A metagenome-wide association study of gut microbiota in type 2 diabetes

Abstract

Assessment and characterization of gut microbiota has become a major research area in human disease, including type 2 diabetes, the most prevalent endocrine disease worldwide. To carry out analysis on gut microbial content in patients with type 2 diabetes, we developed a protocol for a metagenome-wide association study (MGWAS) and undertook a two-stage MGWAS based on deep shotgun sequencing of the gut microbial DNA from 345 Chinese individuals. We identified and validated approximately 60,000 type-2-diabetes-associated markers and established the concept of a metagenomic linkage group, enabling taxonomic species-level analyses. MGWAS analysis showed that patients with type 2 diabetes were characterized by a moderate degree of gut microbial dysbiosis, a decrease in the abundance of some universal butyrate-producing bacteria and an increase in various opportunistic pathogens, as well as an enrichment of other microbial functions conferring sulphate reduction and oxidative stress resistance. An analysis of 23 additional individuals demonstrated that these gut microbial markers might be useful for classifying type 2 diabetes.

Full text

ARTICLE doi:10.1038/nature11450
A metagenome-wide association study of
gut microbiota in type 2 diabetes
Junjie Qin
1
*, Yingrui Li
1
*, Zhiming Cai
2
*, Shenghui Li
1
*, Jianfeng Zhu
1
*, Fan Zhang
3
*, Suisha Liang
1
, Wenwei Zhang
1
,
Yuanlin Guan
1
, Dongqian Shen
1
, Yangqing Peng
1
, Dongya Zhang
1
, Zhuye Jie
1
, Wenxian Wu
1
, Youwen Qin
1
, Wenbin Xue
1
,
Junhua Li
1
, Lingchuan Han
3
, Donghui Lu
3
, Peixian Wu
3
,YaliDai
3
, Xiaojuan Sun
2
, Zesong Li
2
, Aifa Tang
2
, Shilong Zhong
4
,
Xiaoping Li
1
, Weineng Chen
1
,RanXu
1
, Mingbang Wang
1
, Qiang Feng
1
, Meihua Gong
1
, Jing Yu
1
, Yanyan Zhang
1
, Ming Zhang
1
,
Torben Hansen
5
, Gaston Sanchez
6
, Jeroen Raes
7,8
, Gwen Falony
7,8
, Shujiro Okuda
7,8
, Mathieu Almeida
9
,
Emmanuelle LeChatelier
9
, Pierre Renault
9
, Nicolas Pons
9
, Jean-Michel Batto
9
, Zhaoxi Zhang
1
, Hua Chen
1
, Ruifu Yang
1,10
,
Weimou Zheng
1
, Songgang Li
1
, Huanming Yang
1
, Jian Wang
1
, S. Dusko Ehrlich
9
, Rasmus Nielsen
6
, Oluf Pedersen
5,11,12
,
Karsten Kristiansen
1,13
& Jun Wang
1,5,13
Assessment and characterization of gut microbiota has become a major research area in human disease, including type 2
diabetes, the most prevalent endocrine disease worldwide. To carry out analysis on gut microbial content in patients
with type 2 diabetes, we developed a protocol for a metagenome-wide association study (MGWAS) and undertook a
two-stage MGWAS based on deep shotgun sequencing of the gut microbial DNA from 345 Chinese individuals. We
identified and validated approximately 60,000 type-2-diabetes-associated markers and established the concept of a
metagenomic linkage group, enabling taxonomic species-level analyses. MGWAS analysis showed that patients with
type 2 diabetes were characterized by a moderate degree of gut microbial dysbiosis, a decrease in the abundance of some
universal butyrate-producing bacteria and an increase in various opportunistic pathogens, as well as an enrichment of
other microbial functions conferring sulphate reduction and oxidative stress resistance. An analysis of 23 additional
individuals demonstrated that these gut microbial markers might be useful for classifying type 2 diabetes.
Type 2 diabetes (T2D), which is a complex disorder influenced by both
genetic and environmental components, has become a major public
health issue throughout the world
1,2
. Currently, research to parse the
underlying genetic contributors to T2D is mainly through the use of
genome-wide association studies (GWAS) focusing on identifying
genetic components in the organism’s genome
3,4
. Recently, research
has indicated that the risk of developing T2D may also involve factors
from the ‘other genome’, that is, the ‘intestinal microbiome’ (also
termed the gut metagenome)
5
.
Previous metagenomic research on the gut metagenome, primarily
using 16S ribosomal RNA
6
and whole-genome shotgun (WGS)
sequencing
7
, has provided an overall picture of commensal microbial
communities and their functional repertoire. For example, a catalogue
of 3.3 million human gut microbial genes were established in 2010
(ref. 8) and, of note, a more extensive catalogue of gut microorganisms
and their genes were published later
9,10
. Recent research on the gut
metagenome has changed our understanding of human disease and
its potential medical impact as many studies have reported. From the
perspective of both taxonomic and functional composition, the gut
microbiota might be linked to and contribute to many complex
diseases
11
. For example, several studies have indicated that obesity is
associated with an increase in the phylum Firmicutes and a relatively
lower abundance of the phylum Bacteroidetes
7,12–16
. Crohn’s disease
research has revealed that patients had a significant reduction in
the overall diversity of the gut microbiota
17
and had changes in
microbial composition
18
, and a T2D study showed that the proportion
of the phylum Firmicutes and the class Clostridia in the gut of patients
was significantly reduced
19
. However, more work is required to gain
detailed information about gut microbial compositional changes and
their associated impact with these types of diseases, and additional
tools are required to find ways to determine associated changes easily
and rapidly.
To reach these initial goals, we devised and carried out a two-stage
case-control metagenome-wide association study (MGWAS) based
on deep next-generation shotgun sequencing of DNA extracted from
the stool samples from a total of 345 Chinese T2D patients and non-
diabetic controls. From this we pinpointed specific genetic and func-
tional components of the gut metagenome associated with T2D
(Supplementary Fig. 1). Our data provide insight into the character-
istics of the gut metagenome related to T2D risk, a paradigm for future
studies of the pathophysiological role of the gut metagenome in other
relevant disorders, and the potential usefulness for a gut–microbiota-
based approach for assessment of individuals at risk of such disorders.
