 267, 268T
Attention deficit hyperactivity disorder
CARDIoGRAM consortium 130, 131T 132T
(ADHD) 82
Cardiotoxity 270
Autism 39, 80T, 82, 349, 355T
Cardiovascular disease 85, 105, 115, 127–135,
Auto-reactive 210, 302F
137, 259, 295, 296F
Autoantigen 201
CD94 193
Autoimmune disease 9F, 39, 78, 82, 95, 116, 119,
CDNK2A 129, 130, 134
135, 137, 200, 202, 203F, 206F, 207, 210, 215,
CDNK2B 129, 130, 134
216, 218, 297, 299, 300, 301, 304, 312, 322
Celera 329
Autoimmune hepatitis (AIH) 76, 176, 201, 203F,
Cell growth 126T, 134, 135F, 127, 281
204F, 205T, 206F, 322
Cell necrosis 198
Autoimmunity 201, 213, 309, 312
Central nervous system 58, 123, 126T, 199
Autonomy of science 316F, 316
Centromere 2F, 3F, 41, 42, 43F
Autophagy 64–66, 91
Cerebral spinal fluid 58, 200
Autosomal 2F, 2, 10, 11, 18, 19, 31, 36, 44, 45F,
Cervical carcinoma 276, 285T, 357
48, 49, 50, 142F, 142, 143, 196, 198, 201,
Cetuximab 290
260, 355T
Charcot-Marie-Tooth disease 54, 349
Autosome 2F, 27, 31, 40, 41, 50, 51, 53, 167F
Chemokines 201, 242F
Azathioprine 261, 262
Childhood obesity 82
Chip 85F, 93, 96, 97B, 129, 309, 311, 354,
B 357, 358F
β-adrenergic agonist 266, 264, 265 Chromatin 3, 30F, 30, 123, 126T, 282T, 307,
β-pleated sheet 191, 192F, 192 357, 358F, 363
β-2-microglobulin, 179T, 191F, 191, 192, 194 chromatin immunoprecipitation assay
B cell 118, 193, 202F, 277T, 350 (CHiP-chip) 357
Basic extreme trait design 356 Chromosomal abnormalities 35, 36, 40, 41F, 42,
Beckwith-Wiedemann syndrome 31, 49, 50T 57, 58, 58F
Beta (β) chain 119, 179T, 189, 188F, 190T, 191, balanced structural abnormalities 42, 350
192, 198, 203, 206F, 208, 299, 300 Down syndrome 41
Binding groove 115, 117, 118, 192F, 193, 194, Edward’s syndrome 41
198, 199, 209F, 212, 248, 299 Patau syndrome 41
antigen binding groove 115, 191F, 191, 192, somatic chromosomal abnormalities 40
193F, 194, 198, 199, 209F, 299 Turner’s syndrome 41
peptide binding groove 198, 206F, 248F, 248 Chromosomal banding 2F
Binding pocket 198, 199, 203, 204, 205, 206F, Chromosomal breaks 41
209F, 209, 212 Chronic mylegenous leukemia (CML) 285T, 288,
Bioinformatics 341 289, 350
Biological fitness 16, 18, 21 Cis 187, 188F, 188
Biological pathways 60, 66, 74, 81, 88 Clinical expression 94
Biological systems 44, 60, 91, 241, 302, 303, 364 Clinical heterogeneity 37, 48, 54, 60
Bipolar disorder 80T, 82, 105, 115, 122–127, 137, Clopidogrel 257, 259, 260, 258T, 258, 272
288, 349 Cochran-Armitage test 152, 155, 156F, 156, 157F,
Bladder cancer 277F, 280, 284F, 285F, 285, 286, 157, 163, 164, 171
287F, 287, 291 Codeine 113, 257, 259, 258T, 258, 323, 349
Blastocyst 79F Codon 118
Blood group 142F, 142, 143, 176, 177F, 178F, Cold spots 167, 196, 362
179B, 301 Colorectal cancer 275, 278, 284F, 285T, 285, 290,
ABO blood groups 142F, 142, 176, 177F, 178F, 293, 353
179B Comparator populations 95
Carl Landsteiner 176, 177F Complement 37, 101 163, 169, 180B, 214F, 214,
James Blundell 176, 177F 215, 217F, 219 323
Bonferroni’s correction 96, 148 C2 187F, 189F, 190, 214F, 214, 215, 218
Bottleneck effect 15F, 15, 21, 28, 31 C3 214F, 214
BRAF 289 C3-covertase 214F
Breast cancer 255T, 275, 277, 278, 279F, 281, C4A 9F, 101, 190, 202, 214F, 214, 215, 217F,
282T, 282, 283, 284F, 285T, 285, 288, 290, 217, 218, 299
293, 312, 328, 350 C4B 101, 190, 214F, 215, 218
BRCA (BRCA1 and BRCA2) 278, 290, 323, C5 215
328, 330 factor B (Bf ) 163, 189F, 190, 214F, 214, 215,
Breeding population 11, 22, 26 218

