The ISME Journal (2011) 5, 82–91

& 2011 International Society for Microbial Ecology All rights reserved 1751-7362/11
www.nature.com/ismej

ORIGINAL ARTICLE
Toward defining the autoimmune microbiome
for type 1 diabetes
Adriana Giongo1, Kelsey A Gano1, David B Crabb1, Nabanita Mukherjee2, Luis L Novelo2,
George Casella2, Jennifer C Drew1, Jorma Ilonen3,4,5, Mikael Knip5,6,7, Heikki Hyöty5,8,
Riitta Veijola5,9, Tuula Simell5,10, Olli Simell5,10, Josef Neu11, Clive H Wasserfall12,
Desmond Schatz11, Mark A Atkinson12 and Eric W Triplett1
1
Department of Microbiology and Cell Science, University of Florida, Gainesville, FL, USA; 2Department of
Statistics, University of Florida, Gainesville, FL, USA; 3Department of Clinical Microbiology, University of
Kuopio, Kuopio, Finland; 4Immunogenetics Laboratory, University of Turku, Turku, Oulu, and Tampere,
Finland; 5Juvenile Diabetes Research Foundation (JDRF), Center of Prevention of Type 1 Diabetes, Turku,
Finland; 6Hospital for Children and Adolescents, University of Helsinki, Helsinki, Finland; 7Department of
Pediatrics and Research Unit, Tampere University Hospital, Tampere, Finland; 8Department of Virology,
University of Tampere, Medical School and Center for Laboratory Medicine, Tampere University Hospital,
Tampere, Finland; 9Department of Pediatrics, University of Oulu, Oulu, Finland; 10Department of Pediatrics,
Turku University Hospital, Turku, Finland; 11Department of Pediatrics, University of Florida, Gainesville, FL,
USA and 12Department of Pathology, Immunology, and Laboratory Medicine, University of Florida,
Gainesville, FL, USA

Several studies have shown that gut bacteria have a role in diabetes in murine models. Specific
bacteria have been correlated with the onset of diabetes in a rat model. However, it is unknown
whether human intestinal microbes have a role in the development of autoimmunity that often leads
to type 1 diabetes (T1D), an autoimmune disorder in which insulin-secreting pancreatic islet cells are
destroyed. High-throughput, culture-independent approaches identified bacteria that correlate with
the development of T1D-associated autoimmunity in young children who are at high genetic risk for
this disorder. The level of bacterial diversity diminishes overtime in these autoimmune subjects
relative to that of age-matched, genotype-matched, nonautoimmune individuals. A single species,
Bacteroides ovatus, comprised nearly 24% of the total increase in the phylum Bacteroidetes in cases
compared with controls. Conversely, another species in controls, represented by the human
firmicute strain CO19, represented nearly 20% of the increase in Firmicutes compared with cases
overtime. Three lines of evidence are presented that support the notion that, as healthy infants
approach the toddler stage, their microbiomes become healthier and more stable, whereas, children
who are destined for autoimmunity develop a microbiome that is less diverse and stable. Hence, the
autoimmune microbiome for T1D may be distinctly different from that found in healthy children.
These data also suggest bacterial markers for the early diagnosis of T1D. In addition, bacteria that
negatively correlated with the autoimmune state may prove to be useful in the prevention of
autoimmunity development in high-risk children.
The ISME Journal (2011) 5, 82–91; doi:10.1038/ismej.2010.92; published online 8 July 2010
Subject Category: microbe-microbe and microbe-host interactions
Keywords: Bacteroidetes; Firmicutes; gut microbiota; seroconversion