Construction of a gut metagenome reference
To identify metagenomic markers associated with T2D, we first
developed a comprehensive metagenome reference gene set that
included genetic information from Chinese individuals and T2D-
specific gut microbiota, as the currently available metagenomic ref-
erence (the MetaHIT gene catalogue) did not include such data. We
*These authors contributed equally to this work.
1
BGI-Shenzhen, Shenzhen 518083, China.
2
Shenzhen Second People’s Hospital, The First Affiliated Hospital of Shenzhen University, Shenzhen 518035, China.
3
Peking University Shenzhen Hospital,
Shenzhen 518036, China.
4
Medical Research Center of Guangdong General Hospital, Guangdong Academy of Medical Sciences, Guangzhou 510080, China.
5
The Novo Nordisk Foundation Center for
Basic Metabolic Research, Faculty of Health Sciences, University of Copenhagen, DK-2100 Copenhagen, Denmark.
6
Department of Integrative Biology and Department of Statistics, University of California
Berkeley, Berkeley, CA 94820, USA.
7
Department of Structural Biology, VIB, 1050 Brussels, Belgium.
8
Department of Applied Biological Sciences (DBIT), Vrije Universiteit Brussel, 1050 Brussels, Belgium.
9
Institut National de la Recherche Agronomique, 78350 Jouy en Josas, France.
10
State Key Laboratory of Pathogen and Biosecurity, Beijing Institute of Microbiology and Epidemiology, Beijing 100071,
China.
11
Institute of Biomedical Sciences, University of Copenhagen & Faculty of Health Science, University of Aarhus, DK-8000 Aarhus, Denmark.
12
Hagedorn Research Institute, DK-2820 Gentofte,
Denmark.
13
Department of Biology, University of Copenhagen, DK-2200 Copenhagen, Denmark.
4OCTOBER2012|VOL490|NATURE|55
Macmillan Publishers Limited. All rights reserved
©2012

carried out WGS sequencing on individual faecal DNA samples from
145 Chinese individuals (71 cases and 74 controls, Supplementary
Table 1) and obtained an average of 2.61 gigabases (Gb) (15.8 million)
paired-end reads for each, totalling 378.4 Gb of high-quality data that
was free of human DNA and adaptor contaminants (Supplementary
Table 2). We then performed de novo assembly and metagenomic
gene prediction for all 145 samples. We integrated these data with
the MetaHIT gene catalogue, which contained 3.3 million genes that
were predicted from the gut metagenomes of individuals of European
descent, and obtained an updated gene catalogue with 4,267,985 pre-
dicted genes. A total of 1,090,889 of these genes were uniquely
assembled from our Chinese samples, which contributed 10.8% addi-
tional coverage of sequencing reads when comparing our data against
that from the MetaHIT gene catalogue alone (Supplementary Fig. 2).
Having a more complete gene reference, we carried out taxonomic
assignment and functional annotation for the updated gene catalogue
using 2,890 reference genomes (IMG v3.4; Supplementary Table 3),
KEGG (Release 59.0) and eggNOG databases (v3). Here, 21.3% of the
genes in the updated catalogue could be robustly assigned to a genus,
which covered 26.4%–90.6% (61.2% on average) of the sequencing
reads in the 145 samples (Supplementary Methods); the remaining
genes were likely to be from currently undefined microbial species.
For assessment at a functional level, we identified 6,313 KEGG ortho-
logues and 38,641 eggNOG orthologue groups in the updated gene
catalogue, which covered 47.1% and 60.9%, respectively, ofthe genes in
the catalogue. In addition, 14.0% of genes that were not mapped to
eggNOG orthologue groups could be clustered into 7,042 novel gene
families; however, these do not yet have any functional annotation
information, but were still included (as in-house eggNOG orthologue
groups) in our analyses. For each metagenomic sample, on average,
48.7% and 68.8% sequencing reads were covered, respectively, by these
KEGG orthologues- andeggNOG orthologue groups-annotated genes.
Marker identification using a two-stage MGWAS
To define T2D-associated metagenomic markers, we devised and
carried out a two-stage MGWAS strategy. Using a sequence-based
profiling method, we quantified the gut microbiota in the 145 samples
for use in stage I. On average, with the requirement that there should
be $90% identity, we could uniquely map 77.4 60.6% (mean 6s.e.m.;
n5145) paired-end reads to the updated gene catalogue (Supplemen-
tary Fig. 2 and Supplementary Table 2). To normalize the sequencing
coverage, we used relative abundance instead of the raw read count
to quantify the gut microbial genes (Supplementary Methods). With
nearly 16 million sequencing reads on average per sample, our
sequence-based profiling method could reliably detect very low-
abundance genes. For example, given a gene with a real relative abund-
ance of 1 310
26
, the detected value ranged from 0.7 310
26
to
1.5 310
26
based on a theoretical estimation (Supplementary Fig. 3).
To facilitate the subsequent statistical analyses at both genetic and
functional levels, we further defined and prepared three types of
profiles using the quantified gene results: (1) a gene profile; (2) a
KEGG orthologues profile; and (3) an eggNOG orthologue groups
profile (Supplementary Methods).
We investigated the subpopulations of the 145 samples in these
different profiles. Applying the same identification method as used
in the MetaHIT study
20
, we identified three enterotypes in our
Chinese samples (Supplementary Figs 4 and 5). A principal component
analysis (PCA) showed that these three enterotypes were primarily
made up of several highly abundant genera, including Bacteroides,
Prevotella,Bifidobacterium and Ruminococcus (Fig. 1a). However,
we found no significant relationship between enterotype and T2D
disease status (P50.29, Fisher’s exact test). We examined the top five
principal components (Pvalue in Tracy–Widom test ,0.05 and con-
tribution .3%): the first and second principal components were sig-
nificantly correlated with enterotype (P,0.001, Kruskal–Wallis test),
and the fifth principal component was significantly correlated with
T2D (P,0.001, Wilcoxon rank-sum test; Supplementary Fig. 5d),
indicating that T2D, in addition to enterotype, was a determining
factor in explaining the gut microbial differences in our samples.
The third and fourth principal components, however, did not correlate
with any known factors.