2697_Index.indd 398 08/07/2015 09:27

 Index 399

Complement pathways 190, 214F Depression 85, 122

alternative pathway 214F, 214, 215 Diagnostic plot 165
classical pathway 214F, 214, 215 Dicentric 42
lectin pathway 214 Differentially spliced 2
Commensal bacteria 66, 91 Diploid 4, 10, 11, 13, 20, 40, 41F, 46B, 339
Common disease 1, 2, 32, 52, 53F, 61T, 73, 85, Direct consumer testing (DCT) 325
86, 95T, 107, 110, 114, 116, 163, 176, 261, Direct sequencing 96, 100, 118, 349
315, 326, 338, 348, 353, 359, 361, 362 Directional selection 9, 10
Common disease common variant hypothesis Disease causing allele 2, 86, 143
(CDCVH) 67, 86 Disease causing gene 68, 162
Common disease rare variant hypothesis (CDRVH) Disease causing mutation 5, 31, 45, 47, 48, 49, 53,
67 54, 56, 300, 311
Complex trait 36, 55T, 157, 250, 253, 349, DNA
356, 362 crosslinking 2
Concordance 61, 62F, 79, 80, 80T, 102, 122, deoxynucleotide triphosphate (dNTP) 339
185T, 186T, 200, 210, 211, 226, 245T, 246, deoxyribose 2
280, 281, 279 dideoxynucleotide diphosphate (ddNTP)
Confidence intervals (CI) 62T, 86, 98B 339
Conjugation 256F, 261 DNA bank 102
Conquistadors 226, 249 DNA methylation 272, 310, 311, 357, 363
Continuous inheritance 67 DNA sequence 6, 10, 30, 51, 185, 333, 338,
Copy number variation (CNV) 6, 39, 88, 100, 339F, 339, 340F, 341, 343, 345, 357, 358
215, 256, 293, 349 DNA sequencing 289, 311, 357–367
Coronary artery disease 82, 115, 127, 128F, 129T, DNA triplet 185
133T, 135F linear double stranded molecules 2
Coronary heart disease 128F linker DNA 30F
Co-segregate 84 Drug induced liver injury (DILI) 203F, 204F, 267,
Crohn’s disease 56, 57, 61–67, 78, 80T, 82, 86, 268, 269
91, 94, 110, 115, 121F, 136, 163, 297, 298T, Drug transporter 261, 263
304, 305F, 305T, 321, 367 Duchene muscular dystrophy 11
ATG16L1 gene 62, 64, 65, 66F, 66, 91 Duplication 5, 6, 41, 42T, 42, 100, 101, 254T,
CARD15 gene 61T, 62T, 62F, 62, 63, 64, 65, 257F, 259T, 350
66F, 66, 86, 91, 163
NOD2 gene 62F, 62, 63F, 63, 64, 65, 66F, 91
Paneth cells 63, 66
Crossover 7, 8, 9, 143 E
CYP2C9 258T, 259, 260, 264, 271 Electrophoresis 6, 339, 340F
CYP2C19 255T, 258T, 259, 260, 264, 271, 272 agarose electrophoresis 7F
CYP2D6 254T, 255, 256, 257F, 258T, 258, 259, agarose gel 6
272 polyacrylamide gel 6
Cystic fibrosis 2, 5, 18F, 18, 19, 48, 49F, 78T Electrostatic charge 117, 194, 208, 210, 312
CFTR (membrane chloride channel protein Emery-Dreifuss muscular dystrophy 54
gene) 18, 49F ENCODE 100, 337, 348F, 360, 363, 364, 366
Cytochrome P450 255, 255T, 256, 257F, 258T, Endoplasmic recticulum (ER) 62, 66, 116, 194,
258, 259, 260, 261 195F
Cytokines 63, 63F, 116, 137, 200, 235, 236T, 241, Enrichment strategies 351
242, 299, 301, 304T, 364 Epidemic 15, 218, 206, 240, 241
Cytosol 63, 194, 195F, 195 Epidermal growth factor receptor (EGRF) 289,
Cytotoxic T lymphocyte associated protein 4 290, 289
(CTLA4) 300, 304T Epigenetics 1, 30F, 30, 31, 32, 35, 57F, 100, 250,
278, 284, 290, 310, 311, 313, 337, 348, 357,
359, 366, 359F
D epigenetic methylation 30, 31, 32
Data protection act (UK) 332 epigenome-wide association studies (EWAS)
De novo mutations 53, 315, 355T 310, 311
Debrisoquine 254T, 255, 258T epigenomics 239, 311
deCODEme 329, 330 Epistasis 35, 101, 105, 161, 310
Degeneration 163, 199, 200 Ethical board (committee) 315, 316, 317, 320,
Deletions 5, 6, 39, 41, 42T, 42, 52, 59, 215, 257T, 322, 330, 332, 333
257F, 257, 258T, 287, 341T, 350, 354, 362 Ethical enquiry 315
Dendritic cells 118, 119, 193 Ethical issues 32, 75, 82, 315

2697_Index.indd 399 08/07/2015 09:27

 400 Index

Ethical principle 316 Golgi stacks 194, 195

Aristotelian ethical philosophy 316 Gram-positive bacteria 63, 91
consequentialism 316 Grave’s disease 301, 304, 326
Eugenics 318, 319, 320F, 320, 322, 332 Gregor Mendel 43, 44B
Evolution 1, 5, 9, 10, 11, 13, 26, 27, 28, 31, 32, GSTM1 280, 287T, 287, 288
183T, 188, 218, 223, 249
Evolutionary 1, 5, 9, 10, 11, 13, 26, 27, 31, 50,
162, 189T, 223, 319
Exome sequencing 293, 310, 350, 351F, 351, 354, H
355T, 356 Haldane 226, 232, 233
Exons 2, 244F, 256, 262F, 265F Hansen 226
Extended haplotype 88, 91, 92T, 196, 200, 216, Haplotype map (HapMap) 35, 85, 88, 100, 101,
217, 219, 299, 366 107F, 107, 125, 151, 170, 236–284, 337,
349F, 353, 360, 361T, 361, 362, 363F, 366
Hardy-Weinberg equilibrium (HWE) 11T, 11, 12,
F 21, 22, 23F, 23B, 24B, 151, 152T, 157
False negative 86 94, 98, 141, 143, 146T, 147, Hardy-Weinberg principle (HWP) 11, 12, 21, 22,
150, 163, 171, 185T, 186T, 233, 280 23B
False positive 27, 86, 94, 96, 141, 143, 146T, 146, HCP5 247
147, 148, 149, 150, 158, 162, 163, 165, 171, Helices 191, 192F, 192
185T, 186T, 228, 233, 235, 280, 309, 343, Helicos platform 345
353, 356 HeliScope sequencer 345, 347F
Family based association testing (FBAT) 86, Helix 2
87F, 87 Hemochromatosis 2, 44, 45, 47, 55T, 176, 189,
Family risk ratio 74B 196, 201
FGFR2 281, 282T, 282 Hemoglobin 229, 230, 231T, 231T, 232, 234T,
Fibrosis 108, 112, 134 135F, 137, 207, 210 235, 236
Fine mapping 62, 80, 98, 99F, 196, 307, 349 Hemolysis 176, 254T
Fisher’s (exact probability) test 152, 154, 155, Hepatitis A virus (HAV) 203F, 205, 213, 249
157, 171 Hepatitis B virus (HBV) 203F, 213, 226, 249
Fixation 9, 13, 14, 15 Hepatitis C virus (HCV) 105, 203, 213, 249, 255T
Fluorescence in situ hybridisation (FISH) 40, 290 Hereditary 35, 226
Founder effect 15, 16, 31 Heritable 10, 30, 35, 40, 56, 78, 83F, 112, 225,
Founder populations 8 226, 227, 249, 278, 357
Fragile X syndrome 48 Heritability 35, 36, 61, 67, 77, 78, 79, 85F, 112,
Fulfilment 316, 332 127, 225, 227, 249, 278, 313
Functional deficiencies 364 Heterodimer 192
Fusion transcripts 350 Heterogeneity 37, 38F, 48, 162, 241, 281
Heterogeneous 27, 37, 57
Heterozygosity 19, 21, 22
G Heterozygote 18, 21, 22, 23F, 24B, 27, 44, 47T,
Gastric cancer 350 150, 151, 152T, 154, 156F, 157F, 158, 185T,
Gastric epithelium 209, 216 188, 215, 231T, 231, 232F, 233, 240, 242
Gefitinib 289 Heterozygote advantage 18, 20
Gene duplication 101, 254T Heterozygous 4, 4F, 17F, 18, 24B, 87F, 87, 118,
Gene expression 29, 30, 32, 65, 278, 282, 290, 142F, 143, 154, 156F, 232, 233, 246, 256,
291, 307, 350, 357 270, 354
Gene expression profile 291 High resolution genotyping 92T, 92, 99F, 100,
Gene flow 12, 17, 22, 23B 184, 213, 219, 234, 280, 348, 361, 366
Gene-gene interactions 35, 101, 161, 310 Hippocampus 126, 127
Gene pool 5, 9, 10, 12, 16, 17, 31, 32 Hirschsprung’s disease 56, 57, 58–61, 67, 78T
Gene sequence 6, 16, 215 ENBR (Endothelin receptor type B) 59T, 59,
Genetic anticipation 48 60, 60F
Genetic divergence 9, 10T, 10, 12, 14 neural crest stem cell regulation 58, 59, 60, 60F
Genetic drift 1, 9, 10T, 10, 13, 14F, 14, 15, 16, peristalsis in gut 58
22, 23B, 26, 27, 28, 31, 163 RET oncogene 59T, 59, 60F, 60
Genetic load 67, 200 toxic mega-colon 58
Genetic portfolio 37 Histone 2, 30, 30F, 31, 278, 311, 357
Genetic protection 39 histone acetylation 311
Genetic underclass 328 histone modification 278, 357
Gilbert’s syndrome 261 histone proteins 2, 30, 31

