Opinion

Human Gut Microbiome: Function Matters

Anna Heintz-Buschart1,2,3 and Paul Wilmes1,*,@

The human gut microbiome represents a complex ecosystem contributing Trends

essential functions to its host. Recent large-scale metagenomic studies have Functional omics are becoming more
accessible, and increasing numbers of
provided insights into its structure and functional potential. However, the
studies have employed them, demon-
functional repertoire which is actually contributed to human physiology remains strating their potential in identifying
largely unexplored. Here, by leveraging recent omics datasets, we challenge functional traits of the microbiome
related to health and disease.
current assumptions regarding key attributes of the functional gut microbiome,
in particular with respect to its variability. We further argue that the closing of Functional omes display greater varia-
existing gaps in functional knowledge should be addressed by a most-wanted bility and sensitivity to perturbation,
also in cases of where changes in taxo-
gene list, the development and application of molecular and cellular high- nomic composition are minimal, and
throughput measurements, the development and sensible use of experimental they can resolve gut-compartment-
specific information.
models, as well as the direct study of observable molecular effects in the human
host. Methods for resolving functional differ-
ences in meta-omic datasets to the
taxa contributing them have been
The Functional Microbiome developed and are necessary to
The complex assemblages of microorganisms which populate the human gastrointestinal tract understand the impact of microbial
are emerging as key players in governing human health and disease. Several essential functions functions on human physiology.
conferred by the gut microbiome on the human host testify to its importance. These include the
fermentation of indigestible food components into absorbable metabolites, the synthesis of
essential vitamins, the removal of toxic compounds, the outcompetition of pathogens, the
strengthening of the intestinal barrier, and the stimulation and regulation of the immune system
(see recent reviews [1–7]). Most of these functions are interconnected and tightly intertwined
with human physiology. For example, products of microbial fermentation, such as short-chain
fatty acids, represent essential substrates for intestinal cells and play important roles in
immunomodulatory processes, such as T cell differentiation, which, in turn, may affect the
gut microbiome. Although much has been learnt about these tight interrelationships through
carefully conducted mechanistic studies, the extensive diversity of microorganisms and
molecules in the gut implies that our understanding of this expanse requires a comprehensive
toolset to enable new discoveries. More specifically, the emergent functional complement,
which is actually contributed to human physiology by the gut microbiome, requires detailed 1
Luxembourg Centre for Systems
assessment and systematic study. Biomedicine, University of
Luxembourg, 7 avenue des Hauts-
Fourneaux, L-4362 Esch-sur-Alzette,
A widely applied strategy to deconvolute the complex of interactions and to provide avenues to Luxembourg
2
improve human health is constituted by a triad comprising (i) high-resolution, high-fidelity, and Current addresses: Helmholtz-Centre
for Environmental Research GmbH –
high-throughput omics (see Glossary) of microbial biomass and comparative analyses, (ii) UFZ, Department of Soil Ecology,
hypothesis testing in relevant model experimental systems, and (iii) intervention studies in Theodor-Lieser-Str. 4, 06120 Halle
humans. Ideally, the first type of study should yield testable hypotheses relating to the nature of (Saale), Germany
3
German Centre for Integrative
functions conferred by specific microbiota on human physiology, how and why these functions Biodiversity Research (iDiv) Halle-
differ between individuals (most notably between diseased and healthy individuals), and their Jena-Leipzig, Deutscher Platz 5e,
impact on human health. In this context, much has been described regarding the structural 04103 Leipzig, Germany
@
Twitter: @wilmeslab
characteristics of the gut microbiota through the application of 16S rRNA gene amplicon
sequencing and metagenomics [8]. However, to formulate concrete hypotheses for *Correspondence:
mechanistic studies aimed at understanding dependencies between host and microbes, [email protected] (P. Wilmes).

Trends in Microbiology, July 2018, Vol. 26, No. 7 https://doi.org/10.1016/j.tim.2017.11.002 563

© 2017 Elsevier Ltd. All rights reserved.
observational studies should pinpoint specific functions of the microbiome to specific microbial Glossary
populations conferring these. In addition, these studies should identify biologically relevant and Experimental systems: systems
informative read-outs of health status. Here, functional omics are indispensable. that can be employed as a model for
the human microbiome and which
are amenable to manipulation; they
range from mixed-species and
Variability and Information Content of the Functional Microbiome automated culturing, cell-culture-
The generation and integration of functional omics read-outs derived from metatranscrip- based coculture systems, and, as all
tomic, metaproteomic, and metabolomic analyses allow a detailed functional assessment animals carry a microbiome, animal
models, including germ-free or
of the human gut microbiome [9–18]. It has been observed that the functional omes display
gnotobiotic animals, including those
greater variability and sensitivity to perturbation when compared to the information content of with a ‘humanized’ microbiome.
the metagenome [9–12,14–16,18,19]. Therefore, functional omics are expected to more Functional categories: are
accurately portray health and disease states [12,13,18,20]. For example, changes in gene commonly applied to describe the
functions encoded or carried out by
expression have been detected in response to dietary interventions, such as fermented milk
the microbiome (the functional
products [19], and the oral intake of medication [14], despite only minimal changes in complement), usually by grouping
observed community structure in both cases. These observations are seemingly in contra- genes into metabolism-centered
diction to the widely accepted [21,22] interpretation of metagenomic data whereby meta- frameworks, such as the orthology
put forth by the Kyoto Encyclopedia
genomic functional profiles are less variable compared to taxonomic profiles [23] of Genes and Genomes (KEGG) and
(Figure 1A). However, the latter notion may not faithfully reflect reality and may be due to the MetaCyc database, or
several confounding factors. From a methodological point of view, commonly applied nor- functionally annotated orthologous
malization techniques which do not take into account the taxonomic profiles have been groups based on sequence similarity,
such as eggNOG.
shown to underestimate functional variability [24]. In addition, aggregation of genes into broad Functional diversity: is a measure
functional categories, such as whole metabolic modules, based primarily on homology of the number (richness) and
irrespective of the direction of metabolic flux, contributes to the impression of stability. Finally, distribution of different functions
within the community. It is related to
large proportions of the functional genes in a metagenome are not known, and their potential
gene richness but also to
variability is not taken into account at all (see also discussion below). Besides this, we [18] and phylogenetic diversity, as microbial
others [16] have observed that the functional profiles in metatranscriptomes are more variable communities with phylogenetically
compared to metagenomic profiles, even when based on very broad functional categories diverse members often have a wider
functional potential. Phylogenetic
[25] (Figure 1B,C).
diversity and functional diversity have
been observed as traits of human
The central questions that determine whether functional omics can reveal important functional gut microbiota which are relatively
elements of the microbiome are: (i) is the observed variability biologically meaningful?, and (ii) is stable over time.
Functional plasticity: the ability of
a measured microbial functional state informative beyond a single snapshot? For the meta-
the microbial community or its
genome, individual-specific taxonomic profiles have been demonstrated [26,27]. It is notewor- members to adapt to perturbations
thy that, also for functional profiles, greater inter-individual than intra-individual variation is by changing gene expression; it can
observable, at the metagenomic, metatranscriptomic, and metaproteomic levels (Figure 1D stabilize the taxonomic community
structure as well as ecosystem
and Box 1) [18]. Differences in functional profiles provide direct pointers to the functions functions.
involved in microbiome–host interactions. Given the differences between metagenomic and Functional redundancy: is a
metatranscriptomic profiles, the discriminatory power of metatranscriptomics needs to be measure of the number of different
assessed. Based on our own datasets [18] and estimation methods for sample sizes necessary populations within a community that
are able to perform the same
to reach a targeted power [28,29], metatranscriptomic functional profiles are at least as (if not functions. Functional redundancy can
more) powerful in resolving differences as metagenomic profiles (Figure 1E,F). Therefore, the increase functional resilience, in case
functional omes can provide insights into microbial activity and highlight significant microbiome- perturbations affect the taxonomic
community structure; this allows for a
conferred traits. Our own observations also indicate that the functional omes reflect persistent
return to community function, and
individual-specific physiology, and that this signal is not dominated by nonspecific momentary therefore can increase stability.
fluctuations. Intervention studies: seek to
manipulate the human microbiome in
situ, by means of nutrition,
Another fundamental question with respect to microbiome research is: which gut compartment
probiotics, antibiotics, or faecal
is reflected by meta-omics data obtained from faecal samples? Metagenomic data reflect a transplants.
mixture of different locations along the gastrointestinal tract as well as spores [30]. By contrast, Metagenomics: refers to the
the relative abundance of housekeeping transcripts has been found to be related to the site of analysis of genomic DNA for