We then corrected for population stratification, which might be
related to the non-T2D-related factors. For this we analysed our data
using a modified EIGENSTRAT method
21
; however, unlike what is
done in a GWAS subpopulation correction, we applied this analysis
to microbial abundance rather than to genotype. For gene profile, after
adjustment, we found that the effects that correlated with non-T2D-
related factors disappeared (Supplementary Table 4). A Wilcoxon
rank-sum test was done on the adjusted gene profile to identify differ-
ential metagenomic gene content between the T2D patients and con-
trols. The outcome of our analyses showed a substantial enrichment of
a set of microbial genes that had very small Pvalues, as compared with
the expected distribution under the null hypothesis (Fig. 1b), indi-
cating that these genes were true T2D-associated gut microbial genes.
To validate the significant associations identified in stage I, we
carried out the stage II analysis using an additional 200 Chinese
individuals (one of these samples had a very low within-sample
diversity, which was probably owing to the presence of a high fraction
of Escherichia and Klebsiella, and was therefore excluded in later
analyses; Supplementary Tables 1 and 2). We also used WGS sequen-
cing in stage II and generated a total of 830.8 Gb sequence data with
23.6 million paired-end reads on average per sample. We then assessed
the 278,167 stage I genes that had Pvalues ,0.05 and found that the
majority of these genes still correlated with T2D in these stage II study
samples (Supplementary Fig. 6). We next controlled for the false
discovery rate (FDR) in the stage II analysis, and defined a total of
52,484 T2D-associated gene markers from these genes corresponding
to a FDR of 2.5% (stage II Pvalue ,0.01; Fig. 1c, Supplementary Fig. 7
and Supplementary Table 5).
We applied the same two-stage analysis using the KEGG orthologues
and eggNOG orthologue groups profiles and identified a total of 1,345
KEGG orthologues markers (stageII P,0.05 and 4.5% FDR) and 5,612
eggNOG orthologue groups markers (stage II P,0.05 and 6.6% FDR)
that were associated with T2D (Supplementary Tables 6 and 7).
Development of a metagenomic linkage group
To reduce and structurally organize the abundant metagenomic data
and to enable us to make a taxonomic description, we devised the
Null hypothesis
Null hypothesis
ab
P values
Density
0.0 0.2 0.4 0.6 0.8 1.0
0
1
2
3
4
•
•
•••
•
•
•
•
•
•
•
•
•
•
•
•
••
•
•
•
•
•
••
••
•
•
••
•
•
••
•
••
••
•
•
•
••
•
•
•
•
•
•
•
•
••
••
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
Bacteroides
•
Bifidobacterium
Bifidobacterium
Ruminococcus
Ruminococcus
Bifidobacterium
Ruminococcus
Prevotella
Prevotella
Prevotella
PC1
PC2
0.0 0.4 0.8
0
2
4
6
0
0.4
0.8
c
Null hypothesis
Estimated FDR
Estimated power
•
T2D patients
Controls
Figure 1
|
Identification of T2D-associated markers from gut metagenome.
a, The T2D patients (n571) and controls (n574)from stage I were plotted on
the first two principal components of the genus profile. Lines connect
individuals determined to have the same enterotype (usingthe PAM clustering
method of refs 20,36), and coloured circles cover the individuals near the centre
of gravity for each cluster (,1.5s). The top four genera as the main
contributors to these clusters were determined and plotted by their loadings in
these two components. b, Density histogram showing the P-value distribution
of all genes tested in stage I. The horizon line represents the distribution of P
values under the null hypothesis. c, Density histogram showing the P-value
distribution of genes in stage II, which were identified from stage I. The blue
and red curves denote the estimated statistical power and false discovery rate
(FDR), respectively, for a particular Pvalue.
RESEARCH ARTICLE
56 | NATURE | VOL 490 | 4 OCTOBER 2012
Macmillan Publishers Limited. All rights reserved
©2012

generalized concept of metagenomic linkage group (MLG) in lieu of a
species concept for a metagenome. Here a MLG is defined as a group
of genetic material in a metagenome that is probably physically linked
as a unit rather than being independently distributed; this allowed us
to avoid the need to completely determine the specific microbial
species present in the metagenome, which is important given there
are a large number of unknown organisms and that there is frequent
lateral gene transfer (LGT) between bacteria. Using our gene profile,
we defined and identified a MLG as a group of genes that co-exists
among different individual samples and has a consistent abundance
level and taxonomic assignment (Supplementary Methods).
To assess the reliability of our MLG identifying method, we first
constructed a subset of bacterial genes from the updated metagenome
gene catalogue (n5130,605) that were independently derived from
50 known gut bacterial species (Supplementary Methods). We used a
threshold for the minimum gene number for a MLG of 100, above
which all 50 bacterial species could be identified with an average
genome coverage of 83.0% and with an accuracy in the taxonomic
classification of genes in the constructed subset of 99.8% (Supplemen-
tary Fig. 8 and Supplementary Table 8).
We identified 47 MLGs in the T2D-associated gene markers, which
covered 84.4% of these markers (Supplementary Table 9). Of these, 17
MLGs could be assigned to known bacterial species on the basis of
strong alignment sequence similarity with sequenced bacterial
genomes at the nucleotide level (Table 1). Using the taxonomic char-
acterization from these MLGs, we found that almost all of the MLGs
enriched in the control samples were from various butyrate-
producing bacteria, including Clostridiales sp. SS3/4, Eubacterium
rectale,Faecalibacterium prausnitzii,Roseburia intestinalis and
Roseburia inulinivorans. By contrast, most of T2D-enriched MLGs
were from opportunistic pathogens, such as Bacteroides caccae,
Clostridium hathewayi,Clostridium ramosum,Clostridium symbiosum,
Eggerthella lenta and Escherichia coli, which have previously been
reported to cause or underlie human infections such as bacteraemia
and intra-abdominal infections
22–25
. Of interest, the known mucin-
degrading species Akkermansia muciniphila and sulphate-reducing
species Desulfovibrio sp. 3_1_syn3 were also enriched in T2D samples.
The MLGs that were of unknown species origin will be of interest for
isolation and analysis in future studies to obtain information on their
relevant taxonomy.