2697_Index.indd 400 08/07/2015 09:27

 Index 401

HLA (human leucocyte antigen) 9F, 39, 76, 108, HLA genotyping 212
109B, 158 170, 175–220, 236T, 245T, 245, HLA haplotypes 91, 207, 208
246F, 246, 247, 248F, 248, 249, 255T, 266, HLA 7.2 ancestral haplotype 183, 200, 208,
267, 268T, 268, 269F 269, 270, 271, 285, 209, 210, 211, 213
297, 298F, 298, 299, 300, 307, 312, 321, HLA 8.1 ancestral haplotype 9F, 183T, 207,
322, 324, 325 208, 210, 211, 213, 215, 217, 299
HLA alleles 39, 46, 76, 118, 175–220, 267, 268T, HLA types 177
269, 325 A1 9F, 109B, 183T, 183, 202, 207
DQB1*02 9F, 188, 208, 218, 268T, 269, 299, B7 200, 298
304T B8 9F, 176, 202, 207, 298
DQB1*03:01 112F, 112, 137, 267, 299, 304T Bw4 245T, 246, 246F, 247
DQB1*06:02 183T, 196, 198, 199T, 199, 200, Bw6 246
286T, 299 DR2 183T, 185T, 190, 198, 200, 211, 298
DRB1*03 9F, 181, 183T, 183, 188F, 203–208, DR3 183T, 185T 186T, 188, 190T, 202, 211,
218, 299 298, 299
DRB1*04 82, 112, 112F, 117, 118T, 118, 181, DR4 183T, 185T 186T, 190T, 202, 205T, 211,
181, 183T, 202–208, 299, 322, 324 298, 299
DRB1*07 181, 187T, 188F, 197, 204F, 208T, DR7 183T, 186T, 188, 190T
267, 268, 269 DR8 185T, 186T, 211
DRB1*08 181, 182, 203, 211, 212 DR13 190T, 205T, 213
DRB1*11 112F, 112, 181, 183T, 188, 204F, DR15 186T, 190T, 198, 298
211, 212 Holistic view 127, 359
DRB1*13 181, 183T, 203F, 204F, 205T 205, Homozygosity 14, 19, 20, 22, 45, 154, 245T, 245,
208T, 211, 212, 213 246, 247, 261, 355
DRB1*14 117, 118T, 181, 183T Homozygote 18, 21, 27, 44, 45, 62T, 154, 156T,
DRB1*15 181, 183T, 196, 198, 200, 204F, 157, 158, 231, 232, 233, 240, 241, 246, 270
268T, 299 Homozygous 4, 4F, 14, 21, 23F, 24B, 29, 45, 87,
HLA-A*01 9F, 109B, 181B, 218 118, 143, 154, 155, 156F, 230, 240, 246,
HLA-B*07 196, 200 256, 259, 260, 261, 287, 288, 354
HLA-B*08 9F, 118, 183T, 197, 218 Hot spots 167, 186, 361
HLA-B*15:02 267 Hot springs 6
HLA-B*27 46, 47F, 108–110, 115, 116F, Human disease 1, 6, 36, 40, 48, 49, 53, 101, 107,
197, 247 114, 217, 226, 349, 361, 366
HLA-B*57 110, 245T, 247, 248, 255, 267, Human genetics commission (UK) 331
269T, 269, 271, 321, 322, 324, 325 Human microbiome project (HMP) 348, 358
HLA-C*02 197 Human tissue act (UK) 333
HLA-C*06:02 197 Huntington’s disease 48
HLA binding groove 117 Hutchingson-Gilford progeria syndrome (HGPS)
HLA class I genes 181, 245, 267 53
HLA-A gene 109B Hypersensitive reaction 266, 267, 269
HLA-B gene 5, 109B Hypertension 115
HLA-C gene 109B, 248, 249
HLA class II genes 190T, 245, 267
DPA1 gene 189F, 190T, 216 I
DPB1 gene 189F, 190T, 216 Identical twins (monozygotic twins) 61, 79F, 79,
DQA1 gene 187–190, 196, 197, 198, 200, 202, 80T, 178B
203F, 205T, 207, 208T, 211, 212, 216 IGF2 31, 49
DQB1 gene 5, 112, 183T, 185, 187–190, Illumina 309, 341T, 343, 346F, 348F, 354, 358F
196–200, 202, 203F, 205T, 207, 208T, 208, Imatinib 289
211 212, 216, 218 Immune inflammatory disease 65, 82, 116, 128,
DRA1 gene 179T, 181, 185, 188F, 189F, 190T, 367
191, 192, 211 Immune mediated disease 36, 198, 199, 200, 201,
DRB1 gene 5, 119, 120T, 175–220 268, 299, 312
DRB3 gene 179T, 181, 182F, 183T, 184, 189F, Immune regulation 63, 201, 214, 219, 302, 306F
190T, 205T, 205–208, 216, 218 Immune regulatory genes 64, 201, 216, 219,
DRB4 gene 179T, 181, 182F, 183T, 184, 189F, 312,  313
190T, 206–208, 216 Immune regulatory pathways 63, 64, 116, 119
DRB5 gene 181, 182F, 184, 189F, 190T, 207, Immune response 63, 64, 65, 66F, 82, 115, 117,
208, 216 134, 135, 191, 193, 196, 202, 215, 217, 224,
HLA genes 65, 108, 109B, 117, 118, 180B, 184, 227, 228, 233, 235, 238, 240, 241, 242F,
185, 187T, 189, 213, 217 244, 246, 249

