ArticlePDF Available

Alterations of the human gut microbiome in liver cirrhosis

July 2014
Nature

DOI:10.1038/nature13568

Authors:

Nan Qin

Zhejiang University

Fengling Yang

Li Ang

First affiliated hospital of zhengzhou university

Edi Prifti

Institute of Research for Development

Show all 27 authorsHide

Liver cirrhosis occurs as a consequence of many chronic liver diseases that are prevalent worldwide. Here we characterize the gut microbiome in liver cirrhosis by comparing 98 patients and 83 healthy control individuals. We build a reference gene set for the cohort containing 2.69 million genes, 36.1% of which are novel. Quantitative metagenomics reveals 75,245 genes that differ in abundance between the patients and healthy individuals (false discovery rate < 0.0001) and can be grouped into 66 clusters representing cognate bacterial species; 28 are enriched in patients and 38 in control individuals. Most (54%) of the patient-enriched, taxonomically assigned species are of buccal origin, suggesting an invasion of the gut from the mouth in liver cirrhosis. Biomarkers specific to liver cirrhosis at gene and function levels are revealed by a comparison with those for type 2 diabetes and inflammatory bowel disease. On the basis of only 15 biomarkers, a highly accurate patient discrimination index is created and validated on an independent cohort. Thus microbiota-targeted biomarkers may be a powerful tool for diagnosis of different diseases.

Content uploaded by Li Shao

Content may be subject to copyright.

ARTICLE doi:10.1038/nature13568

Alterations of the human gut microbiome

in liver cirrhosis

Nan Qin

1,2

*, Fengling Yang

*, Ang Li

*, Edi Prifti

*, Yanfei Chen

*,LiShao

1,2

*, Jing Guo

, Emmanuelle Le Chatelier

, Jian Yao

1,2

Lingjiao Wu

, Jiawei Zhou

, Shujun Ni

, Lin Liu

, Nicolas Pons

, Jean Michel Batto

, Sean P. Kennedy

, Pierre Leonard

Chunhui Yuan

, Wenchao Ding

, Yuanting Chen

, Xinjun Hu

, Beiwen Zheng

1,2

, Guirong Qian

,WeiXu

, S. Dusko Ehrlich

3,4

Shusen Zheng

2,5

& Lanjuan Li

1,2

Liver cirrhosis occurs as a consequence of many chronic liver diseases that are prevalent worldwide. Here we character-

ize the gut microbiome in liver cirrhosis by comparing 98 patients and 83 healthy controlindividuals. We build a reference

gene set for the cohort containing 2.69 million genes, 36.1%of which are novel. Quantitative metagenomics reveals 75,245

genes that differ in abundance between the patients and healthy individuals (false discovery rate ,0.0001) and can be

grouped into 66 clusters representing cognate bacterial species; 28 are enriched in patients and 38 in control individuals.

Most (54%) of the patient-enriched, taxonomically assigned species are of buccal origin, suggesting an invasion of the gut

from the mouth in liver cirrhosis. Biomarkers specific to liver cirrhosis at gene and function levels are revealed by a

comparison with those for type 2 diabetes and inflammatory bowel disease. On the basis of only 15 biomarkers, a highly

accurate patient discrimination index is created and validated on an independent cohort. Thus microbiota-targeted bio-

markers may be a powerful tool for diagnosis of different diseases.

Cirrhosis is an advanced liver disease resulting from acute or chronic

liver injury, including alcohol abuse, obesity and hepatitis virus infec-

tion. The prognosis for patients with decompensated liver cirrhosis is

poor, and they frequently requireliver transplantation

. The liver inter-

acts directly with the gut through the hepatic portal and bile secretion

systems. Enteric dysbiosis, especially the translocation of bacteria

and

their products

4,5

across the gut epithelial barrier, is involved in the pro-

gression of liver cirrhosis. However, the phylogenetic and functional com-

position changes in the human gut microbiota that are related to this

progression remain obscure

. Some studies have revealed that altera-

tions in the gut microbiota are important in complications of end-stage

liver cirrhosis

(such as spontaneous bacterial peritonitis

and hepatic

encephalopathy

) and the induction and promotion of liver damage

in early-stage liver disease

(such as alcoholic liver disease

and non-

alcoholic fatty liver disease

), but definitive associations of gut micro-

biota andliver pathology in humans are still lacking

. Studiesof patients

with liver cirrhosis

and of mouse models for alcoholic liver disease

have revealed a similar and substantial alteration in the gut microbiota,

as measured by sequencing of 16S ribosomal RNA genes. How these

phylogenetic alterations relate to changes in the functioningof this eco-

system is, however, unclear.

The role of gut microbiota in human health and disease

has recently

received considerable attention. Chronic diseases, such as obesity

15–18

inflammatory bowel disease (IBD)

19,20

, diabetes mellitus

21,22

, metabolic

syndrome

, symptomatic atherosclerosis

and non-alcoholic fatty liver

disease

, have been associated with gut microbiota. The US National

Institutes of Health Human Microbiome Project (HMP) generated a

large data set from different anatomical sites among 242 healthy indivi-

duals and created a large human microbiome gene resource

25,26

. Quanti-

tativemetagenomicsanalysis

27,28

developed by the MetaHIT consortium

revealed a significant loss of gut microbial richness associated with the

risk of metabolic syndrome related co-morbidities. Here we apply a

similar analysis to contrast microbiota from 123 patients with liver cir-

rhosis and 114 healthy counterparts of Han Chinese origin.