564 Trends in Microbiology, July 2018, Vol. 26, No. 7

 (A) (B) (C) (G) (H) mixtures of (often unknown) species.
100 100 100 Its purpose can be to assess the
80 80 80 taxonomic composition of a mixed
t microbial community or to elucidate
Reads (%)

Reads (%)

Reads (%)
60 60 60 i1 i2 i3 i4 i5 the functional potential of its
members.
40 40 40 Intra-individual Inter-individual
over Ɵme (Meta)metabolomics: (also referred
to as metabonomics, mainly in the
20 20 20
t1 i1 context of research on single
0 0 0 organisms) technologies that
(D) t2 i2 measure intra- and/or extracellular
0.5
metaG metaT * metabolites in and around microbial
* t3 i3 communities.
*
Jensen shannon div.

0.4 Metaproteomics: aims to

t4 i4 characterize microbial activity by
*
0.3
* * applying the analysis of proteomes to
* * * t5 i5 mixed-species assemblages.
0.2 * * Metatranscriptomics: is the term
Stable potenƟal Similar potenƟals applied to the analysis of RNA of
0.1 communities, usually with the aim of
inferring activity.
metaP t1 i1
0.0 Omics: a group of methodologies
that aim at the characterization of the
Intra family
Inter family

Intra family
Inter family

Intra family
Inter family
Intra individual
Inter individual

Intra individual
Inter individual

Intra individual
Inter individual

t2 i2 total pool of a class of biomolecules,

including metagenomics, (meta)
t3 i3 metabolomics, metaproteomics, and
metatranscriptomics.
t4 i4 Taxonomic and functional
(E) (F) profiling: to quantify the taxa and
t5 i5 functions detected in a sample form
1.0 1.0
0.8 0.8 riable expression
Variable express di id l expression
Individual part of most meta-omic studies of
Power

Power

0.6 0.6 the human microbiome. Increasingly,

0.4 metaG 0.4 metaG taxonomic resolution of functions of
0.2 metaT 0.2 metaT interest within the microbiome is also
0.0 0.0
achieved.
0 25 50 75 100 0 25 50 75 100
Sample size Sample size

Figure 1. Community-Wide View of the Variability of Encoded and Expressed Functions in the Human Gut
Microbiome. (A) High-level functional profiles of 1267 human gut microbiome metagenomes retrieved from the integrated
gene catalogue (IGC) of the human gut microbiome [25]. (B) High-level functional profiles from metagenomes of our own
smaller integrated multi-omics study [18] annotated using the IGC [25] in comparison to the (C) profiles from metatran-
scriptomes of the same samples. (D) Comparison of intra-individual to inter-individual and intra-family to inter-family
distances (Jensen–Shannon divergence) based on functional metagenomic (MG), metatranscriptomic (MT), and meta-
proteomic (MP) profiles [18]; *P < 0.05, Wilcoxon rank sum test. (E,F) Estimation of power to distinguish functional profiles
from members of different families based on metagenome and metatranscriptome measurements [18] applying limma/
voom assumptions and the statistical model of Bi et al. [29] (E) and van Iterson et al. [28] (F). (G) Summarizing scheme,
illustrating functional potentials with limited variability (middle) and functional expression profiles with greater plasticity
(bottom) within an individual over time (t), compared to (H) the variability between different individuals’ (i) microbiomes’
functional potentials (middle), and functional expression profiles (bottom).

activity of specific microbial taxa, such that oral species have very low transcript levels in stool
samples while colonic organisms are highly active [16,18]. Resolving gene expression to the
taxon of origin, and relating this to the overall activity of that taxon, should further help in
distinguishing in situ activity from noise in functional profiles. Discovery of compartment-
specific functional features, which are important in the context of health and disease [31],
may therefore be facilitated by metatranscriptomics (but see also Box 1 for a discussion of other
omics technologies).