2697_Index.indd 401 08/07/2015 09:27

 402 Index

Immune stress 182, 218 Likelihood 6, 9F, 32, 39, 46B, 48, 77, 82, 84, 94,
Immune surveillance 196 96, 98, 110F, 115, 144B, 144, 145B, 147,
Immune synapse 192F, 193, 194F, 219, 299 148, 160, 161, 169, 201, 224, 249, 290,
Immune tolerance 64, 91, 300, 301, 359 328, 332
Immunoglobulin (Ig) 119, 198 Lipid metabolism 82, 101, 134, 137
Immunoglobulin domains 191F, 191, 193F, 216 LOD score 56, 84, 144B, 144, 145B, 145, 171
Immunology 176, 177, 178B Logistic regression 152, 158, 160, 162, 171
Immunoregulatory genes 175, 176, 301 Lung cancer 261, 275, 277F, 278, 281, 283T, 283,
Imprint control elements 31, 49 286, 289, 291F, 292F, 292, 293
Imprinted genes 31
Imputation 90F, 91, 100, 118, 125, 236, 237, 250,
349, 352F, 352, 356, 357, 362, 366
M
Imputation analysis 236, 248, 337, 349, 352F, 361 Macrophage 65, 129, 227T, 237, 238F, 239F,
Impute 88, 89, 90F, 212, 352F, 353 240, 242F
Inbreeding depression 21 Major histocompatibility complex (MHC) 5, 9F, 9,
Index cases 84, 94, 95, 356 45, 86, 108, 109B, 175–220, 285, 246, 247,
Indirect association 165, 353 249, 269F, 285, 298, 299, 301T, 302F, 303,
Infectious disease 15F, 39, 40, 76, 83F, 83, 88, 304T, 305F, 306F, 307, 312, 324, 362
111, 112, 176, 216, 223–250, 278 MHC class I 189, 190, 193, 246
Inflammatory bowel disease (IBD) 61, 94, 108, MHC class II 66, 189, 217F, 302F
261 MHC class III 189, 209, 213, 214, 215, 216
Influenza virus 199, 226 MHC class I chain-related (like) 190, 214,
Innate immunity 66F, 193, 197, 216, 219 215, 216
Insertions 5, 6, 39, 41, 42T, 42, 62, 261, 341, MICA 187F, 189F, 190, 204F, 207, 208T, 209,
350, 362 210, 214, 215, 216, 217F, 218, 219, 299
Insulin gene (INS) 16, 300, 312 MICB 167F, 190, 209, 215, 216
Interactive networks 122 Malaria 19, 76, 176, 223–237, 239, 247,
Interferon (IFN) genes 112, 200 248, 250
Interferon-gamma (IFN-γ) 200 Malaria Genomic Epidemiological Network
Interleukin 7F, 63, 116, 229, 300, 323, 364 (MalariaGEN) 235
Intermediate metabolizers 256F, 256 Manhattan plot 166, 167F, 171, 269F
International cancer group (iCOGS) 293 Maturity onset diabetes of the young (MODY)
International Genetics of Ankylosing Spondylitis 284, 295, 296F, 296, 311T, 311, 312
Consortium (IGAS) 116 MCY oncogene 284F, 285T, 285, 353
International Histocompatibility Testing Medical Research Council (MRC) 317
Workshop(s) (IHTW) 177, 189B Meiosis 7, 8F, 8, 10, 27, 31, 40, 42, 126T, 357
International HIV controllers study 248 asymmetric 10
Interstitial deletions 43F diploid spermatocyte (sperm cells) 10, 41F
Intracellular proteins 194 haploid gametes 10, 40
Intracellular transport 123 polar body 10
Intronic sequences 4, 181B, 284 Melanoma 130, 286, 289, 227F, 285T, 286, 289,
Invariant chain 195 291, 292F
Inversions 41, 42T, 42, 43F, 350 Mendelian patterns 37, 44, 47, 54, 106, 110, 137,
Irinotecan 261, 288 227, 250, 278
Isoform 101, 182, 215, 258, 287 additive dominant effect 29, 35
Isoniazid 254B, 269 autosomal dominant 10, 31, 36, 44, 45F, 45, 48,
49, 78T, 78, 142F, 142, 143, 335T
autosomal recessive 2, 11, 18, 19, 21, 36, 44,
K 45F, 45, 48, 196, 198, 260, 355T
Karyotype(s) 2F, 43F semi-dominant 44
Karyotyping 8F sex-linked 36, 44
Killer cell immunoglobulin like receptor (KIR) X-linked 44, 45F, 227T, 355T
193, 245T, 246 Y-linked 44, 45F
Kilobases (kb) 2, 353 Meta-analysis 62F, 86, 96, 97B, 117, 119, 124T,
125, 130, 163, 171, 163, 171, 309
Metagenomics 337, 357, 359, 366
L Methylation chips 311
Lamin A gene (LMNA) 54 Migration 1, 9, 10T, 10, 12, 13, 14, 16,
Leukocytes 109B, 158, 175, 178B, 179B, 180B 22, 23B, 27, 28F, 31, 65, 123, 126T,
Ligand 60, 60F, 119, 126T, 129, 239F, 243T, 134, 282
244, 245, 246F, 247, 249, 283T, 302, 303, Miller syndrome 349, 355T
307, 364 Missing heritability 67, 77