Gene catalogue of gut microbes

We constructed a gene catalogue from 98 Chinese patients with liver

cirrhosis and 83 healthy Chinese control individuals (Supplementary

Table 1) using the methodology developed by MetaHIT. The liver cirrho-

sis catalogue contained 2,688,468 non-redundant open reading frames

(ORFs). We compared it with three other gut microbial catalogues:

MetaHIT

,HMP

and T2D

. To facilitate this comparison, genes were

predicted from the original contigs using the same criteria. The MetaHIT

catalogue contained 3,452,726 genes, HMP 4,768,112 genes and T2D

2,148,029 genes. In total 674,131 genes were common to all catalogues

(Extended Data Fig. 1a). The liver cirrhosis catalogue, MetaHIT, HMP

and T2D gene sets contained 794,647, 1,419,517, 2,620,096 and 623,570

unique genes, respectively. Genes from the liver cirrhosis, T2D and

MetaHIT catalogues were merged; the HMP was not included, as it

contained Sanger, 454 or Illumina-based 16S sequences, in addition to

whole metagenomic data. The merged non-redundant catalogue con-

tained 5,382,817 genes (Extended Data Fig. 1b).

Phylogenetic profiles of gut microbes

The sequencing reads (36.67%) were aligned against 4,398 reference

genomes from the National Center for BiotechnologyInformation and

the HMP (Supplementary Table 2). After correction for population strat-

ification that could be related to non-liver cirrhosis-related factors (see

Methods), the relative abundances of phylum, class,order, family, genus

and species between liver cirrhosis and control groups were compared

(Extended Data Fig. 2). Phylotypes with a median relative abundance

larger than 0.01% of the total abundance in either the healthy control

*These authors contributed equally to this work.

State Key Laboratory for Diagnosis and Treatment of Infectious Disease, The First Affiliated Hospital, College of Medicine, Zhejiang University, 310003 Hangzhou, China.

Collaborative Innovation Center

for Diagnosis and Treatment of Infectious Diseases, Zhejiang University, 310003 Hangzhou, China.

Metagenopolis, Institut National de la Recherche Agronomique, 78350 Jouy en Josas, France.

King’s

College London, Centre for Host-Microbiome Interactions, Dental Institute Central Office, Guy’s Hospital, London Bridge, London SE1 9RT, UK.

Key Laboratory of Combined Multi-organ Transplantation,

Ministry of Public Health, the First Affiliated Hospital, Zhejiang University, 310003 Hangzhou, China.

00 MONTH 2014 | VOL 000 | NATURE | 1

group or the liver cirrhosis group were included for comparison. At the

phylum level, Bacteroidetes and Firmicutes dominated the faecal micro-

bial communities of both groups (Fig. 1a, b). Compared with healthy

controls, patients with liver cirrhosis had fewer Bacteroidetes (Fig. 1a),

but higher levels of Proteobacteria and Fusobacteria (Fig. 1b).

At the genus level, Bacteroides was the dominant phylotype in both

groups, but was significantly decreased in the liver cirrhosis group. Of

the remaining genera, Veillonella,Streptococcus,Clostridium and Prevotella

were enriched in the liver cirrhosis group, while Eubacterium and Alistipes

were dominant in the healthy controls (Fig. 1a, b). The most abundant

species in both liver cirrhosis and the healthy control groups were pri-

marily from the Bacteroides genus. Of the 20 species that increased the

most in abundance in the liver cirrhosis group, four were Streptococcus

spp. and six were Veillonella spp., suggesting that the two genera might

play an important role in liver cirrhosis. Of the species that decreased

the most inabundance in theliver cirrhosis group, 12 wereBacteroidetes

and seven were Firmicutes, specifically from the order Clostridiales.

Gut microbial species associated with cirrhosis

Our investigation included two phases. The first was discovery, where

we compared 98 patients with liver cirrhosis and 83 healthy controls.

The second was validation, with additional 25 patients and 31 controls.

In the discovery phase, a Wilcoxon rank-sum test corrected for mul-

tiple testing by the Benjamini and Hochberg method was used to iden-

tify differentially abundant genesin patients and controls. At a stringent

threshold (false discovery rate (FDR) ,0.0001), 75,245 genes were found:

49,830 were more abundant in the patients and 25,415 in the controls

(Methods). Patients and controls could be clearly separated by princi-

pal component analysis based on the 75,245 genes; this was confirmed

with the validation samples (Supplementary Table 3 and Extended Data

Fig. 1c).

To explore further the microbial genes associated with livercirrhosis

we grouped them into clusters, denoted metagenomic species (MGS)

here, on the basis of their abundance profiles

27,30

. Of the 66 MGS, 38 and

28 were enriched in healthy individuals and patients, respectively. The

significantly different abundance distribution between healthy and liver

cirrhosis subjects is shown in Fig. 2 and Supplementary Table 4. A majority

(82%) were also differentially abundant in the validation cohort (q,0.05),

in spite of the reduced statistical power due to the smaller cohort size.

Composition of bacterialcommunities varies considerably as a func-

tion of the overall gene richness

27,28

and the loss of richness is associated

with obesity and IBD

27,28,31

. A large majority of the 38 MGS enriched in

the healthy individuals (33, 86.8%) was correlated with the richness at

q,10

in the Chinese cohort; 26 of these (78.8%) were similarly cor-

related in a Danish cohort (Extended Data Fig. 3). These observations

indicate that gut communities of bacteria in healthy individuals across

continents may be largely similar. Furthermore, gene richness was much

lower in patients with liver cirrhosis than in healthy individuals (on

average389,000 and 497,000 genes,respectively;Supplementary Table5

and Extended Data Fig. 4, top left). Interestingly, among the species

enriched in healthy Chinese, were Faecalibacterium prausnitzii, which

has anti-inflammatory properties and was foundin a ‘healthy’ gene-rich

microbiome

27,28

,andCoprococcus comes, which might contribute to gut

health through butyrate production. A similar butyrate production role

may be played by three Lachnospiraceae and five Ruminococcaceae

enriched in healthy individuals. A lower abundance of these species in

patients with liver cirrhosis indicates that these individuals have a less

healthy gut microbiome.