A co-occurrence network on these MLGs was generated to assess
potential relationships between the T2D-associated gut bacteria
(Fig. 2a and Supplementary Methods). In this result, some types of
butyrate-producers, from clostridial cluster XIVa and IV, showed a
positive correlation with one another and were negatively correlated
with a group of the T2D-enriched bacteria from Clostridium, which
may indicate an antagonistic relationship between these different
clostridial clusters. Another interesting finding was the presence of
a small MLG from Haemophilus parainflu enzae, which is not a butyrate-
producer but was significantly enriched in the control samples, even in
an independent analysis comparing the coverage of its sequenced
bacterial genome (the highest genome coverage in all samples was
94.5%; P,0.001 between case and control groups, Student’s t-test).
In the co-occurrence network, this MLG was clearly separate from the
cluster of butyrate producers, and may have an unknown antagonistic
relationship with a T2D-enriched bacterium that is unknown but
appears closely related to the Subdoligranulum genus. These data
presented various patterns indicating relationships between the
T2D-associated gut bacteria and suggested it may be important to
determine, in a case-by-case manner, the different roles gut bacteria
may have in maintaining or interacting with their environment.
Functional characterization related to T2D
Using the T2D-associated KEGG orthologues and eggNOG ortholo-
gue groups markers, we assessed the potential microbial functional
roles in the gut microbiota of T2D patients. In general T2D-enriched
markers were typically involved in the KEGG categories of membrane
transport (P,0.001, Fisher’s exact test). This result is consistent with
Table 1
|
The list of T2D-associated MLGs that could be assigned to previously known phylotypes
MLG ID No. of genes Pvalues*Odds ratios (95% CI){Taxonomy assignment (level) Percentage similarity{
Stage I Stage II
T2D-enriched
T2D-154 337 0.0014 2.54 310
24
1.52 (1.05, 2.19) Akkermansia muciniphila 98.2
T2D-140 148 3.97 310
24
0.0029 1.50 (1.15, 1.97) Bacteroides intestinalis 98.2
T2D-139 3,386 0.0013 2.11 310
24
1.66 (1.26, 2.20) Bacteroides sp. 20_3 99.3
T2D-11 5,113 4.16 310
28
7.58 310
25
5.89 (1.39, 25.0) Clostridium bolteae 99.4
T2D-5 2,378 4.21 310
25
1.97 310
26
23.1 (2.08, 257) Clostridium hathewayi 99.3
T2D-80 2,381 1.30 310
24
1.41 310
25
1.68 (0.97, 2.89) Clostridium ramosum 99.8
T2D-57 821 4.00 310
27
2.21 310
25
2.62 (1.14, 6.03) Clostridium sp. HGF2 99.6
T2D-15 2,492 4.74 310
25
2.97 310
24
1.13 (0.88, 1.44) Clostridium symbiosum 99.6
T2D-1 949 6.01 310
24
0.0036 1.41 (0.93, 2.13) Desulfovibrio sp. 3_1_syn3 98.0
T2D-7 1,056 6.01 310
24
2.80 310
24
1.57 (0.95, 2.58) Eggerthella lenta 99.6
T2D-137 425 6.71 310
27
0.0012 1.72 (1.16, 2.57) Escherichia coli 99.0
T2D-165 131 0.0096 0.0017 1.46 (1.07, 1.99) Alistipes (genus) 99.51
T2D-12 364 4.52 310
26
8.04 310
28
2.22 (1.12, 4.40) Clostridium (genus) 91.0
T2D-8 5,272 7.08 310
210
9.95 310
26
1.12 (0.86, 1.45) Clostridium (genus) 88.8
T2D-93 1,590 2.01 310-4 0.0020 1.84 (1.03, 3.29) Parabacteroides (genus) 80.51
T2D-62 2,584 7.63 310
26
6.88 310
24
2.41 (1.43, 4.08) Subdoligranulum (genus) 98.71
T2D-2 2,430 3.14 310
25
0.0019 4.06 (1.28, 12.9) Lachnospiraceae (family) 97.31
Control-enriched
Con-107 1,677 1.12 310
27
0.0018 1.44 (1.13, 1.84) Clostridiales sp. SS3/4 98.0
Con-112 232 0.0064 1.99 310
24
1.51 (1.13, 2.03) Eubacterium rectale 97.6
Con-129 1,440 0.0033 0.0010 1.55 (1.19, 2.00) Faecalibacterium prausnitzii 98.2
Con-166 273 3.80 310
25
1.94 310
24
1.25 (0.93, 1.69) Haemophilus parainfluenzae 94.8
Con-121 3,507 6.11 310
25
4.90 310
26
3.10 (1.92, 5.03) Roseburia intestinalis 98.9
Con-113 345 2.85 310
24
9.72 310
24
1.45 (1.11, 1.89) Roseburia inulinivorans 98.2
Con-120 116 1.90 310
24
5.41 310
24
1.55 (1.17, 2.06) Eubacterium (genus) 89.0
Con-130 670 0.0134 0.0018 1.59 (1.21, 2.08) Faecalibacterium (genus) 89.4
Con-131 202 8.99 310
24
0.0017 1.58 (1.16, 2.15) Faecalibacterium (genus) 96.9
Con-133 1,555 3.43 310
25
0.0015 1.52 (1.15, 2.01) Erysipelotrichaceae (family) 66.91
Con-109 378 0.0135 1.67 310
24
1.41 (1.09, 1.83) Clostridiales (order) 87.0
*The stage I Pvalue was calculated after adjustment for population structures, stage II Pvalue was one-side.
{Calculated by logistic model.
{Similarity at nucleic acid level or, when marked with 1at the protein level.
ARTICLE RESEARCH
4 OCTOBER 2012 | VOL 490 | NATURE | 57
Macmillan Publishers Limited. All rights reserved
©2012

the previous findings in studies of inflammatory bowel disease and
obese patients
26
. By contrast, control-enriched markers were fre-
quently involved in cell motility and metabolism of cofactors and
vitamins (P,0.002; Supplementary Fig. 9).
At the module or pathway level, the gut microbiota of T2D patients
was functionally characterized with our T2D-associated markers and
showed enrichment in membrane transport of sugars, branched-chain
amino acid (BCAA) transport, methane metabolism, xenobiotics
degradation and metabolism, and sulphate reduction. By contrast,
there was a decrease in the level of bacterial chemotaxis, flagellar
assembly, butyrate biosynthesis and metabolism of cofactors and
vitamins (Fig. 2b and Supplementary Table 10; see Supplementary
Fig. 10 for the detailed information on butyrate-CoA transferase).