Introduction biota, a leaky intestinal mucosal barrier and an

altered intestinal immune responsiveness. The
A recent review by Vaarala et al. (2008) describes a interplay of these factors seems to have a crucial
trio of factors that create a perfect storm of events role in the onset of several allergenic and auto-
leading to autoimmunity in type 1 diabetes (T1D). immune diseases, including Crohn’s disease, celiac
These factors include an aberrant intestinal micro- disease, T1D and multiple sclerosis (Frank et al.,
2007; Wen et al., 2008; Willing et al., 2009).
Insulin-dependent T1D is relatively common,
Correspondence: EW Triplett, Department of Microbiology with no female dominance. Chronic and auto-
and Cell Science, University of Florida, 1052 Museum Road, immune diseases are usually diagnosed in young
Gainesville, FL 32611-0700, USA.
E-mail: [email protected] children and are caused by T-cell-mediated destruc-
Received 24 February 2010; revised 25 May 2010; accepted 28 tion of insulin-producing pancreatic b cells in the
May 2010; published online 8 July 2010 islets of the pancreas (Harrison et al., 2008). A role
 Characteristics of the autoimmune microbiome
A Giongo et al
83
for bacteria in the onset of diabetes has been shown months and blood samples are collected and
in two murine models. For example, feeding pro- assayed for the presence of specific autoantibodies.
biotic bacterial strains, usually lactic acid bacteria, Once two autoantibodies are detected, the child is
to non-obese diabetic mice or biobreeding diabetes- diagnosed as having seroconverted to autoimmunity
prone (BB-DP) rats can delay or prevent diabetes for T1D.
(Matsuzaki et al., 1997; Calcinaro et al., 2005; Yadav
et al., 2007). Feeding antibiotics to nonobese diabetic
mice or BB-DP rats can also increase survival in Materials and methods
these models (Brugman et al., 2006; Schwartz et al., The samples used in this study came from a total of
2007). In addition, pathogen-free nonobese diabetic eight Finnish children, each represented by three
mice lacking an adaptor protein for multiple toll-like stool samples collected at three time points, in a
receptors known to bind to bacterial ligands fail to total of 24 separate samples. The case children all
develop diabetes (Wen et al., 2008). developed autoimmunity and eventually T1D over-
Roesch et al. (2009b) conducted a culture- time (Table 1). Autoimmunity was diagnosed by
independent analysis of gut bacteria in BB-DP the appearance of at least two autoantibodies. Each
and biobreeding diabetes-resistant (BB-DR) rats of these cases is matched with three samples from
and showed that, at the time of diabetes onset, a child of the same age and HLA-DQ genotype who
the bacterial communities in these two rat strains did not become autoimmune during the study.
differed significantly. Stool from BB-DR rats con- Stool samples were collected by parents at home
tained much higher populations of probiotic-like and delivered to the repository for frozen storage
bacteria, such as Lactobacillus and Bifidobacterium, within 48 h. Recent work has shown that storage at
whereas BB-DP rats had higher numbers of Bacter- room temperature for up to 72 h has a minimal effect
oides, Eubacterium and Ruminococcus. A total of 24 on stool bacterial community structure (Roesch
bacterial species were found to differ significantly et al., 2009a). The case children became autoim-
in abundance between the BB-DP and BB-DR rat mune at or near the time of the third sampling
samples. Five species of Clostridium were higher (Table 1). Autoimmunity in T1D is defined as the
in BB-DP rats, whereas Clostridium hylemonae appearance of two autoantibodies in the serum as
was higher in BB-DR rats. In addition, hundreds of described by The Environmental Determinants
bacterial taxa that could not be classified to genus of Diabetes in The TEDDY Study Group, (2007).
level were also found to differ. Many Lachnospiraceae Each case subject was matched with a healthy
were in higher abundance in BB-DP rats, whereas (that is, non-T1D-associated autoantibody positive,
many unclassified Clostridiaceae were more common nondiabetic) child of the same genotype and of
in BB-DR rats. The differences in Lactobacillus and approximately the same age.
Bifidobacterium observed by pyrosequencing were DNA extraction, 16S rRNA amplification and
confirmed by quantitative PCR. pyrosequencing were performed as described pre-
All these results from Roesch et al. (2009b) are viously (Roesch et al., 2009a). An average of 15 709
consistent with the notion that beneficial bacteria sequences were obtained for each of the 24 samples
seem to provide a protective effect in rodent models (Table 2). The barcodes used in this study to
by delaying or preventing the onset of diabetes. As differentiate the samples are described in Supple-
BB-DP rats have lower populations of species that mentary Table 1. The original sequences and the
contain known probiotic strains than do BB-DR rats, corresponding quality scores are submitted to
potentially beneficial bacteria may be necessary for GenBank as study accession number SRP002359.1.
the maintenance of a healthy microbiome essential
in preventing a leaky gut. A Lactobacillus johnsonii Table 1 Age (days after birth) of each subject at the three time
strain has since been isolated from the stool of the points of collection
same set of BB-DR rats as was used in Roesch et al.
(2009b). This strain of L. johnsonii prevents diabetes Subject Days after birth
when fed to BB-DP rats (Valladares et al. 2010).
These results encouraged a close examination of Sample 1 Sample 2 Sample 3 Autoimmunity T1D
gut bacteria in humans who are at high risk for
autoimmunity and T1D. Human stool samples for Case 1 125 351 512 513 513
Control 1 244 370 599
such an analysis have been collected by the Diabetes Case 2 131 447 659 902 1430
Prediction and Prevention study (DIPP) in Finland Control 2 139 445 655
(Nejentsev et al., 1999; Kupila et al., 2001). DIPP has Case 3 123 552 832 298 731
been collecting stool samples from children on Control 3 149 611 1093
the basis of their genotype since 1994. At birth, the Case 4 147 242 696 470 1921
Control 4 147 238 693
HLA-DQ genotypes of babies are determined. Those
infants who possess specific HLA genotypes are Average 150.6 407 717.4
considered to be at high risk for autoimmunity and
progression toward T1D early in life. When children For each case subject, the ages of diagnosis for autoimmunity and
enter the study, stool samples are collected every 3 type 1 diabetes (T1D) are also listed.