Trends in Microbiology, July 2018, Vol. 26, No. 7 565

 Box 1. Which Functional Omes to Look at?
Metatranscriptomics, by highlighting changes in expression, reveals a more dynamic picture of the microbiome than
metagenomics. The technology allows high sampling depths and high taxonomic resolution of functional processes.
Although metatranscriptomic profiles confer essential information on functional gene expression within microbiomes,
metaproteomic profiles may be a better indicator of the actual phenotype. However, metaproteomic analyses are not
yet able to achieve the information content or sampling depth of sequencing-based technologies. Due to this fact, highly
abundant, stably expressed conserved proteins make up the majority of data points, leading to a higher apparent
stability of metaproteomic profiles [13]. However, the current limitations may be overcome through improvements in
protein preparation [17], identification [120,121], and the adaptation of quantitative methods [122]. Metabolomics, by
directly measuring metabolic outcomes, should be the most sensitive with respect to resolving the functional micro-
biome, and quantitative methods – such as proton nuclear magnetic resonance (NMR) analyses – are able to capture
some of the most important, high-abundance microbial metabolites, such as short-chain fatty acids. Although
advanced metabolomic methods allow the resolution of thousands of metabolite features from human microbiome
samples, current limitations include a very large fraction of unknown metabolites (well in excess of 90% of measured
features may be unknowns, even when searching metabolomic data against comprehensive databases [123]) as well as
the difficulty in linking specific metabolite features to their microbial provenance. Advances in computational mass
spectrometry, as well as in de novo metabolic network reconstruction and modelling, will allow some of these limitations
to be addressed in the future. Given the different limitations of the single omic levels, as well as their complementary
information content, the integration of multi-omic data can also help to close gaps when assessing gut microbial activity
in situ by bridging genomic content to final phenotype.

The observation that metatranscriptomic functional profiles are more variable than might be
inferred based solely on metagenomic information suggests that nonhousekeeping genes,
even those with high genomic copy numbers, are not stably expressed in situ [10,11,16,18].
We have recently developed an approach which allows taxon-specific resolution of expressed
genes [18]. When applying this method to link functional genes to the genomes which encode
them, we observed that functions of interest may be contributed to the community-wide
phenotype by single or multiple microbial populations in the absence of observable differences
in the respective populations’ abundances [18]. The identity of these populations may differ in
different individuals, as the microbiota may have widely divergent taxonomic compositions [18].
The variability observed at the level of gene expression may very well be a reflection of
functional plasticity and a prerequisite for stable community function. Consequently, resolv-
ing functional differences at multiple omic levels to the taxa contributing them is necessary in
order to understand when and how these functions may impact human physiology.

The Unknowns
One challenge for microbiome research in relation to elucidating phenotypic impacts on the
host is posed by unknown taxa and functions. While the overall proportion of protein-coding
genes for which a molecular function cannot be predicted in the human microbiome (40–70%,
depending on the prediction method [18,32,33]), is still generally high, this proportion is higher
the rarer a microbial gene is in the human population (Figure 2A). Furthermore, this is especially
the case when encoded in taxa which are not well described or even uncharacterized
(Figure 2B). In many recent studies, genes without known functions, or those from uncultured
taxa, have been completely ignored, because metagenomic data were analysed by mapping to
annotated reference genomes. These approaches often make inefficient use of the data [34],
are likely to introduce biases in the interpretation [35], and do not have a handle on the large
proportion of horizontally transferred functions in the microbiome [36] as well as on strain-
specific functional gene complements [37,38] which make up taxa-specific pangenomes.
Horizontally transferred and strain-specific genes may be essential [39], in particular when
they code for medically relevant functions such as antibiotic resistance [40] or toxins [41]. In this
light, the prediction of functional potential [42,43] or even metabolic outcome [44] based on
rough (i.e., genus-level) taxonomic profiles must be regarded as questionable.

566 Trends in Microbiology, July 2018, Vol. 26, No. 7

 (A) (B)

100 1.4 × 106

1.2 × 106

Number of genes
Annotated genes (%)

80
1 × 106

60 8 × 105
6 × 105
40 4 × 105
AnnotaƟon by: 2 × 105
20 KO BLAST eggNOG
KO HMM MuSt HMM 0
FOAM all

0 to 0.11
0.11 to 0.23
0.23 to 0.45
0.45 to 0.87
0.87 to 1.7
1.7 to 3.4
3.4 to 6.8
6.8 to 13
13 to 26
26 to 51
51 to 100
0
0 20 40 60 80 100
Frequency (%)
Frequency (%)
(C)
1 × 10−2 Key: Genus known,
funcƟon predicted
Abundance metaT (%)

1×10−3 Genus known,

no funcƟon predicted
1×10−5 Genus unknown,
funcƟon predicted
Genus unknown,
1×10−7 no funcƟon predicted
Phylum known,
funcƟon predicted
1×10−9 Phylum known,
no funcƟon predicted
Phylum unknown,
0 funcƟon predicted
0 0.1 1 10 100 Phylum unknown,
no funcƟon predicted
Frequency (%)

Figure 2. Genes of Unknown Function. (A) Relationship between the fraction of functionally annotated genes and the
frequency of their occurrence according to the integrated gene catalogue (IGC) [25]. Annotations: ‘KO BLAST’: KEGG
orthologous group (KO) annotations included in the IGC [25]; ‘KO HMM’: HMM-based annotations using KOs [18];
‘FOAM’: HMM-based annotations using FOAM [32]; ‘eggNOG’: eggNOG-based [33] annotations included in the IGC [25];
‘MuSt HMM’: HMM-based annotations using KOs, Pfam-A-families, TIGR-families, Swiss-Prot- or MetaCyc enzymes [18];
‘all’: all annotations by either of the named methods. (B) Relationship between the number of annotated genes (by any of
the methods displayed in (A), their relative frequency of occurrence, and the level of taxonomic assignment in the IGC [25].
(C) Frequency of occurrence [25] and maximum observed expression [18] of genes in the IGC. Pink dots highlight genes
annotated with orthologous groups or protein domains of unknown function.