Most interestingly, a high proportion of MGS enriched in patients

belong to taxa such as Veillonella (n58) or Streptococcus (n56), known

to include species of oral origin (Supplementary Table 4). However, the

small intestine also harbours such species

and small-intestinal bacterial

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

Veillonella

Streptococcus

Prevotella

Haemophilus

Lactobacillus

NULL(Lachnospiraceae bacterium 2 1 58FAA)

Fusobacterium

Megasphaera

Genus

0.00

0.10

0.20

Proteobacteria

Fusobacteria

0.00

0.01

0.02

0.03

0.04

0.05

0.06

Abundance

Streptococcus salivarius

Veillonella parvula

Veillonella atypica

Ruminococcus gnavus

Haemophilus parainﬂuenzae

Veillonella sp. 6 1 27

Veillonella sp. 3 1 44

Lachnospiraceae bacterium 2 1 58FAA

Veillonella dispar

Streptococcus parasanguinis

Veillonella sp. oral taxon 158

Lactobacillus salivarius

Streptococcus vestibularis

Streptococcus anginosus

Species

0.0

0.2

0.4

0.6

0.8

1.0

Bacteroidetes

Phylum

Healthy

Liver cirrhosis

0.0

0.2

0.4

0.6

0.8

Abundance

Bacteroides

Eubacterium

Alistipes

Faecalibacterium

Roseburia

Parabacteroides

Odoribacter

Ruminococcus

Dorea

Bilophila

Coprococcus

Null(Lachnospiraceae bacterium 6 1 63FAA)

Tannerella

Subdoligranulum

Null(Lachnospiraceae bacterium 5 1 63FAA)

Null([Bacteroides] pectinophilus ATCC 43243)

Holdemania

Null(Lachnospiraceae bacterium 3 1 46FAA)

Phascolarctobacterium

Null(Ruminococcaceae bacterium D16)

Genus

0.00

0.02

0.04

0.00

0.05

0.10

0.15

Faecalibacterium prausnitzii

Bacteroides sp. D20

Alistipes putredinis

Bacteroides sp. 4 1 36

Bacteroides uniformis

Eubacterium eligens

Eubacterium rectale

Parabacteroides merdae

Bacteroides ﬁnegoldii

Odoribacter splanchnicus

Roseburia intestinalis

Bacteroides eggerthii

Ruminococcus sp. 5 1 39BFAA

Bacteroides sp. 1 1 30

Roseburia hominis

Bacteroides sp. 9 1 42FAA

Bilophila wadsworthia

Eubacterium hallii

Alistipes ﬁnegoldii

Parabacteroides distasonis

Species

Abundance

Phylum

Abundance

Healthy

Liver cirrhosis

Healthy

Liver cirrhosis

Healthy

Liver cirrhosis

Healthy

Liver cirrhosis

Figure 1

Differentially abundant phyla in patients (

598) and healthy

individuals (

583). The phylotypes decreased (a) and increased (b)in

patients with liver cirrhosis at the phylum, genus and species levels. Blue and

red represent healthy controls and patients with liver cirrhosis, respectively.

Only the 20 most abundant species in each group are shown for clarity. The

phylotypes with median relative abundances greater than 0.01% of total

abundance in either the healthy control group or the liver cirrhosis group are

included (FDR ,0.01, Wilcoxonrank-sum test corrected by the Benjamini and

Hochberg method).The boxes representthe interquartile range (IQR), from the

first and third quartiles, and the inside line represents the median. The whiskers

denote the lowest and highest values within 1.5 IQR from the first and third

quartiles. The circles represent outliers beyond the whiskers. The notches show

the 95% confidence interval for the medians. If the notches of two boxes do not

overlap, it gives evidence of a significant difference between the medians.

RESEARCH ARTICLE

2 | NATURE | VOL 000 | 00 MONTH 2014

overgrowth is frequently found in patients with liver cirrhosis

.To

explore the origin of the patient-enriched species, we used information

from the HOMD

and GOLD

databasesabout the origin of the closely

related sequenced isolates. We also constructed a catalogue of 114 pub-

licly available genomes for Streptococcus,Fusobacterium,Lactobacillus,

Veillonella and Megasphaera strains, originating mostly from mouth

or gut (57 or 28, respectively; Supplementary Table 6) and used it for

blastN and blastP analysis (Methods). Thirteen of the species were closest

to an oral isolate whereas only six were closest to the gut isolates, a single

species being from the ileum (Supplementary Table 4 and Extended Data

Fig. 4, top right). Comparison with the three ileum metagenomes failed

to reveal identity above that detectedby comparison with the sequenced

genomes (Methods). We conclude that oralcommensals invade the gut

in patients with liver cirrhosis. Possibly, an altered bile production in

cirrhosis renders the gut morepermissible and/or accessible to ‘foreign’