The ISME Journal

 Characteristics of the autoimmune microbiome
A Giongo et al
84
Table 2 Characteristics of samples collected from the DIPP study Table 3 Mean percent of total reads for all phyla identified in the
case and control samples
Reads Number
Phylum Cases Controls
Total number of reads 406 979
Number of useable reads 390 759 1 2 3 1 2 3
Average of read length 264
Average number of reads/sample 15 709 Bacteroidetes 53.27 62.19 69.17 76.13 60.07 54.65
Minimum number of reads/sample 9 025 Firmicutes 43.41 25.60 20.66 21.78 22.44 25.89
Maximum number of reads/sample 42 942 Proteobacteria 1.92 6.24 1.74 0.37 0.39 0.65
Fusobacteria 0.02 0.40 0 0 0.05 0.37
Abbreviation: DIPP, Diabetes Prediction and Prevention study.
Phyla were identified by clustering at the 80% level of similarity
using the Megablast results. The number following the case–control
The sequences were processed and analyzed designation refers to time point of collection. The time point 3
to determine differences at all taxonomic and samples were collected at the time of autoimmunity diagnosis for the
community levels using PANGEA (Giongo et al., case children. Those phyla that are statistically significantly higher
(Pp0.01) in cases or controls are depicted in bold.
2010a). In PANGEA, small sequences are discarded,
poor quality ends are trimmed, 16S rRNA sequences
are separated by barcode, the closest cultured relative
of each sequence is identified using Megablast, data communities in healthy children are more similar
are collated into stables, statistically significant to each other.
differences in the abundance of taxa are determined To test whether the microbial communities from
and the Shannon diversity index (Shannon and controls are more similar to each other than those
Weaver, 1949) of each community is calculated. of cases during the time course, samples from the
Classification was performed using an up-to-date same time point were grouped before the clustering
RDP-II database modified using TaxCollector (Giongo using cluster database at high identity with tolerance
et al., 2010b). The Shannon diversity index was (Li and Godzik, 2006) and then submitted to principal
chosen as it considers both the presence and abun- coordinate analysis using weighted UniFrac (Lozupone
dance of operational taxonomic unit. The significant et al., 2006).
differences between taxa are determined in a
manner that includes a false discovery rate determi-
nation and is carried out using a modified w2-test. Results
To test whether the microbial communities from
controls are more similar to each other than those Stool samples were obtained from eight children
from cases, UniFrac (Lozupone et al., 2006) pairwise in the DIPP in Finland. These children represented
distances in between all pairs of cases and all four matched case–control pairs in which the
pairs of controls were determined. Specifically, for samples of each autoimmune child were paired
each one of the six possible pairs of individuals in with samples from a nonautoimmune child of
the same group, a corresponding phylogenetic tree approximately the same age and HLA genotype
was generated using MUSCLE (Edgar, 2004). The (Table 1). Samples from each individual were taken
sum of the UniFrac distances of the trees in the case at three time points (Table 1). The first sample was
group was computed, as well as the sum of the six taken between 4 and 8 months of age before any
UniFrac differences in the control group. The child was found to have autoantibodies. In two of
difference between these sums (D_observed) was the four case subjects, the first autoantibody
used to calculate the difference between cases and appeared about 6 months before the second collec-
controls. A permutation test was calculated to verify tion point. In all four case subjects, the second
whether the difference was statistically significant. autoantibody was detected within several months of
To assess the significance of the UniFrac distances the third collection point. Ultimately, all four cases
described above, we ran a permutation test as were diagnosed with T1D.
follows (R code is available from the authors). For After trimming the 16S rRNA sequences for
each phylogenetic tree, we randomly permute the low-quality sequences and bases, a total of 390 759
labels of the individuals (children) and calculate D sequences were useful in the analysis of the 24 stool
for the data with permuted labels. This is repeated samples examined for this case–control experiment
1000 times to obtain a sample of differences (Table 2). This represents an average of 15 709
(D_permuted), and to create a histogram that sequences per sample. The number of sequences
estimates this distribution. As the labels are per- identified at six taxonomic levels shows that the
muted, the populations are now equivalent, and the number of species identified in all samples was
average value of D should be closer to zero. 377, with the species defined at the 99% level of
If the observed distance, D_observed, is in the similarity. The classification of sequences at each of
tail of the sample of D_permuted, pairwise distances the six phylogenetic levels for all cases and controls
in the case group are significantly greater than those at each collection time point was determined
of the control. In other words, the microbial (Tables 3 and 4; Supplementary Tables 2–5).

The ISME Journal

 Characteristics of the autoimmune microbiome
A Giongo et al
85
Table 4 Mean percent of total reads for those species (or OTU, operational taxonomic unit) identified that showed significant differences
in abundance between cases and controls at any time point of collection