Ignoring functional unknowns also limits the potential that metagenomic and metatranscrip-
tomic approaches possess in creating new knowledge. For example, approaches to compare
abundances and genomes of uncultured taxa, which contribute approximately 40% of the
metagenomic data, are well established [45,46]. Similarly, collections of orthologous groups
and protein families without known functions have been established [32,47,48], allowing for
cross-sample comparisons. These approaches facilitate the identification of biologically signif-
icant entities, for example, because they are found to be enriched or depleted in individuals with
a disease or consistently highly abundant and/or expressed. For instance, in our recent multi-
omics study, 9% of the differentially abundant transcripts (between families or between

Trends in Microbiology, July 2018, Vol. 26, No. 7 567

individuals with type 1 diabetes and their healthy family members) were from genes encoding
proteins with domains of unknown function. Likewise, in the integrated gene catalogue (IGC) of
the human gut microbiome [25], 14% of the genes without predicted known functions can be
associated with orthologous groups without known function and, importantly, we found 28% of
these genes to be expressed in our own data [18] (Figure 2C).

Several experimental approaches to gain knowledge on ‘the dark matter’ of the human
microbiome have been proposed, in addition to the proven combination of classical microbio-
logical techniques with functional genomics. ‘Functional metagenomics’ involving the large-
scale in vitro screening of metagenomic sequences has been developed [49–51], including use
of microfluidics to assay millions of metagenomic variants of apparently similar genes [52].
‘Culturomics’, the combination of miniaturized cultivation and advanced sequencing
approaches, for example, to generate metagenomes from enrichment cultures, allows for
the detailed characterization of organisms that are not culturable at a traditional laboratory scale
[53,54]. The elucidation of unknowns that differ in health and disease, as well as the specific role
they play in microbiome–host interactions, is an important challenge for the coming years.

Beyond Single Functions

Another crucial question regarding the contributions of microbiome-conferred functions to
human physiology is related to the dynamics that govern community function and to the
functioning of the microbial ecosystem as a whole [55]: does ecosystem functioning, in addition
to or independent of specific microbial functions, play a role in human health, and are
generalizable patterns discernable from multi-omics? At a fine scale, the gene content of
the gastrointestinal microbiome is remarkably different between individuals. For example, the
majority of the unique genes in the IGC are not found in more than a few percent of the samples
(Figure 3A) [25]. These unique genes, however, carry common functions. In fact, functional
annotations in the IGC are usually carried by many unique genes, with respect to both the whole
catalogue and the subset present in single samples (Figure 3B). In our recent study [18], we also
observed that most functions in microbial metabolism are encoded (Figure 3C) and expressed
(Figure 3D) by a number of different microbial populations in any given sample. In addition, we
have observed that the expression of genes by the same population can change over time,
even when the population’s relative abundance does not change. Finally, the relative transcript
abundance of a gene function with respect to the whole community is independent of the
number of different microbial populations that carry it (Figure 3D). These results imply that
microbiome-conferred services need to be explored with respect to their structural and spatial
dimensions in relation to their effect on host physiology.

The above observations are likely a reflection of functional redundancy within the healthy
human microbiome. Functional redundancy can confer resilience [56] and therefore can
stabilize ecosystem functionality during perturbations [57], which, in the context of the human
microbiome, is generally assumed to lead to both stability and health [58]. However, the actual
relationship between functional redundancy and stability has not been studied in the human gut
microbiome, in contrast to other microbial ecosystems [59,60]. It is not even known whether
there is true redundancy, as different genomic contexts may determine the impact of genes [61]
and, within the gut microbiota, the interaction with the host [62,63]. The assumption that
functional redundancy of the microbiome is related to human health is primarily based on an
apparent relationship between taxonomic stability and the maintenance of taxonomic and
functional diversity over time [18,64,65]. However, for the human gut microbiome, it is currently
unclear whether diversity is a prerequisite for stability [66], which has been shown in other
contexts [67–69]. Functional richness has also been suggested to positively impact human

568 Trends in Microbiology, July 2018, Vol. 26, No. 7

(A) (B)
1 × 107 1 × 106

Genes per annotaon

No. of unique genes

8 × 106 1 × 105

1 × 104
6 × 106
1 × 103
4× 106
1 × 102
2 × 106 1 × 101

0 1 × 100
0 20 40 60 80 Annotaons of:
Frequency (%) Single sample IGC

(C) (D)
25 25

20 20
No. of bins

No. of bins

15 15

10 10

5 5

0 0
1 10 100 10 000 0.1 10 1000
Total MG depth Total MT depth

Figure 3. Functional Redundancy in the Human Microbiome (A) Relationship between the cumulative
number of unique genes and the frequency of their occurrence. The graph is based on the 1267 human gut
microbiome metagenomes retrieved from the integrated gene catalogue (IGC) [25]. (B) Numbers of genes with the same
functional annotation, based on KEGG orthologous groups (KO) and eggNOG [33] orthologous groups, as published with
the IGC [25]. (C) Relationship between the number of population-level genomes (‘bins’) containing genes annotated with a
function in microbial metabolism and the corresponding cumulative metagenomic (MG) depth of coverage of the genes. (D)
Relationship between the number of population-level genomes (‘bins’) expressing annotated genes and the corresponding
cumulative metatranscriptomic (MT) depth of coverage. (C,D) Graphs are based on one representative sample from a
healthy individual [18].

health [70], and decreased functional diversity has been observed in several diseases [22],
although the observed functional richness may also be influenced by colonic transit time [71]. A
higher metabolic diversity ensures digestibility of a wider range of nutrients [72] and potentially
increases overall energy harvest. Metabolic diversity may also offer a protective potential
against environmental toxic substances [3]. Despite the likely importance of functional diversity,
the exact mechanism by which the human host benefits from redundant, diverse, and/or stable

Trends in Microbiology, July 2018, Vol. 26, No. 7 569

communities has not been systematically studied [73–75]. In order to resolve any clear
relationships, future analyses of the microbiome in the context of health and disease will have
to assess functional redundancy. While most existing studies have examined single time points,
the significance of stability of the microbiome will have to be addressed by time series studies
during health and disease as well as experimental perturbations. These studies are also
necessary to infer causal links and determine whether and how ecosystem functions of the
gut microbiome can be shaped by interventions.