bacteria, as bile resistance may be required for survival in the human

gut

36,37

. As patient-enriched MGS include pathogens such as Campylo-

bacter and Haemophilus parainfluenzae, these also might use the oral

route to invade the gut, possibly via contaminated food. The invasion

species foreign to the niche may occur not only in the colon but also in

the ileum, and contribute to the small-intestinal bacterial overgrowth

associated with liver cirrhosis. Among the patient-enriched species were

Streptococcus anginosus,Veillonella atypica,Veillonella dispar,Veillonella

sp. oral taxon and Clostridium perfringens, which have been reported

to cause opportunistic infections

38–40

To analyse the relations between the liver-cirrhosis-associated MGS,

we generated networks based on co-abundance, for healthy individuals

and patients with liver cirrhosis (Fig. 2b). A striking featureis that tax-

onomically related species tend to cluster, as reported previously

.These

observations indicate thatthe gut environment becomes permissive for

Healthy

n = 33

Liver cirrhosis

n = 25

LPA HPA

100

200

300

500

600

1.0

1.2

1.8

2.0

2.2

MELD

CTP

INR

Crea

Alb

Healthy Liver cirrhosis

LPA HPA

Discovery cohort Validation cohort

0.5

LPA HPA

Healthy Liver

cirrhosis

–0.5

MELD : P < 10–5

LPA HPA LPA HPA LPA HPA LPA HPA

400

CTP : P < 2 × 10–4 TB : P < 2 × 10–4 PT : P < 0.02

INR : P < 0.06

1.6

1.4

L_44

Fusobacterium

L_32

S. oralis

L_12

S. anginosus

L_15

S. parasanguinis

L_24

Streptococcus

sp. 2_1_36FAA

L_14

S. vestibularis

L_42

Veillonella

L_18

C. concisus

L_20

A. segnis

L_10

M. micronuciformis

L_55

Veillonella

L_4

V. atypica

L_19

V. dispar

L_6

S. salivarius

L_2

B. dentium

L_7

V. parvula

L_59

Veillonella

L_17

Veillonella

sp._oral_taxon_158

L_39

Veillonella

L_8

P. buccae

L_9

H. parainﬂuenzae

L_1

L. mucosae

L_11

L. fermentum

L_3

L. salivarius

L_40

Lactobacillus

L_44

Fusobacterium L_32

S. oralis

L_12

S. anginosus L_15

S. parasanguinis

L_24

Streptococcus

sp. 2_1_36FAA

L_14

S. vestibularis

L_42

Veillonella

L_18

C. concisus

L_20

A. segnis

L_10

M. micronuciformis

L_55

Veillonella

L_4

V. atypica L_19

V. dispar

L_6

S. salivarius

L_2

B. dentium

L_7

V. parvula L_59

Veillonella

L_17

Veillonella

sp._oral_taxon_158

L_39

Veillonella

L_8

P. buccae

L_9

H. parainﬂuenzae

L_1

L. mucosae

L_11

L. fermentum

L_3

L. salivarius

L_40

Lactobacillus

H_33

Ruminococcaceae

H_37

Ruminococcaceae

H_24

Bacteroidales

H_16

Oscillibacter

H_9

Clostridiales

H_12

Clostridiales

H_3

Clostridiales

H_2

Clostridiales

H_11

Clostridiales

H_40

Clostridiales

H_29

Ruminococcaceae

H_8

Clostridium

H_10

Eubacterium

H_43

Eggerthella

H_18

Eubacterium

H_22

Eubacterium

H_15

Clostridiales

H_7

Clostridiales

H_5

Alistipes

H_28

Bacteroidales

H_21

Bacteroidales

H_26

Porphyromonadaceae

H_30

Ruminococcaceae

H_6

Subdoligranulum

H_32

C. comes

H_42

A. indistinctus

H_25

H_23

Lachnospiraceae

H_34

Lachnospiraceae

H_14

Lachnospiraceae

H_17

Ruminococcaceae

H_20

F. prausnitzii

H_36

Parabacteroides

H_33

Ruminococcaceae H_37

Ruminococcaceae

H_24

Bacteroidales

H_16

Oscillibacter

H_9

Clostridiales H_12

Clostridiales

H_3

Clostridiales

H_2

Clostridiales

H_11

Clostridiales H_40

Clostridiales

H_29

Ruminococcaceae

H_8

Clostridium

H_10

Eubacterium

H_43

Eggerthella

H_18

Eubacterium

H_22

Eubacterium

H_15

Clostridiales

H_7

Clostridiales

H_5

Alistipes H_28

Bacteroidales

H_21

Bacteroidales

H_26

Porphyromonadaceae

H_30

Ruminococcaceae

H_6

Subdoligranulum

H_32

C. comes

H_42

A. indistinctus

H_25

H_23

Lachnospiraceae

H_34

Lachnospiraceae

H_14

Lachnospiraceae

H_17

Ruminococcaceae

H_20

F. prausnitzii

H_36

Parabacteroides

Figure 2

Differentially abundant MGS in patients (

5123) and healthy

individuals (

5114). a, Abundance of 50 ‘tracer’genes for each species in the

discovery (n

patients

598, n

healthy

583) and validation cohorts (n

patients

525,

healthy

531); oral species are highlighted in red. Genes are in rows, abundance

is indicated by colour gradient (white, not detected; red, most abundant); the

enrichment significance is shown (qindicates the Mann–Whitney Pvalues

corrected by the Benjamini and Hochberg method). Individuals are shown in

columns, ordered by increasing abundance of patient-enriched species.

Correlation of the species abundance and patients’ clinical parameters in the

discovery cohort are indicated in colour code (red and blue for positive and

negative correlations; intensity reflects the level of correlation). MELD, model

for end-stage liver disease; CTP, Child–Turcotte–Pugh score; TB, total

bilirubin; PT, prothrombintime test; INR, international normalized ratio

describing coagulation of the blood in patients with liver cirrhosis; Crea,

creatinine level; Alb, albumin level. b, Top, clinical parameters of patients for

the lowest and highest patient-enriched species abundance (LPA and HPA,

respectively; n524 foreach). Pvalues indicate the significance of the difference

by Mann–Whitney U-test except MELD (Student’s t-test). Middle and bottom,

abundance-based species correlation network enriched in patients with liver

cirrhosis (n525) and healthy individuals (n533), respectively. Two nodes are

linked if the pooled variance z-test shows an FDR ,10

when accounting

for the compositionality effect (see Methods). The edge width is proportional to

the correlation strength. The node size is proportional to the mean abundance

in the respective population. Nodes with the same colour are classified in the

same phylogenetic order level.

ARTICLE RESEARCH

00 MONTH 2014 | VOL 000 | NATURE | 3

the development and maintenance of the related taxa in many indivi-

duals.Obviously, taxonomically unrelated speciescan also thrive in such

environments, as observed with Campylobacter concisus,H. parainflu-

enzae or Fusobacterium, whichtend to be associated with Veillonella in

patients. The overall abundanceof species enriched in patients reached

high levels,exceeding 5% in over a quarter andapproaching the extreme

of 40%, whereas it was very low in healthy individuals (Extended Data

Fig. 4, bottom). Interestingly, the severity of the disease was positively

correlatedwith the abundance of a number of MGS enrichedin patients

and negatively correlated with those of the MGS enriched in controls

(and therefore under-represented in patients; Fig. 2a). Thedisease status

of the patients with the highest load of these bacteria was significantly

worse than that of the patients with the lowest load (Fig. 2b, top). Sucha

‘dose response’ is consistent with an active role of the enriched species

in liver cirrhosis.