Species (or OTU) Cases Controls

1 2 3 1 2 3

Alistipes onderdonkii 0 0 0.305 0.005 0.022 0.017

Bacterium mpn-isolate group 18 0.019 0.144 1.078 0 0.034 0.057
Bacteroides caccae 0.226 0.398 1.278 0.390 0.137 0.813
Bacteroides dorei 13.50 6.34 13.01 25.44 10.24 12.69
Bacteroides fragilis 3.95 1.10 3.80 12.42 11.28 6.99
Bacteroides galacturonicus 0 0.337 0 0 0.345 0
Bacteroides ovatus 1.53 9.82 9.94 1.75 1.10 0.62
Bacteroides sp. B2 1.17 1.25 0.09 1.46 0.02 0.04
Bacteroides sp. CJ78 2.95 0.84 5.10 2.57 0.69 0.33
Bacteroides thetaiotaomicron 0.34 3.65 8.29 0.35 0.71 1.95
Bacteroides uniformis 0.57 9.35 2.55 0.02 0.02 1.31
Bacteroides vulgatus 22.17 9.53 0.001 12.14 1.77 4.54
Citrobacter sp. CK3 0.015 0.009 0.009 0.002 0.006 0.010
Clostridiaceae DJF VR07 0.246 0.006 0.019 0.315 0.069 0.004
Clostridiales 80/3 0 0 0.485 0 0.231 0.019
Clostridiales 80/4 0.053 0.083 0.103 0.023 0.043 0.042
Clostridium aldenense 0 0.091 0.049 0.142 0.014 0.066
Clostridium hathewayi 0 0.288 0.050 0 0.007 0.058
Clostridium innocuum 0.004 0.040 0.260 0 0.007 0.007
Clostridium orbiscindens 0.082 0.071 0.249 0.103 0.029 0.116
Clostridium sp. 0 0.219 0.149 0 0.017 0.040
Clostridium sp. CJ67 0 0.006 0.011 0 0.001 0.002
Dialister sp. E2 20 0 1.296 0.187 0 0.002 0
Enterobacter sp. NII-26 0.041 0.026 0.019 0.016 0.015 0.013
Escherichia coli 1.195 4.094 0.689 0.149 0.087 0.078
Eubacterium eligens 0 0.582 0.483 0 1.114 1.480
Eubacterium rectale 0 0 0.706 0 0.716 2.859
Faecalibacterium prausnitzii 0 7.40 0.79 0 1.18 3.32
Faecalibacterium sp. DJF VR20 0 0.101 0.013 0 0.267 0.882
Granulicatella para-adiacens 0.09 0 0.003 0.005 0.007 0
Haemophilus parainfluenzae 0.141 0.005 0.016 0.044 0.148 0.082
Human intestinal firmicute CB47 0 0 0.002 0 0.443 1.066
Human intestinal firmicute CO19 0 0.749 1.800 0 6.316 7.114
Human intestinal firmicute CO35 0 0.050 0.125 0 1.342 0.554
Lachnospiraceae DJF RP14 0.03 0.166 0.194 0 0.042 0.105
Lactobacillus rhamnosus 0.737 0 0.012 0.046 0 0.019
Odoribacter splanchnicus 0.004 0.138 0.035 0.288 0.006 0.075
Parabacteroides distasonis 0.97 0.77 0.04 4.09 0.49 1.07
Parabacteroides sp. DJF B086 0.04 0 0 0.12 0.04 0.02
Parasutterella excrementihominis 0 0.60 0.29 0 0.01 0.01
Ruminococcus bromii 0 0.016 0.085 0 2.516 0.037
Ruminococcus gnavus 12.02 0.39 0.54 12.61 0.55 0.14
Ruminococcus sp. CB3 0.026 0.162 0.703 0 0.067 0.120
Ruminococcus sp. CO27 0 0 0.456 0 0.037 0.128
Ruminococcus sp. ID8 0 0.025 0.051 0 0.097 0.023
Ruminococcus sp. M10 0 0.011 0.009 0 0.022 0.015
Streptococcus salivarius 0.27 0.03 0.08 0.01 0.28 0.05
Swine fecal bacterium RF3E-Xyl1 0.008 0.187 0.037 0 0 0.013
Veillonella atypica 2.31 0.10 0.05 0.31 0.08 0.02
Veillonella dispar 5.49 0.06 0.19 0.99 0.21 0.02
Veillonella sp. ADV 3107.03 1.841 0.022 0.054 0.011 0.013 0.002

Species/OTUs were identified by clustering at the 99% level of similarity using the Megablast results. The number following the case/control
designation refers to time point of collection. The time point 3 samples were collected at the time when the case children were positive for at least
two autoantibodies. Those genera that are statistically significantly higher (Pp0.01) in cases or controls are depicted in bold. Species that
represent at least 1% of the total reads in either cases or controls at the time of autoimmunity are highlighted in grey.

Phylum level all sequences at the first collection point to 69.17%

By far, the two most striking differences between the of all sequences at the third collection point, whereas
healthy (nonautoimmune) and autoimmune stool in control samples, Bacteroidetes sequences decreased
microbial communities were the differences within overtime from 76.13% to 54.65% of all sequences
the two most abundant phyla, the Bacteroidetes (Table 3). In contrast, the second most abundant
and the Firmicutes (Table 3). In case samples, the phylum, the Firmicutes, expressed an inverse pattern.
Bacteroidetes sequences increased from 53.27% of The Firmicutes sequences declined in case samples

The ISME Journal

 Characteristics of the autoimmune microbiome
A Giongo et al
86
overtime from 43.41% to 20.66% of all sequences, 38.63% of the bacteria present in stool samples.
whereas they increased in control samples from Two genera in the Firmicutes, Eubacterium and
21.78% to 25.89% of all sequences (Table 3). The Faecalibacterium, increase dramatically overtime in
differences in abundance of both Bacteroidetes and controls compared with cases, and together repre-
Firmicutes observed between cases and controls sent more than 13% of the total population of
were significant at all three time points. bacteria in autoimmune children at time point 3
(Supplementary Table 5).