2697_Index.indd 402 08/07/2015 09:27

 Index 403

Missing-self hypothesis 246 Natural selection 1, 9, 10T, 10, 15F, 16, 17F,
Mitochondrial disease 40 163, 323
Kearns-Sayre syndrome 52T, 52 balancing selection 16, 17F, 18
Leber hereditary optic neuropathy 52T, 52 positive (or adaptive) Darwinian selection 16,
mitochondrial myopathy 52T, 52, 53F 17F, 18
Pearson’s syndrome 52T, 52 purifying (or negative) natural selection 16, 17F
Mitochondrial genes 51 New Germania 319
Mitochondrial genome 1, 27 - 28, 36, 50, 51F, 51, Next generation sequencing (NGS) 337, 340, 349
52T, 52, 53F Nicotine 258T, 283
heteroplasmy 27 Nicotinic acetylcholine receptor 283T, 283
homoplasmic 27 Non-concentric heterochromatin 3F
hot spot positions 28, 167, 196 Non-Hodgkin’s lymphoma 175, 277, 285T, 291F,
mitochondrial DNA (mtDNA) 2, 27, 31, 292F
28, 28F Non-identical twins (dizygotic twins) 61, 79F, 79,
molecular clock 28 80T, 80, 122, 200
Mitochondrial genotype 27, 28F Non-random mating 21, 151
Mitosis 40, 42 Nortriptyline 256, 257F, 259
Modifier 33, 47, 56, 59T, 59, 60 Novel somatic mutations 349
Molecular chaperone 195 Nuclear genome 51
Molecular mimicry 116F, 137 Nucleosome 30, 30F
Molecular signature 357 Nucleosome core particle (NCP) 30
Molecular structures 117, 183T, 191, 199 Nucleus 2, 30, 31, 50, 54, 195F, 238F, 358F
Molecular weight markers 7F
Monist League 319, 320F
Monogenic 37, 44, 47, 54, 55, 56, 57, 68, 84,
O
227T, 227, 296, 311, 312, 313 Odds ratio (OR) 22, 46, 46B, 86, 108, 129T, 154,
Monomorphic 5, 14, 22, 134, 188F 161, 213, 227, 281, 299, 324, 354
Morphine 257, 258, 259 Oligogenic 37, 54–59, 66, 68, 78T, 78
Morula 79F Online mendelian inheritance in man (OMIM)
Mosaic 40, 42T, 107, 349 11, 22, 44, 58, 59T, 66, 119, 121T, 123T,
MSMB 283, 284T 124T, 311
Multi-hit hypothesis 206, 217F, 217, 218, 299 Over dominance 9
Multicenter AIDS Cohort Study (MACS) 342
Multiple sclerosis (MS) 78T, 80T, 119, 121F, 176, P
197, 199–202, 220, 298T, 299, 313
demyelination 199, 200 P value 119, 120T, 124T, 125, 129T, 131T, 133T,
international multiple sclerosis genetic 146, 147, 148, 151, 152T, 164, 165, 166,
consortium (IMSGC) 201 170, 213, 269F, 303, 304, 308T
myelin 199, 200 Paired-end mapping 350
sclerotic plaque Paired-end sequencing 341T, 350
Mycobacterium 226 Paleo 16, 28
Myocardial infarction 127, 266 Pancreas 268F, 291F, 330, 307, 308, 309, 311
Myopathy 52T, 52, 263, 270 pancreatic β cells 198, 297, 307, 308, 311, 312
Myotonic dystrophy 48 Paternity and maternity testing 327
Myriad genetics 330, 331 Pathogenic 63, 66, 116, 125, 194, 212, 223, 225
Pathogens 63, 66, 137, 193, 195F, 214, 223, 224,
225, 226, 240, 241, 249, 250, 358, 359,
N 360
N-acetyltransferase 2 (NAT2) 269, 280, 287T, 287, Patient management 105, 107F, 111, 113, 120,
288 137, 250, 260, 313, 357, 354, 367
Nanopore 345 PCR (polymerase chain reaction) 6, 7F, 109B, 181,
Narcolepsy 176, 197–200, 208, 299, 313 177F, 186T, 207, 211, 229, 298, 299, 338,
axonal pruning 198 341T, 342F, 345, 366
cataplexy 198 polymerase enzyme 6, 338, 339, 340, 341T,
microglial cell expression 198 344F, 345, 346F, 347F
National Center for Biotechnology Information Pearson’s χ2 test 151, 152T, 152, 153T, 153, 154,
(NCBI) 358 157, 163, 164, 171
National Health Service (NHS) 328 Penetrance 44, 45, 46, 47F, 54, 55T, 55, 56, 59,
National human genome research institute 60, 84, 109, 196, 215, 227, 253, 326, 327
(NHGRI) 136, 362 Personalised medicine 2, 325
National Institutes of Health (NIH) 136, 317 Personalised therapy 272, 290, 293, 309, 337
Natural killer (NK) cells 115, 193, 246 Phagocytosis 65, 195F, 195, 233, 234T, 236T