Concluding Remarks: A Map to Bring It All Together

Given the potentials and challenges highlighted above, future functional studies will have to
integrate and compare reference-based alignments and de novo genome reconstructions, the
wealth of existing omics datasets, functional knowledge, and orthology-based annotations to
home in on the functions that really matter (Figure 4, Key Figure). The functional knowledge

Key Figure
Roadmap for Using Functional Omics to Create New Knowledge

Funconal
meta’omics: Context-dependent
metatranscriptome expression paerns
metaproteome
…

-
- Funconal Candidate funcons with Intervenon
- annotaons health-related potenal studies
-
Human gut - Targeted New
microbiome Gene catalogue experiments funconal
model systems knowledge
Genes Orthologous Most wanted Funconal
groups of list for in vitro
Metagenome unknown genes funconal
Genome assays
elucidaon
reconstrucons
DUFXX
DUFXY
Genome
DUFYZ
collecon DUFZX
…

Isolates Isolate genomes

Funconal assays

Human omics
clinical and lifestyle data

Figure 4. Crucial steps are the integration of reference genomes, metagenomic data collections, and de novo gene and genome reconstructions in genome and gene
collections or catalogues. Genes should be linked to functions, taxonomic occurrence, and expression in different hosts. Genes without predicted functions can be
grouped by orthology to enable comparative analyses and derive a list of ‘most wanted’ yet to be determined functions. Genes with functions that are likely to affect
human health and/or display suggestive patterns of expression in different human hosts should be validated in targeted experiments in model systems and human
intervention studies.

570 Trends in Microbiology, July 2018, Vol. 26, No. 7

should also be systematically linked to the taxonomic structure [76–78] of the analysed samples Outstanding Questions
at strain-level resolution [26,79–84] to explain microbiome-conferred phenotypic traits from a Which microbial functions impact
mechanistic point of view [85]. In this context, it will be essential to contrast and identify human physiology, and in which con-
text? Are there specific drivers of, or
phenotypic traits which are widely distributed across constituent taxa, that is, those that are protective functions against,
redundant, to those which are only encoded and expressed by specific taxa. In either case, diseases?
identification of functional genes of interest should be performed first, followed by their linking to
constituent taxa along the premise of ‘form follows function’ or ‘function first, taxa second’. How fast, how strong, and how resil-
ient does the functional microbiome
Analogous to the most-wanted taxa list [86], which was hunted down to a large extent within
react to perturbations?
half a decade [87], a functional most-wanted list should therefore also be established by the
community. Such a functional most-wanted list should explicitly take unknowns into account, What do we not know – what do
based on information from omics experiments, such as when or where the genes and products unknown genes contribute to micro-
are observed. This information, as well as potential interaction partners [88], should result in biome functioning and host
physiology?
hypotheses for assays to elucidate molecular functions [89]. Another needed resource to
understand our ‘second genome’ is an Online Mendelian Inheritance in Man-like framework
Functional diversity, redundancy, and
that would list observed links between (functional) microbial genes and human phentoypes. stability – what is the link to human
Such a resource should draw on existing metagenomic or functional meta-omic data [90], in health? Why is a diverse or a stable
addition to functional reference genome databases [48,91]. Finally, this resource should already microbiome supposed to be beneficial
for human health?
anticipate the omes and readouts which are poised to make an impact in the future, such as
growth of different populations within the microbiome [92,93], regulatory elements [94,95], and
(How) can community function and
sRNAs [11]. Additionally, several recent studies have demonstrated the power of integrating ecology be manipulated?
functional [18,96–102] and genetic data of the human host [103–105], which should likewise be
linked to microbiome data in large-scale databases. This knowledge will be essential to
understand the interaction between the microbiome and the human host.

A detailed representation and understanding of the functional microbiome is an essential

prerequisite for future rational interventions leveraging the gut microbiome to alter host
phenotype (see Outstanding Questions). To assess the impact of specific microbial functions
on human physiology, and explain their mechanism of action, experiments in representative
models will be critical. To model cellular interactions and reach high throughputs, miniaturized in
vitro models of the human gut interface [106,107] should therefore be employed. Animal
models, in spite of notable limitations [108–110], can be used to observe systemic impacts.
These studies have yielded remarkable insights through detailed analysis [111–113]. Finally,
once useful and safe candidates for improving human health have been established, interven-
tion trials in human cohorts, through diet [114,115] and/or faecal transplants [116,117], faecal
components [118], such as small molecules, or faeces-derived selected microbiota [119],
could be performed. End-points of these studies should involve the monitoring of health-related
physiological markers as well as follow, in detail, the induced changes in the microbiome over
time using omics measurements to understand the role of the microbiome-borne functional
complement in governing human health and disease.

Acknowledgments
All analyses were performed on the High Performance Computing platform of the University of Luxembourg. This work was
supported by Luxembourg National Research Fund (FNR) CORE programme grants (CORE/15/BM/10404093 and
CORE/16/BM/11276306) to P.W.

References
1. Karasov, W.H. et al. (2011) Ecological physiology of diet and 3. Claus, S.P. et al. (2016) The gut microbiota: a major player in the
digestive systems. Annu. Rev. Physiol. 73, 69–93 toxicity of environmental pollutants? NPJ Biofilms Microbiomes
2. LeBlanc, J.G. et al. (2013) Bacteria as vitamin suppliers to their 2, 16003
host: a gut microbiota perspective. Curr. Opin. Biotechnol. 24, 4. Kamada, N. et al. (2013) Role of the gut microbiota in immunity
160–168 and inflammatory disease. Nat. Rev. Immunol. 13, 321–335

Trends in Microbiology, July 2018, Vol. 26, No. 7 571

5. Abreu, M.T. (2010) Toll-like receptor signalling in the intestinal 30. Browne, H.P. et al. (2016) Culturing of ‘unculturable’ human
epithelium: how bacterial recognition shapes intestinal function. microbiota reveals novel taxa and extensive sporulation. Nature
Nat. Rev. Immunol. 10, 131–144 533, 543–546
6. Sommer, F. and Bäckhed, F. (2013) The gut microbiota – 31. Zhang, S. et al. (2017) Sampling strategies for three-dimensional
masters of host development and physiology. Nat. Rev. Micro- spatial community structures in IBD microbiota research. Front.
biol. 11, 227–238 Cell. Infect. Microbiol. 7, 51
7. Hooper, L.V. et al. (2012) Interactions between the microbiota 32. Prestat, E. et al. (2014) FOAM (Functional Ontology Assign-
and the immune system. Science 336, 1268–1273 ments for Metagenomes): a Hidden Markov Model (HMM) data-
8. Gilbert, J.A. et al. (2016) Microbiome-wide association studies base with environmental focus. Nucleic Acids Res. 42,
link dynamic microbial consortia to disease. Nature 535, 94–103 e145–e145