Class and order levels

Similar trends occured at the class level for the Species level
two most abundant classes, the Bacteroidetes and At the 99% identity level, 59% and 49% of case and
the Clostridia (Supplementary Table 2). As the control sequences, respectively, were classified to
Bacteroidetes become more abundant in the case named and cultured species. Of the 377 bacterial
samples overtime, they became less abundant in the species identified in these samples, 51 species
control samples overtime. Conversely, as the abun- statistically differed in abundance between cases
dance of Clostridia decreased overtime in the case and controls with a P-value o0.01 in at least 1 time
samples, the Clostridia sequences increased in point (Table 4). At the species level, specific taxa
control samples. These trends repeated at the order make large contributions to the overall differences
level: the Bacteroidales sequences increased in case between cases and controls. Perhaps the most
samples, whereas they decreased in control samples striking example is Bacteroides ovatus. Nearly one-
and the Clostriales sequences increased in control fourth of the difference between cases and controls
samples, whereas they decreased in case samples. within the phylum Bacteroidetes can be explained
Four other orders differed at one time point or by this single species (Figure 1a). At the first time
another between cases and controls, but no clear point, there are slightly more B. ovatus sequences in
trends emerged overtime (Supplementary Table 3). control samples than in case samples, but, at the time
of autoimmunity, there are 16-fold more B. ovatus
sequences in cases than in controls (Table 4).
Family level Most other Bacteroides species are also present in
Analysis at the family level showed that three greater proportions in cases than in controls.
families among the Firmicutes, the Ruminococca- However, two Bacteorides species, B. vulgatus and
ceae, the Lachnospiraceae and the Eubacteriaceae, B. fragilis, are observed at much higher levels in
were significantly more abundant in controls than in controls than in cases at the third time point and
cases at time point 3. In contrast, Veillonellaceae, represent more than 11% of total sequences in the
also in the Firmicutes, were present at significantly control samples.
higher levels in cases than in controls at all time Another bacterium of interest is the human
points, but decline rapidly overtime in cases. firmicute CO19 (Table 4). Although not sufficiently
Bacteroidaceae was the dominant family of the characterized to be given a species name, this
Bacteroidetes phylum and was significantly more organism has been cultured (Hayashi et al., 2002).
abundant in cases than in controls (Supplementary At the 99% similarity level, more than 7% of all
Table 4). Among Bacteroidetes, the Porphyromona- control sequences at the third time point clustered
daceae sequences were present at greater levels in with human firmicute CO19. Although increasing in
controls than in cases at all time points. The both cases and controls overtime, the abundance of
Rikenellaceae family was also significantly more the human firmicute CO19 at the third time point
abundant in cases compared with controls at the seems to be nearly fourfold greater in controls than
second and third time points. in cases.
At the third collection time point, just 37 of 377
species differed significantly between cases and
Genus level controls (Table 4). The 15 species that are higher in
At all time points and in both cases and controls, abundance in controls represent 30% of all control
more than 92% and 81% of all case and control sequences at time point 3. Similarly, the 22 species
sequences, respectively, could be classified to the that are much higher in cases than in controls
genus level. Hence, the majority of organisms in represent nearly half of all sequences present in
these samples are well known to science, and their cases in autoimmune children.
physiology and morphology are understood. The
Bacteroides is by far the dominant genus in these
samples, representing over two-thirds of all se- The unclassified sequences
quences early in controls, and at the third time At all taxonomic levels, the sequences that cannot
point in cases. In cases, the number of Bacteroides be classified to known taxa are significantly more
increases overtime and is significantly higher than abundant in control samples than in cases at
that in controls at time points 2 and 3. In controls, the third time point (Figure 1). As expected, the
however, Bacteroides declines from 66.47% to number of unclassified sequences increases as the

The ISME Journal

 Characteristics of the autoimmune microbiome
A Giongo et al
87

Figure 1 Significant differences in taxa between cases (autoimmune) and controls (healthy). Samples were collected approximately 4
months, 1 year and 2 years after birth, represented, respectively, as time points 1, 2 and 3: (a) increasing numbers of Bacteroidetes in
cases overtime compared with controls; (b) increasing numbers of Firmicutes in controls overtime compared with cases; and (c) higher
proportion of unclassified sequences in controls compared with cases. Significant differences between cases and controls are designated
by a star (Pp0.002).

Figure 2 Bacterial community differences between cases and controls during autoimmunity development in cases; (a) significant
increase in Bacteroidetes with concomitant decrease in Firmicutes in cases compared with controls (P-value p0.01 at all time points;
(b) significantly higher (Po0.05) Shannon diversity index in controls compared with cases in time point 3. Significant differences
between cases and controls are designated by a star. The P-values for time points 1, 2 and 3 are (a) 0.0000, 0.0000 and 0.0000, and (b) 0.80,
0.33 and 0.03, respectively.

phylogenetic classification becomes more restrictive. In addition, the analysis of microbial commu-
Hence, the proportion of unclassified sequences at the nities using principal coordinate analysis shows
phylum level is o1% of all sequences, whereas the that the bacterial communities in control samples
proportion of unclassified sequences at the species are more similar to each other than are the bacterial
level is over 30% in some control samples. communities in case samples (Figures 3 and 4).
No differences in community diversity were observed
at time point 1 at the 10% confidence interval.
Community diversity indices However, the average distance between any pair of
At the genus level, bacterial diversity, as measured cases was significantly higher than between any pair
through the Shannon index, increases overtime of controls at the second and third collection points
in control samples (Figure 2). At the third time at the 5% and 10% level of confidence, respectively.
point, the diversity index is significantly higher Although these confidence intervals are relatively
in control communities than in case communities high, they are remarkable, given that only six pairwise
(P-value o0.05). comparisons were available in cases and control.