Some important functions, including butyrate biosynthesis and sul-
phate reduction, coincided with the T2D-associated bacteria identified
in the MLG analysis. The butyrate-producing bacteria seemed to be the
primary contributors to the cell motility functions (Supplementary
Table 11), potentially indicating some functional enrichment might
be related to the presence of specific species enrichment.
We found that seven of the T2D-enriched KEGG orthologues
markers were related to oxidative stress resistance, including catalase
(K03781), peroxiredoxin (K03386), Mn-containing catalase (K07217),
glutathione reductase (NADPH) (K00383), nitric oxide reductase
(K02448), putative iron-dependent peroxidase (K07223), and cyto-
chrome cperoxidase (K00428), but none of the identified control-
enriched KEGG orthologues markers had similar types of function.
This may indicate that the gutenvironment of a T2D patient is one that
stimulates bacterial defence mechanisms against oxidative stress
(Supplementary Table 10). Similarly, we found 14 KEGG orthologues
markers related to drug resistance that were greatly enriched in T2D
patients, further supporting that T2D patients may have a more hostile
gut environment, andthe medical histories of these patients mayreflect
this (Supplementary Table 10).
T2D-related dysbiosis in gut microbiota
In light of the above MGWAS result and an additional
PERMANOVA
27
(permutational multivariate analysis of variance)
analysis that clearly showed that T2D was a significant factor for
explaining the variation in the examined gut microbial samples
(Supplementary Table 12), we deduced that the gut microbiota in
T2D patients featured dysbiosis, which is a state where the balance
of the normal microbiota has been disturbed. However, the degree of
this T2D-related dysbiosis was moderate, because only 3.8 60.2%
(mean 6s.e.m.; n5344) of the gut microbial genes (at the relative
abundance level) were associated with T2D in an individual.
Additionally, we did not observe a significant difference in the
within-sample diversity between T2D and control groups (Fig. 3a).
Specifically, the degree of gut microbiota change in T2D was not as
substantial as that seen in inflammatory bowel disease (from the
MetaHIT samples
8
; see Fig. 3a) or enterotypes (Supplementary Fig. 11).
A similar result using the eggNOG orthologue groups profile sup-
ported the same conclusion (Supplementary Fig. 12).
a
b
Desulfovibrio
Desulfovibrio
sp. 3_1_syn3
sp. 3_1_syn3
Desulfovibrio sp. 3_1_syn3
E. coli
E. coli
E. coli
A. muciniphila
A. muciniphila
A. muciniphila
Con-142
Con-142
Con-142 Con-180
Con-180
Con-180
C. bolteae
C. bolteae
Bacteroides
Bacteroides
sp. 20_3
sp. 20_3
C. symbiosum
C. symbiosum
T2D-14
T2D-14
Clostridium
Clostridium
sp. HGF2
sp. HGF2
T2D-8
T2D-8
T2D-2
T2D-2
C. hathewayi
C. hathewayi
T2D-16
T2D-16
E. lenta
E. lenta
T2D-62
T2D-62
Clostridium ramosum
Clostridium ramosum
T2D-12
T2D-12
T2D-170
T2D-170
T2D-9
T2D-9
T2D-93
T2D-93
T2D-90
T2D-90
T2D-37
T2D-37
T2D-6
T2D-6
B. intestinalis
B. intestinalis
T2D-165
T2D-165
T2D-79
T2D-79
T2D-73
T2D-73
T2D-30
T2D-30
C. symbiosum
T2D-14
Clostridium sp. HGF2
T2D-8
T2D-2
C. hathewayi
E. lenta
T2D-62
Clostridium ramosum
T2D-12 T2D-9
T2D-79
T2D-73
C. bolteae
Bacteroides sp. 20_3
T2D-16
T2D-170
T2D-93
T2D-90
T2D-37
T2D-6
B. intestinalis
T2D-165
Con-130
Con-130
Con-109
Con-109
F. prausnitzii
F. prausnitzii
Con-131
Con-131
Con-152
Con-152
Con-144
Con-144
Con-133
Con-133
E. rectale
E. rectale
Clostridiales
Clostridiales
sp. SS3/4
sp. SS3/4
Con-101
Con-101
Con-104
Con-104
H. parainfluenzae
H. parainfluenzae
Con-148
Con-148
Con-155
Con-155
Con-120
Con-120
Con-122
Con-122
R. intestinalis
R. intestinalis
R. inulinivorans
R. inulinivorans
T2D-30
Con-130
Con-109
F. prausnitzii
Con-131
Con-152
Con-144
Con-133
E. rectale
Clostridiales sp. SS3/4
Con-101
Con-104
H. parainfluenzae
Con-148
Con-155
Con-120
Con-122 R. intestinalis
R. inulinivorans
Clostridiales
Clostridium
Faecalibacterium
Eubacterium
Roseburia
Subdoligranulum
Lachnospiraceae
Erysipelotrichaceae
Firmicutes
Desulfovibrio
Escherichia
Haemophilus
Proteobacteria
Bacteroides
Alistipes
Bacteroidales
Parabacteroides
T2D-enriched MLGsControl-enriched MLGs
Verrucomicrobia
Akkermansia
Actinobacteria
Eggerthella
Unclassied
b
Butyrate-producing bacteria
Con-343 Con-3380 Con-1831 Con-1697
Butyrate biosynthesis
Akkermansia muciniphila
T2D-317 Mucin degradation
Sulphate-reducing bacteria
T2D-823 H2S biosynthesis
Oxidative stress resistance Drug resistance
Cell motility
Xenobiotics biodegradation and metabolism
CH4 metabolism
Mucin layer integrality
T2D
Gut microbiota Gut environment
Sugar related membrane transport
Metabolism of cofactors and vitamins
BCAA transport
Butyrate
Cofactors
Vitamins
Host tissues
Xenobiotics
Oxidative stress
Mucin layer
BCAA
H2S
CH4
Figure 2
|
Taxonomic and functional characterization of gut microbiota in
T2D. a, A co-occurrence network was deduced from 47 MLGs that were
identified from 52,484 gene markers. Nodes depict MLGs with their ID
displayed in the centre. The size of the nodes indicates gene numberwithin the
MLG. The colour of the nodes indicates their taxonomic assignment.