9. Verberkmoes, N.C. et al. (2008) Shotgun metaproteomics of the 33. Powell, S. et al. (2012) eggNOG v3.0: orthologous groups
human distal gut microbiota. ISME J. 3, 179–189 covering 1133 organisms at 41 different taxonomic ranges.
Nucleic Acids Res. 40, D284–D289
10. Turnbaugh, P.J. et al. (2010) Organismal, genetic, and transcrip-
tional variation in the deeply sequenced gut microbiomes of 34. Narayanasamy, S. et al. (2016) IMP: a pipeline for reproducible
identical twins. Proc. Natl. Acad. Sci. U. S. A. 107, 7503–7508 reference-independent integrated metagenomic and metatran-
scriptomic analyses. Genome Biol. 17, 260
11. Gosalbes, M.J. et al. (2011) Metatranscriptomic approach to
analyze the functional human gut microbiota. PLoS One 6, e17447 35. Timmons, J.A. et al. (2015) Multiple sources of bias confound
functional enrichment analysis of global -omics data. Genome
12. Erickson, A.R. et al. (2012) Integrated metagenomics/metapro-
Biol. 16, 186
teomics reveals human host-microbiota signatures of Crohn’s
disease. PLoS One 7, e49138 36. Brito, I.L. et al. (2016) Mobile genes in the human microbiome
are structured from global to individual scales. Nature 535,
13. Ferrer, M. et al. (2012) Microbiota from the distal guts of lean and
435–439
obese adolescents exhibit partial functional redundancy besides
clear differences in community structure. Environ. Microbiol. 15, 37. Zhang, C. and Zhao, L. (2016) Strain-level dissection of the
211–226 contribution of the gut microbiome to human metabolic disease.
Genome Med. 8, 41
14. Maurice, C.F. et al. (2013) Xenobiotics shape the physiology and
gene expression of the active human gut microbiome. Cell 152, 38. Antony-Babu, S. et al. (2017) Multiple Streptomyces species
39–50 with distinct secondary metabolomes have identical 16S rRNA
gene sequences. Sci. Rep. 7, 11089
15. Pérez-Cobas, A.E. et al. (2013) Gut microbiota disturbance
during antibiotic therapy: a multi-omic approach. Gut 62, 39. McClean, D. et al. (2015) Single gene locus changes perturb
1591–1601 complex microbial communities as much as apex predator loss.
Nat. Commun. 6, 8235
16. Franzosa, E.A. et al. (2014) Relating the metatranscriptome and
metagenome of the human gut. Proc. Natl. Acad. Sci. U. S. A. 40. Sommer, M.O.A. et al. (2014) The human microbiome harbors a
111, E2329–E2338 diverse reservoir of antibiotic resistance genes. Virulence 1,
299–303
17. Xiong, W. et al. (2015) Development of an enhanced metapro-
teomic approach for deepening the microbiome characteriza- 41. Dutilh, B.E. et al. (2013) Screening metatranscriptomes for toxin
tion of the human infant gut. J. Proteome Res. 14, 133–141 genes as functional drivers of human colorectal cancer. Best
Pract. Res. Clin. Gastroenterol. 27, 85–99
18. Heintz-Buschart, A. et al. (2016) Integrated multi-omics of the
human gut microbiome in a case study of familial type 1 diabe- 42. Langille, M.G.I. et al. (2013) Predictive functional profiling of
tes. Nat. Microbiol. 2, 16180 microbial communities using 16S rRNA marker gene sequen-
ces. Nat. Biotechnol. 31, 814–821
19. McNulty, N.P. et al. (2011) The impact of a consortium of
fermented milk strains on the gut microbiome of gnotobiotic 43. Jun, S.-R. et al. (2015) PanFP: pangenome-based functional
mice and monozygotic twins. Sci. Transl. Med. 3, 106ra106 profiles for microbial communities. BMC Res. Notes 8, 479

20. Proctor, L.M. (2016) The National Institutes of Health Human 44. Magnúsdóttir, S. et al. (2017) Generation of genome-scale
Microbiome Project, Seminars in Fetal & Neonatal Medicine. metabolic reconstructions for 773 members of the human gut
Semin. Fetal Neonatal. Med. 21, 368–372 microbiota. Nat. Biotechnol. 35, 81–89

21. Börnigen, D. et al. (2013) Functional profiling of the gut micro- 45. Sunagawa, S. et al. (2013) Metagenomic species profiling using
biome in disease-associated inflammation. Genome Med. 5, 65 universal phylogenetic marker genes. Nat. Methods 10,
1196–1199
22. Cho, I. and Blaser, M.J. (2012) The human microbiome: at the
interface of health and disease. Nat. Rev. Genet. 13, 260–270 46. Nielsen, H.B. et al. (2014) Identification and assembly of
genomes and genetic elements in complex metagenomic sam-
23. Huttenhower, C. et al. (2012) Structure, function and diversity of
ples without using reference genomes. Nat. Biotechnol. 32,
the healthy human microbiome. Nature 486, 207–214
822–828
24. Manor, O. and Borenstein, E. (2017) Revised computational
47. Huerta-Cepas, J. et al. (2016) eggNOG 4.5: a hierarchical
metagenomic processing uncovers hidden and biologically
orthology framework with improved functional annotations for
meaningful functional variation in the human microbiome. Micro-
eukaryotic, prokaryotic and viral sequences. Nucleic Acids Res.
biome 5, 19
44, D286–D293
25. Li, J. et al. (2014) An integrated catalog of reference genes in the
48. Mende, D.R. et al. (2017) proGenomes: a resource for consis-
human gut microbiome. Nat. Biotechnol. 32, 834–841
tent functional and taxonomic annotations of prokaryotic
26. Schloissnig, S. et al. (2013) Genomic variation landscape of the genomes. Nucleic Acids Res. 45, D529–D534
human gut microbiome. Nature 493, 45–50
49. Lakhdari, O. et al. (2010) Functional metagenomics: a high
27. Franzosa, E.A. et al. (2015) Identifying personal microbiomes throughput screening method to decipher microbiota-driven
using metagenomic codes. Proc. Natl. Acad. Sci. U. S. A. 112, NF-kB modulation in the human gut. PLoS One 5, e13092
E2930–E2938
50. Dobrijevic, D. et al. (2013) High-throughput system for the
28. van Iterson, M. et al. (2013) General power and sample size presentation of secreted and surface-exposed proteins from
calculations for high-dimensional genomic data. Stat. Appl. Gram-positive bacteria in functional metagenomics studies.
Genet. Mol. Biol. 12, 449–467 PLoS One 8, e65956
29. Bi, R. and Liu, P. (2016) Sample size calculation while controlling 51. Lam, K.N. et al. (2015) Current and future resources for
false discovery rate for differential expression analysis with RNA- functional metagenomics. Front. Microbiol. 6, 149
sequencing experiments. BMC Bioinformatics 17, 146