The ISME Journal

 Characteristics of the autoimmune microbiome
A Giongo et al
88

Figure 3 Histograms showing the permutation test based on the UniFrac significance obtained from the three time points (a, b and c).
Dashed blue lines represent the 0.10, 0.05 and 0.01 quantiles, and the red line indicates the value of the observed difference. No
differences in community diversity were observed at time point 1 at the 10% confidence interval. However, the average distance between
any pair of cases was significantly higher than that between any pair of controls at the second and third collection points at the 5% and
10% level of confidence, respectively. A summary of data over all time points is shown in (d). (The color version of this figure is available
in online version only).

Figure 4 Principal coordinate analysis for the case and control communities at time points 1 (a), 2 (b) and 3 (c).

Discussion and autoimmune microbiomes in children who

develope T1D. One striking result of this analysis
The data presented here were analyzed at various is the decline in Firmicutes and increase in
taxonomic levels and at the community level to Bacteroidetes in the gut microbiome overtime as
identify specific taxa and community characteristics children become autoimmune; in contrast, Firmi-
that differ between microbiomes in healthy children cutes increase as Bacteroidetes declines in healthy

The ISME Journal

 Characteristics of the autoimmune microbiome
A Giongo et al
89
children (Figures 1a, b and 2a). These two phyla less healthy than those of control children. These
comprise more than 80% of the sequences at all time data, however, do build a framework on which
points in both cases and controls. Phylogenetic specific questions can be asked regarding the
analysis revealed that virtually all changes that autoimmune microbiome. These three lines of
occurred overtime between cases and controls with- evidence are as follows:
in the Bacteroidetes phylum were attributed to a First, as children become older, changes in the
single genus, Bacteroides. More than one-fifth of the composition of the intestinal microbiota take
changes that occurred within the genus Bacteroides place depending on the feeding stage of childhood
can be ascribed to a single species, B. ovatus (Favier et al., 2002). Near 2 years of age, the micro-
(Figure 1a). Nearly all of the increased abundance biota of children become more similar to the adult
in Firmicutes in control samples compared with and present large changes thereafter (Tiihonen et al.,
cases can be attributed to a single order, Clostri- 2009). In this study, healthy children contain a
diales (Supplementary Table 3). Over 17% of that significantly higher number of poorly known,
increase is ascribed to a single species represented unclassified microorganisms than do autoimmune
in the literature by a single strain, human intestinal children overtime (Figure 1c). The presence of a
firmicute CO19 (Figure 1b), which was isolated from greater proportion of unclassified organisms in
the intestine of a healthy human subject (Hayashi microbiomes of healthy subjects suggests that these
et al., 2002). individuals may host a greater proportion of non-
The dysbiosis observed at phylum level between pathogenic organisms than do autoimmune subjects.
Bacteroidetes and Firmicutes in the human gut Historically, microbiology has focused on the
has been described in several human disorders. characterization of microorganisms that cause dis-
As described in previous studies, the ratio between ease, as well as their mechanisms of pathogenesis
Firmicutes and Bacteroidetes in human type 2 (de Kruif, 1926; Lechevalier and Solotorovsky,
diabetes declines compared with controls (Larsen 1974; Sapp, 2009). Thus, if a bacterium is found
et al., 2010). In Crohn’s disease, both Bacteroidetes that causes disease, the scientific community
and Firmicutes seem to decline, whereas Proteo- studies this organism immediately and with great
bacteria increase (Frank et al., 2007; Willing et al., intensity. As a result, if a bacterium is a pathogen, it
2009). The reverse was observed in obesity, the is far more likely to be classified and characterized
imbalance is observed by the reduction in the than if it is not pathogenic. Thus, if the autoimmune
Bacteroides proportion in obese human subjects, microbiome possesses more organisms with a
with a corresponding increase in the Firmicutes/ pathogenic potential than the healthy microbiome,
Bacteroidetes ratio. Among Firmicutes, Lactobacillus the number of unclassified sequences would be
numbers seem to increase in obese patients (Turnbaugh expected to be lower in autoimmune subjects
et al., 2006; Armougom et al., 2009). compared with healthy ones.
In all four pairs of cases and controls at the third Second, bacterial diversity increased in controls,
time point, 22 bacterial species were significantly but remained constant in cases (Figure 2b). The
more abundant (Po0.01) in cases than in controls. same phenomenon was described in Crohn’s dis-
Of those 22 species, 5 species (bacterium mpn- ease, in which decreased diversity of fecal micro-
isolate group 18, B. ovatus, Bacteroides sp. CJ78, biota of cases in comparison with controls
B. thetaiotaomicron and B. uniformis) represented (Manichanh et al., 2006; Sartor, 2008). A decrease
more than 1% of all sequences each. Similarly, 15 in microbial diversity in the gut was also observed
bacterial species were found to be significantly more in obese individuals in relation with lean indivi-
abundant (Po0.01) in controls compared with cases duals (Turnbaugh et al., 2009). This result may be
in all four pairs of children at the third time point. caused by a disturbance, likely an immunological
Of those species (Bacteroides fragilis, B. vulgatus, response, in autoimmune children that reduces the
Eubacterium eligens, E. rectale, Faecalibacterium microbial diversity in these case subjects. In natural
prausnitzii, human intestinal firmicute CB47 and systems, microbial communities are very sensitive
human intestinal firmicute CO19), each represented to disturbance and are not sufficiently resilient to
at least 1% of all sequences. Thus, this study return to their original states (Allison and Martiny,
identified highly abundant bacteria in the gut 2009). Reduced community composition decreases
microbiomes that are either negatively or positively the set of ecosystem processes available in this
correlated with the development of autoimmunity community. In the human intestine, limited diver-
in children who are at high risk for the onset of T1D. sity may lead to a reduced capacity to digest a
Three lines of evidence suggest that, overtime, diverse diet, leading to lower energy levels in
nonautoimmune children are able to build a healthy affected individuals. Hence, a gut microbial com-
and stable gut microbiome, whereas the micro- munity with less diversity may lead to less healthy
biomes of autoimmune children are less diverse individuals or may be an indicator of an unhealthy
and unstable. Hence, we now refer to this unhealthy, state.
unstable microbiome as the autoimmune micro- Finally, the mean molecular distance between any
biome for T1D. The evidence presented here does two case samples is greater than the molecular
not prove that the microbiomes of case children are distance between any two control samples collected