2697_Index.indd 403 08/07/2015 09:27

 404 Index

Pharmacogenetics 6, 40, 86, 113, 196, 253, 254, Product rule 24B, 25B, 25, 26B
263, 271, 272 Prognosis 75F, 75, 78, 111
Pharmacogenomics 253–272, 322, 367 Prostate cancer 275, 277F, 278, 280, 281, 283,
Pharmacological agents 39, 113, 272 284T, 284F, 284, 285T, 285, 286, 292F, 293
Phenotype Protease 195F, 195, 282, 297
phenotypic expression 29 Proteasome 194, 195F
phenotypic traits 1 Protective alleles 39, 161, 163
thrifty phenotype 16 Protein polymorphism 185
Phenylketonuria (PKU) 29, 55T Pseudogene 182F, 182, 188, 189T, 189,
dysfunctional phenylalanine hydroxylase enzyme 190T, 215
(PAH) 29 Psoriasis 108, 176, 196, 197
Phosphodiester bond 339 PTPN22 (Protein tyrosine phosphatase non-
Phylogenetic tree 28 receptor type 22) 86, 115, 117, 118, 119,
Plasmodium 225, 229–237 120T, 121F 122F, 303, 303, 304T, 304,
Pleiotrophy 37, 38F, 39F 305T, 305F
PLINK 89, 125, 150, 151 Publication bias 93, 94, 97, 98
Ploidy 40 Punnett square 22, 24B, 46B, 154, 167
aneuploidy 40, 41, 42T Pyrosequencing technology 342F, 343F, 348F
diploid 4, 10, 11, 13, 20, 40, 41F, 46B, 339
haploid 10, 13, 40
ploid 40 Q
polyploidy 40, 42T Q-Q plot 165, 166F, 166, 171
tetraploidy 40, 41F, 41 Qualitative trait 94
triploid 41F, 41, 42T Quantitative estimation 21
Polygenic 37, 54, 55, 56, 57, 61, 63, 65, 67, Quantitative trait 29, 94, 160, 348
68, 78T
Polymorphic 5, 90F, 109B, 177, 179T, 181B,
182, 187, 192, 247, 262, 264, 271F, 352F, R
364, 365F Random genetic drift 9, 13, 22, 162
Polypeptide 51F, 51, 109B, 117, 15, 188F, 192, Random mating 22, 24B
194, 198, 230, 231F, 305 Receptor(s) 59, 64, 116, 125, 126T, 137, 163, 93,
Poor metaboliser 254T, 255, 256, 257F, 260 216, 224, 228, 235–246, 263, 264 265F, 265,
Population bias 22 282T, 282, 283T, 283, 288, 289, 290, 299,
Population genetics 8, 9, 11, 13, 21, 29, 114 300, 301, 302, 303F, 304T, 304, 305T, 307,
Population size 13, 15, 16, 21, 26, 28, 31, 224 308T, 309, 323, 354, 355F, 360, 364, 365F,
Population stratification (substructure) 26, 27, 365
95T, 95, 151, 161–165, 228, 236 Recombination 7, 8F, 10, 18, 20, 27, 41, 126T,
Population vigor 21 282T, 283T, 284F, 361
Positional cloning 84, 85F Recruitment bias 94
Positive natural selection 9 Red blood cells 19F, 19, 109B, 179B, 229, 230F,
Post-genome 4, 84, 85, 88, 101, 107, 219, 303, 231F, 231T, 231, 232F, 262F
304T, 328 Reductionist 101
Power calculations 93, 94, 97B Redundancy 101, 185, 217F, 241, 288, 302, 364,
Prader-Willi syndrome 31, 49, 50T, 50 366
Pre-genome 84, 85, 115, 147, 301T, 303 Redundant 85
Preferentially bound 117, 199, 203, 210, 247 Regeneration 134, 137
Preferential presentation 299 Regulator 18, 65, 126T, 133, 300, 301, 307
Preponderance 87F, 95, 116, 202, 210 Regulatory cytokines 63F, 63
Prevalence 164, 200, 232, 303 Relative risk (RR) 56, 58, 61, 67, 77, 78T, 78, 85F,
Primary biliary cirrhosis (PBC) 78T, 91, 176, 197, 86, 93, 102, 117, 118T, 122, 127, 164, 210,
201, 203F, 204F, 210–213, 299, 322 227, 233, 297, 298T, 354
Primary ciliary dyskinesia 349 Replication 80, 91, 92T, 93, 94, 97B, 98, 99F, 119,
Primary immune deficiencies 7F, 272 130T, 130, 132, 239, 240, 338, 339F, 350
Primary sclerosing cholangitis (PSC) 82, 176, 197, Reproductive fitness 349
201, 203F, 204F, 207–210, 211, 212, 213, Resistance 19, 39, 88, 175, 203, 205T, 208T, 208,
216, 217, 218, 299, 322 223–250, 280, 289, 290, 293, 322, 353, 358,
biliary obstruction 207 360, 362
cholestasis 297 Restriction enzymes 6
portal hypertension 207 Retinitis pigmentosa 36, 52T, 355
Probability 87, 96, 119, 121T, 128, 132, 144B, RFLP (restriction fragment length polymorphism)
146–149, 152, 154, 155, 157, 158, 160, 161, analysis 6, 109B, 177F, 185T, 186T, 207,
165, 166F, 170, 171, 235, 304 211, 298, 300, 366