Microbial functions enriched in liver cirrhosis

To investigate the functional roleof the gut microbiota in livercirrhosis,

we identified4,801 KEGG (Kyoto Encyclopedia of Genes and Genomes

database) orthologues and 13,970 eggNOG (evolutionary genealogy of

genes: Non-supervised Orthologous Groups database) orthologues asso-

ciated with the disease (Supplementary Tables 7 and 8). The most abun-

dant KEGG orthologues in patients and controls were enzymefamilies.

The most enriched orthologues in patients were membrane transport,

similar to findings for IBDs

19,20

, obesity

and T2D

. In contrast, the

most prevalent markers among the controls included those involved in

carbohydrate metabolism,amino-acid metabolism, energy metabolism,

signal transduction andthe metabolismof cofactors and vitamins (Ex-

tended Data Fig. 5). At the module or pathway level, the liver-cirrhosis-

associated markers included assimilation or dissimilation of nitrate to

or from ammonia, denitrification, GABA(c-aminobutyric acid) biosyn-

thesis, GABA shunt, haem biosynthesis, phosphotransferase systems and

some types of membrane transport, such as amino-acid transport. The

control-enrichedmodules included histidine metabolism, ornithinebio-

synthesis, creatine pathway, carbohydrate metabolism, repair systems and

glycosaminoglycan metabolism (Supplementary Table 9).

The enrichment of the modules for ammonia production in patients

suggests a potential role of gut microbiota in hepatic encephalopathy, a

complication related to liver cirrhosis that is characterized by hyper-

ammonemia. Overproduction of ammonia by gut bacteria might con-

tribute to increased levels of ammonia in blood. Manganese-related

transport system modules enriched in patients possibly contribute to

the changes in concentrations of manganese.The accumulationof man-

ganese within the basal ganglia in patients with end-stage liver disease

may have a rolein the pathogenesis of chronic hepatic encephalopathy

a main complication of liver cirrhosis. The hydrodynamic venous shunt

and liver failure could promote this accumulation, which,in turn, causes

metabolic disorders ofthe nerve cell enzymes, affects transmission func-

tion of neural synapsesand eventuallyleads to hepatic encephalopathy

Finally, the modules for GABA biosynthesis were enriched in the patients.

The GABA neurotransmitter system is involved in the pathogenesis of

hepatic encephalopathy in humans

. Because of the hydrodynamic

venous shunt and liver failure, GABA levels in the blood are increased

and could go through the blood–brain barrier to activate GABA recep-

tor and cause hepatic encephalopathy. Microbiome modulation, aim-

ing at manganese elimination and lowering of GABA levels in the gut,

might provide a new therapeutic option for the treatment of hepatic

encephalopathy.

Microbial dysbiosis in chronic diseases

It is unclear whether a gutmicrobial dysbiosis in type 2 diabetes (T2D)

IBD

and liver cirrhosis

is similar or unique for each disease. We com-

pared the differences between the gut microbiota from patients with

liver cirrhosis, T2D and IBD,and organized the disease-associated gene,

KEGG orthologue group and eggNOG orthologue group markers into

patient- and control-enriched groups. We then identified markers common

to different disease pairs (T2D and liver cirrhosis, liver cirrhosis and

IBD, and IBD and T2D) and to the three diseases (Supplementary

Table 10). Different diseases displayed a relatively unique profile, even

if some markers were shared (Extended Data Fig. 6a, b). Most liver-

cirrhosis-enriched markers had low Pvalues (Extended Data Fig. 6c),

implying that patients with liver cirrhosis had more severe dysbiosis

than patients with T2D. Functional differences between liver cirrhosis

and T2D were also detected at the pathway level, even if there was a sig-

nificant increase in membrane transport markers in both (Extended

Data Figs 7 and 8). Most functional markers in both diseases were from

categories of carbohydrate metabolism, metabolism of cofactors and

vitamins, amino-acid metabolism and signal transduction. In contrast,

most cellmotility markersin the KEGG orthologue group were enriched

in liver cirrhosis or T2D but not both, possibly indicating a unique role

in each disease (Extended Data Fig. 8a, b). However, similar cell motility

markers and pathways in the KEGG orthologue group were enriched

both in liver cirrhosis and in T2D controls, suggesting a possible role

in health (Extended Data Figs 8c, d and 9a, b).

Gene markers that identify patients with liver cirrhosis

We used a pattern recognition techniqueto identify patients by gut mic-

robiota information in the discovery cohort (n5181). For this we selected

46,000 genes, half enriched in patients and half in controls (Supplemen-

taryTable 11). From this set we selected 15 optimal gene markers by a

minimum redundancy–maximum relevance (mRMR) method combined

with an incremental feature search, which showed the highest value of

Matthews correlation coefficient (Extended Data Fig. 9c). A support

vector machine discriminator was constructed using the same samples

and 15 gene markers (Supplementary Table 12), with the training and

leave-one-out cross-validation AUC (area under the receiver operating

characteristic curve) achieving 0.918(confidence interval: 0.881–0.955)

(Fig. 3b) and 0.838, respectively. The validation cohort of 31 healthy

controls and 25 patients with liver cirrhosis showed an AUC value of

0.836 (95% confidence interval 0.730–0.943) (Fig. 3c) for these samples,

confirming that the gut microbiota information could be applied to iden-

tify patients accurately.

To facilitate the clinical application of the 15 optimal gene markers,

we propose a patient discrimination index (PDI). The high correlation

coefficient value between the ratio of patients in our cohort and thePDI

(Fig. 3a and Supplementary Table 13) indicates that the PDI could be

used to identify patients with liver cirrhosis. The discriminatory power

of the PDI was then validated using an independent group (Fig. 3d).