The ISME Journal

 Characteristics of the autoimmune microbiome
A Giongo et al
90
at any of the time points (Figures 3 and 4). Thus, Armougom F, Henry M, Vialettes B, Raccah D, Raoult D.
although control communities are more diverse than (2009). Monitoring bacterial community of human gut
case communities, overall, control communities are microbiota reveals an increase in Lactobacillus in
much more similar to each other and show a obese patients and Methanogens in anorexic patients.
PLOS One 4: e7125.
community stability that is lacking in case commu-
Brugman S, Klatter FA, Visser JTJ, Wildeboer-Veloo ACM,
nities. This community difference between any two Harmsen HJM, Rozing J et al. (2006). Antibiotic
case samples suggests that the autoimmune com- treatment partially protects against type 1 diabetes
munities are much less stable. in the Bio-Breeding diabetes-prone rat. Is the gut
Of all the differences observed between autoim- flora involved in the development of type 1 diabetes?
mune and healthy microbiomes, there are two Diabetologia 49: 2105–2108.
indicators of the autoimmune microbiome that can Calcinaro F, Dionisi S, Marinaro M, Candeloro P, Bonato V,
be observed at the earliest time points, before the Marzotti S et al. (2005). Oral probiotic administration
first appearance of autoantibodies in case subjects. induces interleukin-10 production and prevents
The first of these characteristics is the instability of spontaneous autoimmune diabetes in the non-obese
diabetic mouse. Diabetologia 48: 1565–1575.
the autoimmune microbiome. The second character- de Kruif P. (1926). Microbe Hunters. Harcourt, Brace and
istic is the high ratio of Firmicutes to Bacteroidetes Co.: New York.
in cases, which is observed within the first 6 months Edgar RC. (2004). MUSCLE: a multiple sequence align-
after birth, compared with the low ratio found ment method with reduced time and space complex-
in controls. Both these characteristics may be early ity. Bioinformatics 5: 1–19.
diagnostic markers of pending autoimmunity that Favier CF, Vaughan EE, De Vos WM, Akkermans ADL.
can be observed before the first appearance of (2002). Molecular monitoring of succession of bac-
autoantibodies in serum. This knowledge may allow terial communities in human neonates. Appl Environ
early interventions that can delay or prevent auto- Microb 68: 219–226.
immunity in children by alerting the intestinal Frank DN, St Amand AL, Feldman RA, Boedeker EC,
Harpaz N, Pace NR et al. (2007). Molecular-phylogenetic
microbiome or by other means. characterization of microbial community imbalance in
Taken collectively, these findings support the human inflammatory bowel diseases. Proc Natl Acad
concept of an autoimmune microbiome for T1D. Sci USA 104: 13780–13785.
Although the number of samples used here is Giongo A, Crabb DB, Davis-Richardson AG, Chauliac D,
too low to make any firm conclusions, these data Mobberley JM, Gano KA et al. (2010a). PANGEA:
suggest that the autoimmune microbiome for this pipeline for analysis of next generation amplicons.
disease tends to have more classified members ISME J 4: 852–861.
but decreased diversity and reduced stability when Giongo A, Davis-Richardson AG, Crabb DB, Triplett EW.
compared with a healthy microbiome. The forces (2010b). TaxCollector: tools to modify existing 16S
at work that result in the autoimmune microbiome rRNA databases for the rapid classification at six
taxonomic levels. Submitted.
are unknown, but may be related to the breeching Harrison LC, Honeyman MC, Morahan G, Wentworth JM,
of the epithelial layer combined with abnormal Elkassaby S, Colman PG et al. (2008). Type 1 diabetes:
immune responsiveness. This work provides a lessons for other autoimmune diseases? J Autoimmun
model of the autoimmune microbiome for T1D and 31: 306–310.
delineates four features that can serve as spring- Hayashi H, Sakamoto M, Benno Y. (2002). Phylogenetic
boards for further analysis. The defining charac- analysis of the human gut microbiota using 16S rDNA
teristics of the stable, healthy microbiome and the clone libraries and strictly anaerobic culture-based
unstable, unhealthy autoimmune microbiomes are methods. Microbiol Immunol 46: 535–548.
reminiscent of the famous first line from Tolstoy0 s Kupila A, Muona P, Simell T, Arvilommi P, Savolainen H,
Anna Karenina: ‘All happy families are alike; each Hamalainen AM et al. (2001). Feasibility of genetic
and immunological prediction of Type I diabetes
unhappy family is unhappy in its own way.’ in a population-based birth cohort. Diabetologia 44:
290–297.
Larsen N, Vogensen FK, van den Beg FWL, Nielsen DS,
Acknowledgements Andreasen AS, Pedersen BK et al. (2010). Gut
microbiota in human adults with type 2 diabetes
This research was supported by the Florida Agricultural differs from non-diabetic adults. PLoS One 5: e9085.
Experiment Station. We thank Professor Vassiliki Betty Lechevalier HA, Solotorovsky M. (1974). Three Centuries
Smocovitis of the University of Florida for her advice on of Microbiology. Dover Publications: New York.
the history of microbiology. Li W, Godzik A. (2006). CD-HIT: a fast program
for clustering and comparing large sets of protein
or nucleotide sequences. Bioinformatics 22:
1658–1659.
Lozupone C, Hamady M, Knight R. (2006). UniFrac—An
References online tool for comparing microbial community
diversity in a phylogenetic context. BMC Bioinfor-
Allison SD, Martiny JBH. (2009). Resistance, resilience, matics 7: 371.
and redundancy in microbial communities. Proc Natl Manichanh C, Rigottier-Gois L, Bonnaud E, Gloux J,
Acad Sci USA 105: 11512–11519. Pelletier E, Frangeul L et al. (2006). Reduced diversity