Connecting lines represent Spearman correlation coefficient values above 0.4
(blue) or below 20.4 (red). b, A schematic diagramshowing the main functions
of the gut microbes that had a predicted T2D association. Red text denotes
enriched functions in T2D patients; blue text denotes depleted functions in
T2D patients; black text denotes an uncertain functional role relative to T2D.
The dashed line arrows point to the inference that was not detected directly but
reported by previous studies.
RESEARCH ARTICLE
58 | NATURE | VOL 490 | 4 OCTOBER 2012
Macmillan Publishers Limited. All rights reserved
©2012

To characterize ecologically the gut bacteria involved in the T2D-
related dysbiosis, we compared, in all individual samples, the distri-
bution of the occurrence rate of both T2D-associated gene and func-
tion markers, and these showed the same pattern, which was that the
control-enriched markers had a higher occurrence rate on average
than the T2D-enriched markers (Fig. 3b and Supplementary Figs 13–
15). This may be because the beneficial bacteria lost in the T2D gut
were universally present, whereas some of the harmful bacteria that
appeared in the T2D gut were diverse, and thus had less overall
abundance within the human population.
Gut-microbiota-based T2D classification
To exploit the potential ability of T2D classification by gut microbiota,
we developed a T2D classifier system based on the 50 gene markers that
we defined as an optimal gene set by a minimum redundancy–maximum
relevance (mRMR) feature selection method (Supplementary Fig. 16
and Supplementary Table 13). For intuitive evaluation of the risk of
T2D disease based on these 50 gut microbial genemarkers, we computed
a T2D index (Supplementary Methods), which correlated well with the
ratio of T2D patients in our population (Fig. 4a), and the area under the
receiver operating characteristic (ROC) curve was 0.81 (95% confidence
interval 0.76–0.85) (Fig. 4b), indicating the gut-microbiota-based T2D
index could be used to classify T2D individuals accurately.
We validated the discriminatory power of our T2D classifier using
an independent study group: 11 T2D patients and 12 non-diabetic
controls. In this assessment analysis, the top eight samples with the
highest T2D index were all T2D patients (Fig. 4c and Supplementary
Table 14); the average T2D index between case and control was sig-
nificantly different (P50.004, Student’s t-test). Overall, our cross-
sectional study in overt T2D indicated that it would be worthwhile to
test more extensively gut-microbiota-based classifiers in future lon-
gitudinal studies for their ability to identify subsets of the population
that are at high risk for progressing to clinically defined T2D.
Discussion
T2D is a heterogeneous and multifactorial disease, influenced by a
number of different genetic and environmental factors. By applying
the standard two-stage GWAS strategy to design and carry out a
MGWAS to identify disease-associated metagenomic markers, the
present study highlights how the gut microbial composition,
traditionally considered to be factors of environmental origin
12
, dif-
fers between T2D patients and non-diabetic control subjects in a
Chinese population.
We first established an updated human microbial gene reference set,
adding information from both a new ethnicity and from T2D patients,
which will be a useful resource for future metagenomic analyses. We
also developed the concept of a MLG, which provided various types of
taxonomic information from whole-genome shotgun data, including
bacterial species-specific regions on a chromosome, and mobile genetic
elements, such as plasmids and bacteriophages. Thus, a MLG can
provide metagenomic species-level information even for unknown
species, instead of requiring traditional taxonomic classification
approaches based on sequence composition or similarity
28,29
. The use
of species-level information allows assessment of the relationships
between the T2D-associated bacteria. For example, we identified what
appears to be an antagonistic relationship between beneficial bacteria
and harmful bacteria, highlighted by the large populations of clostridial
clusters. These species-level analyses also showed various patterns: for
example, the MLG from Haemophilus parainfluenzae in the control
samples could be inferred, under these circumstances, to be beneficial;
however, on the basis of relationshippatterns, it was quite distinct from
the other inferred beneficial bacteria, indicating that H. parainfluenzae
may have a different type of impact in this specific biological context
(Fig. 2a).
Our findings indicated that T2D patients had only a moderate
degree gut bacterial dysbiosis; however, functional annotation ana-
lyses indicated a decline in butyrate-producing bacteria, which may be
metabolically beneficial, and an increase in several opportunistic
a b
Occurrence rate (n = 344)
0.0
0.2
0.4
0.6
0.8
1.0
0110
T2D-enriched
markers
Control-enriched
markers
Density
Abundance sum
Within-sample diversity
0.0
0.1
0.2
11.0
11.4
11.8
ControlsIBD patientsControls
T2D patients
***
NS
NS
**
Figure 3
|
Gut microbiota of T2D patients show a moderate degree of
dysbiosis. a, An ecological comparison between T2D patients (n5170) and
control (n5174) in all samples, as well as inflammatory bowel disease (IBD)
patients (n525) and control (n599) from published MetaHIT samples
8
.The
upward bars denote the gross relative abundance of the T2D-associated gene
markers for each sample and the same value computed on the inflammatory-
bowel-disease-associated gene markers (see Supplementary Methods). The
downward bars denote the within-sample diversity (calculated using the
Shannon index) in each group. For an individual sample, a lower proportion of
gut microbiota was implicated in T2D disease and there was no significant
difference in the within-sample diversity between the T2D patients and control
as compared with thedistinct difference seen in the inflammatory bowel disease
analysis. **P,0.01; ***P,0.001 (Student’s t-test); NS, not significant; and
the error bar denote standard error. b, A density histogram showing a
comparison of the occurrence rate distribution between T2D-enriched gene
markers and control-enriched gene markers in all samples (n5344). The
threshold of mapped read number for gene identification is $2.