The average PDI index between the control and the patient groups was

significantly different (P,8.18 310

, Wilcoxon r ank-sum test), con-

firming the potential use of gut microbiota information for identifying

patients with liver cirrhosis.

Discussion

To study gut microbiota in liver cirrhosis we firstestablished a novel gut

gene catalogue (liver cirrhosiscatalogue),including 98 patients withliver

cirrhosis and 83 healthy control individuals. Comparison with the previ-

ously established MetaHIT and T2D

gene catalogues indicated a com-

mon core of approximately 800,000 genes and a considerable propor-

tion of catalogue-specific genes (37.01% of MetaHIT, 36.59% of T2D

and 18.02% of liver cirrhosis), indicating that the current gene sets are

still limited and should be completed by inclusion of more individuals.

Interestingly, although the T2D and liver cirrhosis gene sets are both

derived from Chinese populations, the number of unique genes in each

gene set was large. Thismight be due to the difference in diseaseprofiles

and to the different genotypes, body mass indices, age

and dietary

habits

(Supplementary Table 14 and Extended Data Fig. 10). Never-

theless, there was no significant difference in the abundance of main

phyla (P.0.01);of the top 30 most abundant generaand species, 28 and

26, respectively, were the same in both studies, and there were no signi-

ficant differences in abundance for most of them. Furthermore, the top

four species were exactly the same. These results, and the similarity of

RESEARCH ARTICLE

4 | NATURE | VOL 000 | 00 MONTH 2014

controls withthe healthy Danish population, point towards overall sim-

ilarity of the microbiota in healthy individuals.

Use of the liver cirrhosis gene catalogue, in conjunction with the quan-

titative metagenomics approach, revealed a major change of the gut mic-

robiota in the patients with liver cirrhosis, mainly because of a massive

invasion of the gut by oral bacterial species. Correlation of the severity

of the disease with the abundance of the invading species suggests that

they may play an active role in the pathology. This was not noted in a

previous study, where the 16S-based approach probably lacked the required

species-level resolution, even if similar trends in taxonomy change between

the liver cirrhosis group and the healthy controls at the phylum, class

and order levels were observed

. Some of the MGS depleted in patients

were negatively associated with the severity of the disease (Fig. 2). This

opens avenues to the development of novel probiotics, which might help

combat the aggravation of liver cirrhosis. More generally, modulation

of microbiota to correct the major dysbioses we report might open new

avenues to treatment of liver cirrhosis.

A combination of 15 microbial genes discriminates patientswith liver

cirrhosis from healthy individuals, with a high specificity. This could

lead to a new way of monitoring and preventing liver cirrhosis. None of

the 15 markers found in the liver cirrhosis studyoverlapped with the 50

markers found in the T2D study

, indicating that diagnosis of different

diseases with microbiota-targeted biomarkers may be a powerful tool

for disease detection.

Online Content Methods, along with any additional Extended Data display items

and SourceData, are available in theonline version of the paper;references unique

to these sections appear only in the online paper.

Received 7 April 2013; accepted 9 June 2014.

Published online 23 July 2014.

1. Fouts, D. E., Torralba, M., Nelson, K. E., Brenner, D. A. & Schnabl, B. Bacterial

translocation and changes in the intestinal microbiome in mouse models of liver

disease. J. Hepatol. 56, 1283–1292 (2012).

2. Cesaro, C. et al. Gut microbiota and probioticsin chronic liver diseases. Digest. Liver

Dis. 43, 431–438 (2011).

3. Wiest,R. & Garcia-Tsao, G. Bacterial translocation (BT) in cirrhosis. Hepatology41,

422–433 (2005).

4. Nolan, J. P. The role of intestinal endotoxin in liver injury: a long and evolving

history. Hepatology 52, 1829–1835 (2010).

5. Gill,S. R. et al. Metagenomic analysis of the humandistal gut microbiome. Science

312, 1355–1359 (2006).

6. Garcia-Tsao, G. & Wiest, R. Gutmicroflora in the pathogenesisof the complications

of cirrhosis. Best Pract. Res. Clin. Gastroenterol. 18, 353–372 (2004).

7. Wiest,R., Krag, A. & Gerbes, A. Spontaneous bacterial peritonitis: recent guidelines

and beyond. Gut 61, 297–310 (2012).

8. Bass, N. M. et al. Rifaximin treatment in hepatic encephalopathy. N. Engl. J. Med.

362, 1071–1081 (2010).

9. Benten, D. & Wiest, R. Gut microbiome and intestinal barrier failure–the ‘‘Achilles

heel’’ in hepatology? J. Hepatol. 56, 1221–1223 (2012).

10. Yan, A. W. et al. Enteric dysbiosis associated with a mouse model of alcoholic liver

disease. Hepatology 53, 96–105 (2011).

11. De Filippo, C. et al. Impact of diet in shaping gut microbiota revealed by a

comparative study in children from Europe and rural Africa. Proc. Natl Acad. Sci.

USA 107, 14691–14696 (2010).

12. Cho, I. & Blaser, M. J. The human microbiome: at the interface of health and

disease. Nature Rev. Genet. 13, 260–270 (2012).

13. Chen, Y. et al. Characterization of fecal microbialcommunities in patientswith liver

cirrhosis. Hepatology 54, 562–572 (2011).

14. Nelson, K. E. et al. A catalog of reference genomes from the human microbiome.

Science 328, 994–999 (2010).

15. Ley, R. E., Turnbaugh, P. J., Klein, S. & Gordon, J. I. Microbial ecology: human gut

microbes associated with obesity. Nature 444, 1022–1023 (2006).

16. Turnbaugh, P. J. et al. An obesity-associated gut microbiome with increased

capacity for energy harvest. Nature 444, 1027–1031 (2006).

17. Turnbaugh, P. J. et al.A core gut microbiome in obese and lean twins. Nature 457,

480–484 (2009).

18. Ley, R. E. et al. Obesity alters gut microbial ecology. Proc. Natl Acad. Sci. USA 102,

11070–11075 (2005).