2697_Index.indd 404 08/07/2015 09:27

 Index 405

Rheumatoid arthritis (RA) 65, 80T, 81, 82, 86, SOLiD (support oligonucleotide ligation and
105, 115, 116–122, 127, 137, 176, 198, detection) platform 341T, 342, 343, 345F,
201, 203, 207 209, 262, 300, 304, 305T, 348F
305F, 322 Sparteine 254T, 255
cyclic-citrullinate peptide (anti-CCP) 116 Spleen 118, 233
seropositive arthritis 117, 118 Splits 182, 184F, 197
severe erosive disease 118 Statin 100, 263, 269, 270
synovial fluid 116 Statistical analysis 11, 40, 93, 96, 97B, 141–171,
synovial joints 116 202, 352, 354
Ring chromosomes 41, 43F significance threshold 98, 99F, 115, 116, 119,
Risk gene 44, 58, 62,120T, 130, 135F, 137, 202F, 125, 130, 147, 148, 151, 167F, 201, 237,
217F, 299, 304T, 313, 328, 330 247, 354
Risk portfolio 39F, 55, 67, 110, 111F, 114F, 114, statistical confidence 93, 94, 148, 348, 354
11F, 116, 201, 216, 217, 218 statistical error 146F, 146, 147, 149, 157, 162
RNA 2, 30, 51, 225, 237, 238, 239, 256, 278, statistical likelihood 84
293, 302, 308, 345, 359, 355F statistical power 6, 39, 85, 90, 93, 94, 96,
Roche/454 sequencing technique 341T, 342F, 146–150, 157, 163, 170, 171, 228,
342 348F 280, 348
Rs number (ref-SNP cluster ID number) 5, 258 statistical significance 39, 87F, 90, 98, 354
statistical threshold 171, 303, 353
Stroke 52T, 128F, 259
S Sum rule 25B, 25, 26B
Sacroileitis 108 Survival of the fittest 31, 318
Salt bridge 198, 204, 205, 206F, 299 Synthetic associations 353
Sample size 90, 93, 94, 96, 97B, 147, 150, 171, Systemic lupus erythematosus (SLE) 82, 119, 121F,
307, 349, 353, 354, 356 176, 210, 215, 216, 217
Sanger center 329, 338, 362
Sanger sequencing 338, 339, 340F, 356
Schizophrenia 76, 78T, 80T, 82, 122, 123, 124T, T
126T, 349, 350 T cells 135F, 191, 193, 201, 202F, 227–248, 301,
DISC1 82, 123 302F, 357
GABRB1 82, 123, 124T, 126T B7 (B7.1/B7.2) 301, 302F
GMR7 82, 123, 124T, 125, 126T CD4+ T cells 193, 194F, 210, 237, 238, 239F,
Selective pressure 1, 10, 16, 226 242F, 244, 245, 301, 302F, 302, 357
Segregate 42, 84, 151, 168, 255 CD8+ T cells 191, 193, 194F, 210, 236T, 245T,
Segregation 8, 20, 40, 85F, 91 245, 246, 248
Segregation at meiosis 8 CD28 301, 302F
Serotype 180B, 197 CD80 202F, 301, 302F
Serotyping 184F, 211 CD86 202F, 301, 302F
Serum 58, 109B, 181, 200 γδ T-cells 209, 210, 216
Severity 39, 48, 67, 78, 75F, 94, 95, 110, 111, 113, T cell co-stimulator 201
161, 162, 199, 205, 223, 224, 229, 259 T cell receptor (TCR) 117, 191, 192F, 193,
Shared epitope 117, 118T, 118, 202, 203F, 194F, 195, 199, 203, 299, 300, 302F
204, 205 T cell tolerance 201
Sibling pairs 55, 67, 142, 298T, 299, 300 Tagged SNPs 64, 88, 89, 90, 91, 97B, 100, 124T,
Sibling relative risk (λs) 58,61, 77, 78T, 85F, 102, 125, 129, 130, 150, 167F, 170, 213, 236,
117, 122, 127, 164, 210, 297, 298T 248, 281, 287, 288, 293, 303, 309, 349
Sickle cell anemia 4, 19F, 19, 230–235 Tamoxifen 272, 288
Significance threshold 98, 99F, 115, 116, 119, 125, Telomerase 285T, 285
139, 147, 148, 151, 167F, 201, 237, Thalassemia 230–235
247, 354 The thrifty gene hypothesis 16
Silver syndrome 31, 49 Thermophilic bacteria 6
Skin rash 266, 267, 268T Thiopurine methyltransferase (TPMT) 254T, 255,
SLCO1B1 263, 270 256F, 256, 261, 262F, 262, 263, 271
Smallpox 226, 240 Third generation sequencing 343, 345
SNP chip 91, 93, 96, 97B Thymus 118, 201, 300
SNP frequency 6 Tissue type 176
SNP map 4, 85, 88, 101, 107F, 107, 119, Toll-like receptor 64, 216
360, 366 TOX3 282T, 283
SNURF-SNRPN 31, 49, 50T Trans 187, 188F, 188
Social Darwinism 318 Transcriptome 337, 350
Society 14, 315, 317, 318, 319, 326, 328 Translocations 6, 41, 42, 350, 351

2697_Index.indd 405 08/07/2015 09:27

 406 Index

Transmission disequilibrium testing (TDT) 86, 87, V

145, 300
Variable number tandem repeat sequences (VNTR)
Transplantation 109B, 112, 113, 114F, 114, 175,
6, 7F, 62F, 262F, 300
176, 177, 178B, 180, 207
Vascular adhesion 128, 134
Trastuzumab 255T, 290
Vemurafenib 289
True single-molecule sequencing 345
Vesicle 194, 195F, 195
Tumor immunology 176, 178B
Viral disease 65, 201, 203, 213
Tumor suppressor gene 284
Vitamin K epoxide reductase (VKOR) 259, 260F,
Tumor necrosis factor (TNF) 63, 190, 201, 187F,
263
189F, 190, 202F, 214, 216,217F, 218, 299,
VKORC1 (vitamin K epoxide reductase complex 1)
236T
259, 260F, 264, 271
Type 1 diabetes (T1D) 53F, 65, 78T, 80T, 82, 85,
115, 128F, 136, 176, 177F, 198, 200, 208,
295, 296, 297–306, 307, 309, 310, 311, 312, W
313 326 Wahlund’s principle 27
Type 2 diabetes (T2D) 16, 53F, 80T, 82, 85, 86, Warfarin 255T, 257, 258, 259, 260F, 264, 271
115, 128F, 136, 176, 295, 296F, 296, 297, Well-being 316, 332
298T 306–311, 313, 313 Wellcome Trust (WT) 64, 93, 115, 148, 167, 201,
glucose storage 16 237, 299, 317, 329, 362
insulin release 16 West Nile virus 39, 340, 341
islet (of Langerhans) cells 297, 307, 311, 312 Whole genome 40, 41, 42T, 60, 64, 78, 80, 88,
mature insulin 300 89F, 89, 107, 149, 150, 171, 196, 250, 254,
pre-pro-insulin 297F, 300 278, 293, 331, 338, 347, 349, 350, 351F,
pro-insulin 297F, 300 351, 354, 356, 362, 366
Type 1 neurofibromatosis 11 Whole-transcriptome shot-gun sequencing
(WTSS) 350
U Wild type 10, 18, 19, 21, 45, 53, 142F, 143, 232,
233, 242, 256, 259, 262F, 270, 287, 289
UBE3A gene 49, 50T, 50 Wilson’s disease 55T, 201
UDP-glucuronosyltransferases (UGTs) 256F, 261 World Health Organisation (WHO) 225, 276, 295
UGT1A1 261
Ulcerative colitis (UC) 61T, 61, 78T, 94, 207,
209, 213, 298T, 359 X
Ultrarapid metaboliser 254T, 255, 256F, 256, X-ray crystallography 115, 177, 191, 193
259, 260
US Food and Drug Administration (FDA) 260,
261, 264, 267 Z
Uveitis 108 Zygote 13, 24B, 230