T2D index
Number of individuals
≤–1.5 –1 0 1 2 3 ≥3.5
0
20
40
60
0
0.2
0.4
0.6
0.8
1
Percentage of T2D patients
ab
–20246
•
•
•
•
•
•
•
•
•
•
•
•
•
•
••
•
•
•
•
•••
Controls T2D patients
T2D index
c
1.00.80.60.40.20.0
0.0
0.2
0.4
0.6
0.8
1.0
1 – specicity
Sensitivity
AUC = 0.81
95% CI: 0.76–0.85
Figure 4
|
A trial classification of T2D using gut microbial gene markers.
a, A classifier to identify T2D individuals was constructed using 50 gene
markers selected by mRMR, and then, for each individual, a T2D index was
calculated to evaluate the risk of T2D. The histogram shows the distribution of
T2D indices for all individuals,in which values less than 21.5 and values greater
than 3.5 were grouped. For each bin, the black dotsshow the proportion of T2D
patients in the population of that bin (yaxis on the right). b, The area under the
ROC curve (AUC) of gut-microbiota-based T2D classification. The black bars
denote the 95% confidence interval (CI) and the area between the two outside
curves represents the 95% CI shape. c, The T2D index was computed for an
additional 11 Chinese T2D samples and 12 non-diabetic controls. The box
depicts the interquartile range (IQR) between the first and third quartiles (25th
and 75th percentiles, respectively) and the line inside denotes the median,
whereas the points represent the T2D index in each sample.
ARTICLE RESEARCH
4 OCTOBER 2012 | VOL 490 | NATURE | 59
Macmillan Publishers Limited. All rights reserved
©2012

pathogens. Importantly, the abundance of these categories of
opportunistic pathogens seemed to be quite diverse among our
Chinese study participants. Such changes in the intestinal bacteria
composition have recently been reported for colorectal cancer
patients
30
and ageing population
31
. Thus, a general picture is emerging
where butyrate-producing bacteria seem to have a protective role
against several types of diseases. Additionally, our finding of a general
dysbiosis in T2D patients raises the possibility that there is a ‘func-
tional dysbiosis’, rather than there being a specific microbial species
that has a direct association with T2D pathophysiology. Furthermore,
given that other intestinal diseases show a loss of butyrate-producing
bacteria with a commensurate increase in opportunistic pathogens, it
is possible that dysbiosis that results in a disordered, rather than
directional, alteration of gut microbial composition may itself have
a role in increasing the susceptibility to a variety of diseases.
Our analysis of bacterial gene functions indicating there was an
increase in functions relating to gut oxidative stress response is also
of interest, given that previous studies have shown that a high
oxidative stress level is related to a predisposition for diabetic com-
plications
32
. Finally, our findings that gut metagenomic markers are
able to differentiate between T2D cases and controls with a higher
level of specificity than similar analyses based on human genome
variation
33
raises the possibility for a mode of monitoring gut health
and a complementary approach for risk assessment of this common
disorder.
METHODS SUMMARY
Sample collection and DNA extraction. Faecal samples were obtained from 368
volunteers (345 samples for MGWAS and 23 additional samples for T2D clas-
sification) after signing an informed consent form. The sampling procedure was
approved by the Ethical Committee for Clinical Research from the Peking
University Shenzhen Hospital, Shenzhen Second People’s Hospital and
Medical Research Center of Guangdong General Hospital. The individuals had
not received any antibiotic treatment within 2 months before sample collection.
The samples were frozen immediately and underwent DNA extraction using
standard methods
34
.
Sequencing and data processing. Illumina GAIIx and HiSeq 2000 were used to
sequence the samples. We constructed a paired-end library with insert size of
,350 base pairs for every sample. Adaptor contamination and low-quality reads
were discarded from the raw reads, and the remaining reads were filtered to
eliminate human host DNA based on the human genome reference (hg18).
Full Methods and associated references are available in the Supplementary
Information.
Received 30 August 2011; accepted 27 July 2012.
Published online 26 September 2012.
1. Wellen, K. E. & Hotamisligil, G. S. Inflammation, stress, and diabetes. J. Clin. Invest.
115, 1111–1119 (2005).
2. Rise
´rus, U., Willett, W. C. & Hu, F. B. Dietary fats and prevention of type 2 diabetes.
Prog. Lipid Res. 48, 44–51 (2009).
3. The Wellcome Trust Case Control Consortium.Genome-wide association study of
14,000 cases of seven common diseases and 3,000 shared controls. Nature 447,
661–678 (2007).
4. Scott, L. J. et al. A genome-wide association study of type 2 diabetes in Finns
detects multiple susceptibility variants. Science 316, 1341–1345 (2007).
5. Musso, G., Gambino, R. & Cassader, M. Interactions between gut microbiota and
host metabolism predisposing to obesity and diabetes. Annu. Rev. Med. 62,
361–380 (2011).
6. Eckburg, P. B. et al. Diversity of the human intestinalmicrobial flora. Science 308,
1635–1638 (2005).
7. Turnbaugh, P. J. et al.A core gut microbiome in obese and lean twins. Nature 457,
480–484 (2009).
8. Qin, J. et al. A human gut microbial gene catalogue established by metagenomic
sequencing. Nature 464, 59–65 (2010).
9. The Human Microbiome Project Consortium. Structure, function and diversity of
the healthy human microbiome. Nature 486, 207–214 (2012).
10. The Human MicrobiomeProject Consortium.A framework for humanmicrobiome
research. Nature 486, 215–221 (2012).
11. Vijay-Kumar, M. et al. Metabolic syndrome and altered gut microbiota in mice
lacking Toll-like receptor 5. Science 328, 228–231 (2010).
12. Ba
¨ckhed, F. et al. The gut microbiota as an environmental factor that regulates fat
storage. Proc. Natl Acad. Sci. USA 101, 15718–15723 (2004).
13. Ley, R. E. et al. Obesity alters gut microbial ecology. Proc. Natl Acad. Sci. USA 102,
11070–11075 (2005).
14. Zhang, H. et al. Human gut microbiota in obesityand after gastric bypass.Proc. Natl
Acad. Sci. USA 106, 2365–2370 (2009).
15. Ba
¨ckhed, F., Manchester, J. K., Semenkovich, C. F. & Gordon, J. I. Mechanisms
underlying the resistance to diet-induced obesity in germ-free mice. Proc. Natl
Acad. Sci. USA 104, 979–984 (2007).
16. Turnbaugh, P. J. et al. An obesity-associated gut microbiome with increased
capacity for energy harvest. Nature 444, 1027–1031 (2006).
17. Manichanh, C. et al. Reduced diversity of faecal microbiota in Crohn’s disease
revealed by a metagenomic approach. Gut 55, 205–211 (2006).
18. Joossens, M. et al. Dysbiosis of the faecal microbiota in patients with Crohn’s
disease and their unaffected relatives. Gut 60, 631–637 (2011).
19. Larsen, N. et al. Gut microbiota in human adults with type 2 diabetes differs from
non-diabetic adults. PLoS ONE 5, e9085 (2010).