19. Lepage, P. et al. Twin study indicates loss of interaction between microbiota and

mucosa of patients with ulcerative colitis. Gastroenterology 141, 227–236 (2011).

20. Garrett, W. S. et al. Enterobacteriaceae act in concert with the gut microbiota to

induce spontaneous and maternally transmitted colitis. Cell Host Microbe 8,

292–300 (2010).

21. Wen, L. et al. Innateimmunity and intestinal microbiota in the development of type

1diabetes.Nature 455, 1109–1113 (2008).

22. Qin, J. et al. A metagenome-wide association study of gut microbiota in type 2

diabetes. Nature 490, 55–60 (2012 ).

23. Vijay-Kumar, M. et al. Metabolic syndrome and altered gut microbiota in mice

lacking Toll-like receptor 5. Science 328, 228–231 (2010).

24. Karlsson, F. H. et al. Symptomatic atherosclerosis is associated with an altered gut

metagenome. Nature Commun. 3, 1245 (2012).

25. The HumanMicrobiome Project Consortium. A frameworkfor human microbiome

research. Nature 486, 215–221 (2012).

26. The Human Microbiome Project Consortium. Structure, function and diversity of

the healthy human microbiome. Nature 486, 207–214 (2012).

27. Le Chatelier,E. et al. Richness of human gut microbiomecorrelates with metabolic

markers. Nature 500, 541–546 (2013).

Sensitivity

0.0

0.2

0.4

0.6

0.8

1.0

1-Specicity

Sensitivity

0.0

0.2

0.4

0.6

0.8

1.0

0 0.2 0.4 0.6 0.8 1.0 0 0.2 0.4 0.6 0.8 1.0

–15 –10 –5 0 5

Controls

Case

–1.0 –0.5 0.0 0.5 1.0

Controls

Case

Median

AUC = 91.8%

Condence interval: 88.1–95.5%

AUC = 83.6%

Condence interval: 73.0–94.3%

–1.5 –1 0 1 2 3 3.5

PDI

Number of individuals

−1.5 −1 0 1 2 3 3.5

0.2

0.4

0.6

0.8

1.0

PDI

Percentage of patients

1-Specicity

Figure 3

PDI on the basis of gut microbial biomarkers. a, A PDI was

calculated for each individual from 15 gene markers selected using the mRMR

approach to evaluate the risk of liver cirrhosis. The filled blue circles show the

distribution of liver cirrhosis indices for all individuals (bins of 0.5 PDI units

were used; values less than 21.5 and greater than 3.5 were grouped). Inset, the

proportion of patients with liver cirrhosis in the corresponding bins. b,c,The

AUC is shown for the training (b) and validation (c) samples. d, The liver

cirrhosis PDI was computed for an additional 25 liver cirrhosis samples and 31

healthy control samples. The box depicts the interquartile range between the

first and third quartiles (25th and 75th percentiles, respectively); the line inside

denotes the median. Inset, the PDI without the outliers.

ARTICLE RESEARCH

00 MONTH 2014 | VOL 000 | NATURE | 5

28. Cotillard, A. et al. Dietary intervention impact on gut microbial gene richness.

Nature 500, 585–588 (2013).

29. Qin, J. et al. A human gut microbial gene catalogue established by metagenomic

sequencing. Nature 464, 59–65 (2010).

30. Nielsen, H. B. et al. Identification and assembly of genomes and genetic elements

in complex metagenomic samples without using reference genomes. Nature

Biotechnol. http://dx.doi.org/10.1038/nbt.2939 (2014).

31. Albertsen, M. et al. Genome sequences of rare, uncultured bacteria obtained by

differential coverage binning of multiple metagenomes. Nature Biotechnol. 31,

533–538 (2013).

32. Zoetendal, E. G. et al. The human small intestinal microbiota is driven by rapid

uptake and conversion of simple carbohydrates. ISME J. 6, 1415–1426 (2012).

33. Bauer, T. M. et al. Small intestinal bacterial overgrowth in human cirrhosis is

associated with systemic endotoxemia. Am. J. Gastroenterol. 97, 2364–2370

(2002).

34. Chen, T. et al. The Human Oral Microbiome Database: a web accessible resource

for investigating oral microbe taxonomic and genomic information. Database

2010, baq013 (2010).

35. Pagani, I. et al. The Genomes OnLine Database (GOLD) v.4: status of genomic and

metagenomic projects and their associated metadata. Nucleic Acids Res. 40,

D571–D579 (2012).

36. Saarela, M., Mogensen, G., Fonden, R., Matto, J. & Mattila-Sandholm, T. Probiotic

bacteria: safety, functional and technological properties. J. Biotechnol. 84,

197–215 (2000).

37. Merritt, M. E. & Donaldson, J. R. Effect of bile salts on the DNA and membrane

integrity of enteric bacteria. J. Med. Microbiol. 58, 1533–1541 (2009).

38. Marchandin, H. et al. Prosthetic joint infection due to Veillonella dispar.Eur. J Clin.

Microbiol. Infect. Dis. 20, 340–342 (2001).

39. Hwang, J. J.,Lau, Y. J., Hu, B. S., Shi, Z. Y. & Lin, Y. H. Haemophilus parainfluenzaeand

Fusobacterium necrophorum liver abscess: a case report. J. Microbiol. Immunol.

Infect. 35, 65–67 (2002).

40. Xu, M. et al. Changesof fecal Bifidobacteriumspecies in adult patientswith hepatitis

B virus-induced chronic liver disease. Microb. Ecol. 63, 304–313 (2012).

41. 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).

42. Krieger, D. et al. Manganese and chronic hepatic encephalopathy. Lancet 346,

270–274 (1995).

43. Ferenci,P., Schafer, D. F., Kleinberger, G., Hoofnagle,J. H. & Jones, E. A. Serumlevels

of gamma-aminobutyric-acid-like activity in acute and chronic hepatocellular

disease. Lancet ii, 811–814 (1983).