2697_Index.indd 406 08/07/2015 09:27

 (a) Fragment library

Primers P1<<P2
P1
Mate-paired library OR + + Polymerase + coupled
beads

(b)

PCR
3’

Emulsion
modification

(c)
Deposition
Bead

(d)
1. Prime and Ligate POH 5. Repeat steps 1-4 to extend sequence
+ Ligase Ligation cycle 1 2 3 4 5 6 7 ... (n cycles)

AT TT GT GT TT CA CG
TA AA CA CA AA GT CG
Universal seq primer (n) AT 3’
1μm
bead 3’
P1 adapter TA Template sequence
6. Primer reset
2. Image Exite Fluorescence

Universal seq primer (n-1)

3’ 2. Primer reset 1. Melt off extended
sequence
3’
TA 1μm
bead 3’
3. Cap unextended strand
Phosphatase
PO4
7. Repeat steps 1-5 with new primer

3’
–1
4. Cleave off fluor
Cleavage agent Universal seq primer (n-1) AT AT AT TC AA TA CC
3’
1μm
HO bead 3’
T GT GC AG TT AT GG
AT
P
3’
TA
8. Repeat reset with, n-2, n-3, n-4 primer

Read position 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35
Primer round

Bridge probe

Bridge probe

Bridge probe

Indicates positions of interogation Ligation cycle 1 2 3 4 5 6 7

Figure 12.5: The Applied Biosystems SOLiD sequencing-by-ligation method. Fragments of DNA are ligated
with oligonucleotide adaptors at each end and hybridized to complementary oligonucleotides attached to
magnetic beads (a). Beads are then placed in a water/oil emulsion where DNA amplification is performed (b).
Once amplification is finished, the beads are placed on a glass surface and entered into the sequencer (c). In
the sequencer a universal sequencing primer, complementary to the adaptor sequence, is used to trigger the
sequencing reaction where ligation cycles with fluorescently labeled degenerate probes is performed. Once the
probe anneals to the DNA template, a DNA ligase covalently binds the probe to the sequencing primer and the
fluorescence is recorded. The probe is then cleaved and another cycle starts. After seven rounds of sequencing,
the extended universal primer is removed and a new universal primer is added that is offset by one base (d). The
sequence of the read is inferred by interpreting the ligation results for the 16 possible dinucleotide interrogation
probes. (Courtesy of Applied Biosystems.)

2967_Colour Insert.indd 1 08/07/2015 10:13

 (a)

A
A

T
+
T

T A
A T

Adaptor-modified DNA strand hybridized to

(b)
oligonucleotide anchor

Denature,
cleave

Cluster generated by Sequencing of forward

bridge amplification strands

(c)
T
POL C
POL A
G
G
T T T T
A C
G
Fluor A C
C
Incorporation cleavage G
G
C
Block
removal

Template
strand

Figure 12.6: The principles of the Illumina sequencing technique. DNA fragments are first ligated onto
oligonucleotide adaptors to form double strands (a). Adapter-modified, single-stranded DNA is then added to
a flow cell and immobilized by hybridization. Bridge amplification generates clonally amplified clusters (b). The
cluster fragments are denatured, annealed with a sequencing primer, and subjected to sequencing-by-synthesis
employing DNA polymerase and four reversible dye terminators. Once incorporated, the terminator stops
the sequencing reaction, which restarts immediately by cleavage of the incorporated dye terminator. Post-
incorporation fluorescence is recorded (c).

2967_Colour Insert.indd 2 08/07/2015 10:13

 (b)

(a) (c)
–log10P

–log10P

1 2 3 4 5 6

After imputation

Figure 12.10: Imputation analysis. The two graphs illustrate a genotype sequence before (on the right) and
after (on the left) impute has been applied. The central section of the figure illustrates in detail six different
haplotypes numbered 1 to 6 for a combination of thirteen bi-allelic genes illustrated as double rings. Each gene
is polymorphic six of the genes have been successfully genotyped. There are two alleles for all of genes (both
genotyped and unknown – not genotyped). The alleles are illustrated by gray shading (A allele) or no shading
(white – B allele) in those genes that have been successfully genotyped. The seven genes illustrated in black
have not been genotyped. Imputation analysis can be applied to enable the genotypes of the black genes on
this haplotype to be estimated based on the known pattern of linkage disequilibrium for the surrounding genes
which have been successfully genotyped. To do this, data are extracted from large genotype databases to fill in
the “missing” genotypes. For example if we know that persons with haplotype 1 which runs top to bottom; white,
unknown, unknown, unknown, white, unknown, gray, unknown, white, gray, gray, unknown, unknown (or B-?-
?-?-B-?-A-?-A-B-A-?-?) carries the white (B) alleles at positions 2, 3, 4, and 6 and the gray (A) alleles at position
8, 12, and 13 in the sequence, then the full sequence for this haplotype can be assigned. If we use the same
alphabetical system (A and B) to name the alleles of the seven unknown genes the sequence for the thirteen
genes of haplotype 1 would be: B-B-B-B-B-B-A-A-B-A-A-A-A. Using the known genotypes from positions 1, 5, 7,
9, 10, and 11 we can identify the unknown genotypes for the other seven genes on the remaining five haplotypes
(haplotypes 2 to 6). This imputed data can be included in the analysis. Because linkage disequilibrium is so strong
in certain areas of the human genome, imputation analysis from initial GWAS data can massively extend the
studies. Thus an initial GWAS study of based 300,000 to 400,000 SNPs can generate data for up to one million or
more markers. (From Marchini J & Howie B [2010] Nat Rev Genet 11:499–511. With permission from Macmillan
Publishers Ltd.)

2967_Colour Insert.indd 3 16/07/2015 11:02

 Nucleus

Cross-link and
fractionate
chromatin

ChIP:
Enriched DNA
binding sites

Sequence

Binding site
mapping

Figure 12.11: Illumina ChIP sequence experiment workflow. DNA fragments are first cross-linked in situ,
fractionated, and then immunoprecipitated. DNA is then sequenced using an Illumina platform to identify
genome-wide sites associated with proteins of interest. (Courtesy of Illumina.)

2967_Colour Insert.indd 4 08/07/2015 10:13