The ISME Journal

 Characteristics of the autoimmune microbiome
A Giongo et al
91
of faecal microbiota in Crohn’s disease revealed by a Tiihonen K, Ouwehand AC, Rautonen N. (2009). Human
metagenomic approach. Gut 55: 205–211. intestinal microbiota and healthy ageing. Ageing Res
Matsuzaki T, Nagata Y, Kado S, Uchida K, Kato I, Rev 9: 107–116.
Hashimoto S et al. (1997). Prevention of onset in an The TEDDY Study Group (2007). The Environmental
insulin-dependent diabetes mellitus model, NOD Determinants of Diabetes in the Young (TEDDY) study:
mice, by oral feeding of Lactobacillus casei. APMIS study design. Pediatr Diabetes 8: 286–298.
105: 643–649. Turnbaugh PJ, Ley RE, Mahowald MA, Magrini V, Mardis
Nejentsev S, Sjoroos M, Soukka T, Knip M, Simell O, ER, Gordon JI. (2006). An obesity-associated gut
Lovgren T et al. (1999). Population-based genetic microbiome with increased capacity for energy
screening for the estimation of Type 1 diabetes harvest. Nature 444: 1027–1031.
mellitus risk in Finland: selective genotyping of Turnbaugh PJ, Hamady M, Yatsunenko T, Cantarel BL,
markers in the HLA-DQB1, HLA-DQA1 and HLA- Duncan A, Ley RE et al. (2009). A core gut microbiome
DRB1 loci. Diabetic Med 16: 985–992. in obese and lean twins. Nature 457: 480–485.
Roesch RFW, Casella G, Simell O, Krischer J, Wasserfall Vaarala O, Atkinson MA, Neu J. (2008). The ‘Perfect
CH, Schatz D et al. (2009a). Influence of sample Storm’ for type 1 diabetes. The complex interplay
storage on bacterial community diversity in fecal between intestinal microbiota, gut permeability, and
samples. Open Microbiol J 3: 40–46. mucosal immunity. Diabetes 57: 2555–2562.
Roesch RFW, Lorca GL, Casella G, Giongo A, Naranjo A, Valladares R, Sankar D, Li N, Williams E, Mahmou ASA,
Pionzo AM et al. (2009b). Culture-independent iden- Lai K-K et al. (2010). Lactobacillus johnsonii N6.2
tification of gut bacteria correlated with the onset of mitigates the development of type 1 diabetes in BB-DP
diabetes in a rat model. ISME J 3: 536–548. rats. PLoS One 5: e10507.
Sapp J. (2009). The new foundations of evolution: on the Wen L, Ley RE, Volchkov PY, Stranges PB, Avanesyan L,
tree of life. Oxford University Press: New York. Stonebraker AC et al. (2008). Innate immunity and
Sartor RB. (2008). Microbial influences in inflammatory intestinal microbiota in the development of Type 1
bowel diseases. Gastroenterology 134: 577–594. diabetes. Nature 455: 1109–1113.
Schwartz RF, Neu J, Schatz D, Atkinson MA, Wasserfall C. Willing B, Halfvarson J, Dicksved J, Rosenquist M, Jarnerot
(2007). Comment on: Brugman S et al. (2006). G, Engstrand L et al. (2009). Twin studies reveal
Antibiotic treatment partially protects against type 1 specific imbalances in the mucosa-associated micro-
diabetes in the Bio-Breeding diabetes-prone rat. Is the biota of patients with ileal Crohn’s disease. Inflamm
gut flora involved in the development of type 1 Bowel Dis 15: 653–660.
diabetes? Diabetologia 49:2105-2108. Diabetologia 50: Yadav H, Jain S, Sinha PR. (2007). Antidiabetic effect of
220–221. probiotic dahi containing Lactobacillus acidophilus
Shannon CE, Weaver W. (1949). The mathematical theory of and Lactobacillus casei in high fructose fed rats.
communication. Urbana IL: University of Illinois Press. Nutrition 23: 62–68.

Supplementary Information accompanies the paper on The ISME Journal website (http://www.nature.com/ismej)

The ISME Journal