572 Trends in Microbiology, July 2018, Vol. 26, No. 7

52. Colin, P.-Y. et al. (2015) Ultrahigh-throughput discovery of pro- 78. Manor, O. and Borenstein, E. (2017) Systematic characteriza-
miscuous enzymes by picodroplet functional metagenomics. tion and analysis of the taxonomic drivers of functional shifts in
Nat. Commun. 6, 10008 the human microbiome. Cell Host Microbe 21, 254–267
53. Lagier, J.-C. et al. (2016) Culture of previously uncultured mem- 79. Luo, C. et al. (2015) Constrains identifies microbial strains in
bers of the human gut microbiota by culturomics. Nat. Microbiol. metagenomic datasets. Nat. Biotechnol. 33, 1045–1052
1, 16203 80. Nayfach, S. et al. (2016) An integrated metagenomics pipeline
54. Berdy, B. et al. (2017) In situ cultivation of previously uncultivable for strain profiling reveals novel patterns of bacterial transmission
microorganisms using the ichip. Nat. Protoc. 12, 2232–2242 and biogeography. Genome Res. 26, 1612–1625
55. Cadotte, M.W. et al. (2011) Beyond species: functional diversity 81. Scholz, M. et al. (2016) Strain-level microbial epidemiology and
and the maintenance of ecological processes and services. J. population genomics from shotgun metagenomics. Nat. Meth-
Appl. Ecol. 48, 1079–1087 ods 13, 435–438
56. Walker, B. (1995) Conserving biological diversity through eco- 82. Truong, D.T. et al. (2017) Microbial strain-level population struc-
system resilience. Conserv. Biol. 9, 747–752 ture and genetic diversity from metagenomes. Genome Res. 27,
57. Naeem, S. (1998) Species redundancy and ecosystem reliabil- 626–638
ity. Conserv. Biol. 12, 39–45 83. Costea, P.I. et al. (2017) metaSNV: A tool for metagenomic
58. Lozupone, C.A. et al. (2012) Diversity, stability and resilience of strain level analysis. PLoS One 12, e0182392
the human gut microbiota. Nature 489, 220–230 84. Quince, C. et al. (2017) DESMAN: a new tool for de novo
59. Allison, S.D. and Martiny, J.B.H. (2008) Colloquium paper: resis- extraction of strains from metagenomes. Genome Biol. 18, 181
tance, resilience, and redundancy in microbial communities. 85. Olesen, S.W. and Alm, E.J. (2016) Dysbiosis is not an answer.
Proc. Natl. Acad. Sci. U. S. A. 105, 11512–11519 Nat. Microbiol. 1, 16228
60. Delgado-Baquerizo, M. et al. (2016) Microbial diversity drives multi- 86. Fodor, A.A. et al. (2012) The ‘most wanted’ taxa from the human
functionality in terrestrial ecosystems. Nat. Commun. 7, 10541 microbiome for whole genome sequencing. PLoS One 7,
61. Blount, Z.D. et al. (2012) Genomic analysis of a key innovation in e41294
an experimental Escherichia coli population. Nature 488, 87. Almeida, M. et al. (2016) Capturing the most wanted taxa
513–518 through cross-sample correlations. ISME J. 10, 2459–2467
62. Louwen, R. et al. (2012) A novel link between Campylobacter 88. Mayers, M.D. et al. (2017) Quantitative metaproteomics and
jejuni bacteriophage defence, virulence and Guillain–Barré syn- activity-based probe enrichment reveals significant alterations
drome. Eur. J. Clin. Microbiol. Infect. Dis. 32, 207–226 in protein expression from a mouse model of inflammatory
63. Sahl, J.W. et al. (2015) Genomic characterization of Burkholde- bowel disease. J. Proteome Res. 16, 1014–1026
ria pseudomallei isolates selected for medical countermeasures 89. Prosser, G.A. et al. (2014) Metabolomic strategies for the iden-
testing: comparative genomics associated with differential viru- tification of new enzyme functions and metabolic pathways.
lence. PLoS One 10, e0121052 EMBO Rep. 15, 657–669
64. Flores, G.E. et al. (2014) Temporal variability is a personalized 90. Nayfach, S. et al. (2015) MetaQuery: a web server for rapid
feature of the human microbiome. Genome Biol. 15, 531 annotation and quantitative analysis of specific genes in the
65. de Meij, T.G.J. et al. (2016) Composition and stability of intestinal human gut microbiome. Bioinformatics 31, 3368–3370
microbiota of healthy children within a Dutch population. FASEB 91. Caspi, R. et al. (2013) The MetaCyc database of metabolic
J. 30, 1512–1522 pathways and enzymes and the BioCyc collection of Path-
66. Coyte, K.Z. et al. (2015) The ecology of the microbiome: Net- way/Genome Databases. Nucleic Acids Res. 42, D459–D471
works, competition, and stability. Science 350, 663–666 92. Korem, T. et al. (2015) Growth dynamics of gut microbiota in
67. McCann, K.S. (2000) The diversity–stability debate. Nature 405, health and disease inferred from single metagenomic samples.
228–233 Science 349, 1101–1106