20. Arumugam, M. et al. Enterotypes of the human gut microbiome. Nature 473,
174–180 (2011).
21. Price, A. L. et al. Principal components analysis corrects for stratification in
genome-wide association studies. Nature Genet. 38, 904–909 (2006).
22. Woo, P. C. Y. et al. Bacteremia due to Clostridium hathewayi in a patient with acute
appendicitis. J. Clin. Microbiol. 42, 5947–5949 (2004).
23. Elsayed, S. & Zhang, K. Bacteremia caused by Clostridium symbiosum.J. Clin.
Microbiol. 42, 4390–4392 (2004).
24. McClean, K. L., Sheehan, G. J. & Harding, G. K. Intraabdominal infection: a review.
Clin. Inf. Dis. 19, 100–116 (1994).
25. Brook, I. Clostridial infection in children. J. Med. Microbiol. 42, 78–82 (1995).
26. Greenblum, S., Turnbaugh, P. J. & Borenstein, E. Metagenomic systems biology of
the human gut microbiome reveals topological shifts associated with obesity and
inflammatory bowel disease. Proc. Natl Acad. Sci. USA 109, 594–599 (2012).
27. McArdle, B. H. & Anderson, M. J. Fittingmultivariate models to communitydata: a
comment on distance-based redundancy analysis. Ecology 82, 290–297 (2001).
28. Yang, B. et al. Unsupervised binning of environmental genomic fragments based
on an error robust selectionof l-mers. BMC Bioinformatics11 (suppl. 2), S5 (2010).
29. Krause, L. et al. Phylogenetic classificationof short environmental DNA fragments.
Nucleic Acids Res. 36, 2230–2239 (2008).
30. Wang, T. et al. Structural segregation of gut microbiota between colorectal cancer
patients and healthy volunteers. ISME J. 6, 320–329 (2012).
31. Biagi, E. et al. Through ageing, and beyond: gut microbiota and inflammatory
status in seniors and centenarians. PLoS ONE 5, e10667 (2010).
32. Kashyap, P. & Farrugia, G. Oxidative stress: key player in gastrointestinal
complications of diabetes. Neurogastroenterol. Motil. 23, 111–114 (2011).
33. Lyssenko, V. et al. Clinicalrisk factors, DNA variants,and the development oftype 2
diabetes. N. Engl. J. Med. 359, 2220–2232 (2008).
34. Godon, J. J., Zumstein,E., Dabert, P., Habouzit, F. & Moletta, R. Molecular microbial
diversityof an anaerobic digestor as determined by small-subunit rDNAsequence
analysis. Appl. Environ. Microbiol. 63, 2802–2813 (1997).
35. Li, S. et. al. Type 2 diabetes gut metagenome(microbiome) data from 368 Chinese
samples. GigaScience http://dx.doi.org/10.5524/100036 (2012).
36. Wu, G. D. et al. Linking long-term dietary patterns with gut microbial enterotypes.
Science 334, 105–108 (2011).
Supplementary Information is available in the online version of the paper.
Acknowledgements We thank L. Goodman for editing the manuscript and providing
comments. This research was supported by the Ministry of Science and Technology of
China, 863 program (2012AA02A201), the National Natural Science Foundation of
China (30890032, 30725008, 30811130531, 31161130357), the Shenzhen
Municipal Government of China (ZYC200903240080A, BGI20100001,
CXB201108250096A, CXB201108250098A), the Danish Strategic Research Council
grant (2106-07-0021), the Ole Rømer grant from Danish Natural Science Research
Council, the Solexa project (272-07-0196), and the European Commission FP7 grant
HEALTH-F4-2007-201052. The Lundbeck Foundation Centre for Applied Medical
Genomics in Personalised Disease Prediction, Prevention and Care (LuCamp,
www.lucamp.org). The Novo Nordisk Foundation Center for Basic Metabolic Research
is an independentResearch Center at theUniversity of Copenhagenpartially funded by
an unrestricted donation from the Novo Nordisk Foundation (http://
www.metabol.ku.dk). We are also indebted to many additional faculty and staff of
BGI-Shenzhen who contributed to this work.
Author Contributions The project idea was conceivedand the project was designedby
Ju.W., K.K.,O.P., R.N. and S.D.E.; J.Q.,Y.L., Sh.L. and Ju.W. managedthe project. F.Z., Z.C.,
R.X., Su.L., L.H.,D.L., P.W., Y.D., X.S., Z.L., A.T., S.Z., M.W., Q.F. and T.H. performed sample
collection and clinical study. Wen.Z., M.G., J.Y., Y.Z. and W.X. performed DNA
experiments.Ju.W.,K.K., O.P., R.N., S.D.E., J.Q., Y.L., Sh.L. and J.Z. designed the analysis.
J.Q., Y.L., Sh.L., J.Z., Su.L., Y.G., Y.P., D.S., X.L., W.C., D.Z., Y.Q., M.Z., Z.Z., Z.J., G.S., J.L., J.R.,
S.O., H.C. and W.W.performed the data analysis. J.Q.,Sh.L., J.Z., Y.G., Y.P., M.A.,E.L., P.R.,
N.P. and J.-M.B. worked on metagenomic linkage group method. J.Q., D.S., Su.L., Y.Q.,
J.R., G.F. and S.O. did the functional annotation analyses. J.Q., Sh.L., D.S., J.Z., Y.P. and
Y.L. wrote thepaper. Ju.W., O.P., K.K., R.N.,S.D.E., Ji.W., H.Y., So.L.,Wei.Z. and R.Y. revised
the paper.
Author Information The rawIllumina read data of all 368 samples hasbeen deposited
in the NCBI Sequence Read Archive under accession numbers SRA045646 and
SRA050230. The assembly data, updated metagenome gene catalogue, annotation
information,and MGLs are published in the GigaScience database, GigaDB
35
. Reprints
and permissions information is available at www.nature.com/reprints. The authors
declare no competing financial interests. Readers are welcome to comment on the
online version of the paper. Correspondence and requests for materials should be
addressed to Ju.W. (wangj@genomics.org.cn).
RESEARCH ARTICLE
60 | NATURE | VOL 490 | 4 OCTOBER 2012
Macmillan Publishers Limited. All rights reserved
©2012