68. Erkus, O. et al. (2013) Multifactorial diversity sustains microbial 93. Brown, C.T. et al. (2016) Measurement of bacterial replication
community stability. 7, 2126–2136 rates in microbial communities. Nat. Biotechnol. 34,
1256–1263
69. Carrara, F. et al. (2015) Experimental evidence for strong stabi-
lizing forces at high functional diversity of aquatic microbial 94. Nawrocki, E.P. and Eddy, S.R. (2014) Computational identifica-
communities. Ecology 96, 1340–1350 tion of functional RNA homologs in metagenomic data. RNA
Biol. 10, 1170–1179
70. Le Chatelier, E. et al. (2013) Richness of human gut microbiome
correlates with metabolic markers. Nature 500, 541–546 95. Bao, G. et al. (2015) Strand-specific community RNA-seq
reveals prevalent and dynamic antisense transcription in human
71. Roager, H.M. et al. (2016) Colonic transit time is related to
gut microbiota. Front. Microbiol. 6, 94
bacterial metabolism and mucosal turnover in the gut. Nat.
Microbiol. 1, 16093 96. Kato, T. et al. (2014) Multiple omics uncovers host-gut microbial
mutualism during prebiotic fructooligosaccharide supplementa-
72. Tap, J. et al. (2015) Gut microbiota richness promotes its sta-
tion. DNA Res. 21, 469–480
bility upon increased dietary fibre intake in healthy adults. Envi-
ron. Microbiol. 17, 4954–4964 97. Palm, N.W. et al. (2014) Immunoglobulin A coating identifies
colitogenic bacteria in inflammatory bowel disease. Cell 158,
73. Matsen, F.A. (2014) Phylogenetics and the Human Microbiome.
1000–1010
Syst. Biol. 64, e26–e41
98. Kolmeder, C.A. et al. (2015) Colonic metaproteomic signatures
74. Moya, A. and Ferrer, M. (2016) Functional redundancy-induced
of active bacteria and the host in obesity. Proteomics 15,
stability of gut microbiota subjected to disturbance. Trends
3544–3552
Microbiol. 24, 402–413
99. Debyser, G. et al. (2016) Faecal proteomics: A tool to investigate
75. Sommer, F. et al. (2017) The resilience of the intestinal micro-
dysbiosis and inflammation in patients with cystic fibrosis. J.
biota influences health and disease. Nat. Rev. Microbiol. 15,
Cyst. Fibros. 15, 242–250
630–638
100. Liu, S. et al. (2016) The host shapes the gut microbiota via fecal
76. Jiang, Y. et al. (2016) Metatranscriptomic analysis of diverse
microRNA. CHOM 19, 32–43
microbial communities reveals core metabolic pathways and
microbiome-specific functionality. Microbiome 4, 2 101. Pedersen, H.K. et al. (2016) Human gut microbes impact host
serum metabolome and insulin sensitivity. Nature 535, 376–381
77. Plichta, D.R. et al. (2016) Transcriptional interactions suggest
niche segregation among microorganisms in the human gut. 102. Okai, S. et al. (2016) High-affinity monoclonal IgA regulates gut
Nat. Microbiol. 1, 16152 microbiota and prevents colitis in mice. Nat. Microbiol. 1, 16103

Trends in Microbiology, July 2018, Vol. 26, No. 7 573

103. Turpin, W. et al. (2016) Association of host genome with intesti- 114. David, L.A. et al. (2013) Diet rapidly and reproducibly alters the
nal microbial composition in a large healthy cohort. Nat. Genet. human gut microbiome. Nature 505, 559–563
48, 1413–1417 115. O’Keefe, S.J.D. et al. (2015) Fat, fibre and cancer risk in African
104. Bonder, M.J. et al. (2016) The effect of host genetics on the gut Americans and rural Africans. Nat. Commun. 6, 6342
microbiome. Nat. Genet. 48, 1407–1412 116. van Nood, E. et al. (2013) Duodenal infusion of donor feces for
105. Wang, J. et al. (2016) Genome-wide association analysis iden- recurrent Clostridium difficile. N. Engl. J. Med. 368, 407–415
tifies variation in vitamin D receptor and other host factors 117. Li, S.S. et al. (2016) Durable coexistence of donor and recipient
influencing the gut microbiota. Nat. Genet. 48, 1396–1406 strains after fecal microbiota transplantation. Science 352,
106. Kim, H.-J. et al. (2016) Contributions of microbiome and 586–589
mechanical deformation to intestinal bacterial overgrowth and 118. Ott, S.J. et al. (2017) Efficacy of sterile fecal filtrate transfer for
inflammation in a human gut-on-a-chip. Proc. Natl. Acad. Sci. treating patients with Clostridium difficile infection. Gastroenter-
U. S. A. 113, E7–E15 ology 152, 799–811.e7
107. Shah, P. et al. (2016) A microfluidics-based in vitro model of the 119. Petrof, E.O. et al. (2013) Stool substitute transplant therapy for
gastrointestinal human-microbe interface. Nat. Commun. 7, the eradication of Clostridium difficile infection: ‘RePOOPulating’
11535 the gut. Microbiome 1, 3
108. Fritz, J.V. et al. (2013) From meta-omics to causality: experi- 120. Tanca, A. et al. (2016) The impact of sequence database choice
mental models for human microbiome research. Microbiome 1, on metaproteomic results in gut microbiota studies. Microbiome
14 4, 51
109. Nguyen, T.L.A. et al. (2015) How informative is the mouse for 121. Tang, H. et al. (2016) A graph-centric approach for metage-
human gut microbiota research? Dis. Models Mechan. 8, 1–16 nome-guided peptide and protein identification in metaproteo-
110. Arrieta, M.-C. et al. (2016) Human microbiota-associated mice: mics. PLoS Comput. Biol. 12, e1005224
a model with challenges. Cell Host Microbe 19, 575–578 122. Gillet, L.C. et al. (2012) Targeted data extraction of the MS/MS
111. Furusawa, Y. et al. (2013) Commensal microbe-derived butyrate spectra generated by data-independent acquisition: a new con-
induces the differentiation of colonic regulatory T cells. Nature cept for consistent and accurate proteome analysis. Mol. Cell.
504, 446–450 Proteomics 11, O111.016717
112. Atarashi, K. et al. (2013) Treg induction by a rationally selected 123. Roume, H. et al. (2013) A biomolecular isolation framework for
mixture of Clostridia strains from the human microbiota. Nature eco-systems biology. ISME J. 7, 110–121
500, 232–236
113. Turroni, F. et al. (2016) Deciphering bifidobacterial-mediated
metabolic interactions and their impact on gut microbiota by
a multi-omics approach. ISME J. 10, 1656–1668

574 Trends in Microbiology, July 2018, Vol. 26, No. 7