44. Minuk,G. Y., Winder, A., Burgess, E. D. & Sarjeant,E. J. Serum gamma-aminobutyric

acid(GABA) levels in patientswith hepaticencephalopathy.Hepatogastroenterology

32, 171–174 (1985).

45. Yatsunenko, T. et al. Human gut microbiome viewed across age and geography.

Nature 486, 222–227 (2012).

46. 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 This work was supported by the National Program on Key Basic

Research Project (2013CB531401),the National Natural ScienceFoundation of China

(81301475 and 81330011), the Science Fund for Creative Research Groups of the

National Natural Science Foundation of China (81121002), the Technology Group

Project for Infectious Disease Control of Zhejiang Province (2009R50041) and the

Metagenopolis grant ANR-11-DPBS-0001. We thank Q. Cao, K. Su, J. Shao and

A. Ghozlane for help with data computation, and H. Zhang, H. Lu, Q. Bao, J. Ge, J. Jiang,

Z. Ren and M. Ye for assistance with sample collection.We are thankful to the MetaHIT

consortium for generating the gut gene set and the Human Microbiome Project for

generating the reference genomes from human gut microbes.

Author ContributionsL.J.L., S.D.E., S.S.Z.and N.Q. designed the project. L.J.L.,S.P.K. and

N.Q. managed the project. F.L.Y., N.Q., Y.F.C., J.G., G.R.Q., X.J.H. and B.W.Z. collected

samples and performed clinical study. J.G., Y.T.C. and W.X. performed DNA extraction

experiments. Y.J., L.J.W., J.W.Z. and S.J.N. performed library construction and

sequencing. L.J.L. and S.D.E. designed the analysis. N.Q., A.L.,E.P., E.L.C., L.L., N.P., P.L.,

J.M.B., C.H.Y. and W.C.D. analysed the data. A.L. and N.Q. did the functional annotation

analyses. L.S.,E.P., E.L.C. and A.L. analysed the statistics. N.Q., F.L.Y., L.S. and E.P. wrote

the paper. L.J.L. and S.D.E. revised the paper.

Author Information The raw Illumina read data for all samples have been deposited in

the European Bioinformatics Institute European Nucleotide Archive under accession

number ERP005860. 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 onlineversion of the paper. Correspondence

and requests for materials should be addressed to L.J.L. (ljli@zju.edu.cn),

S.S.Z. (zyzsss@zju.edu.cn) or S.D.E. (dusko.ehrlich@jouy.inra.fr).

RESEARCH ARTICLE

6 | NATURE | VOL 000 | 00 MONTH 2014

METHODS

Patient information. Liver cirrhosis was diagnosedaccording to the international

guidelines by comprehensive consideration of liver biopsy, imaging examination,

clinical symptoms, physical signs, laboratory tests, medical history, progress notes

and cirrhosis-associated complications. Biopsy as the ‘gold standard’ for cirrhosis

diagnosis was used for 46 out of the 123 (37.4%) patients. As biopsy was counter-

indicated for patientswith conditions such as refractory ascites and obviousbleed-

ing tendency, the remaining77 (62.6%) were diagnosed using all other approaches

combined.To confirm diagnoses, we solicitedoutside expert opinions for each case.

Borderline or otherwise inconclusive cases were excluded from the study. After

discharge of the patient from the hospital, their case history was further reviewed

for medication history. Cases that progressed to hepatic carcinoma or those found

to suffer from other diseases such as hypertension and diabetes were excluded.

The control group included 114 healthy volunteers who visited the First Affiliated

Hospital of Zhejiang University in China for their annual physical examination.

The liver imaging and liver biochemistry results of all healthy controls were in the

normal range. Physical examination, routine e xamination of blood, urine and stools,

preoperative serological tests (including the detection of hepatitis B surface antigen,

hepatitis C virus antibody, Treponema pallidum antibody, human immunodefi-

ciency virus antibody), liver function, renal function, electrolyte, liver ultrasound,

electrocardiogram andchest X-ray results were checked in the healthy controls to

exclude any abnormal samples. Comprehensive clinical information for each enrolled

individual was recorded (Supplementary Table 1). Exclusion criteria for the con-

trol groupincluded hypertension, diabetes, obesity, metabolic syndrome, IBD, non-

alcoholic fatty liver disease, coeliac disease and cancer. Individuals who received

antibiotics and/or probiotics within 8 weeks before enrolment were also excluded.

All participants, or their legally authorized representatives, provided a written informed

consent upon enrolment.The study conformed to the ethical guidelinesof the 1975

Declaration of Helsinkiand was approved by the Institutional ReviewBoard of the

First Affiliated Hospital of Zhejiang University.

Human faecal sample collectionand DNA extraction. Each cirrhotic patient and

healthy individual provided a fresh stool sample that was delivered immediately

from our hospital to the laboratory in an ice bag using insulating polystyrene foam

containers. In the laboratory it was divided into five aliquots of 200 mg and imme-

diately stored at 280 uC. A frozen aliquot (200mg) of each faecal sample was pro-

cessed by phenol trichloromethane DNA extraction

16,47

as previously described. DNA

concentration was measured by NanoDrop (Thermo Scientific) and its molecular

size was estimated by agarose gel electrophoresis.

DNA library construction and sequencing. DNA libraries were constructed accord-

ing to the manufacturer’s instructions (Illumina). The same workflows from Illumina

were used to perform cluster generation, template hybridization, isothermal amp-

lification, linearization, blocking, denaturing and hybridization of the sequencing

primers. We performed paired-end sequencing on 2 3100 base pairs (bp) for all

libraries. The base-calling pipeline (Casava 1.8.2 withparameters ‘-use-bases-mask

y100n, I6n,Y100n, -mismatches1, -adaptor-sequence’) was used to processthe raw

fluorescent images and callsequences. The sameinsert size inferredby Agilent 2100

was used for all libraries (ranging from 275 to 450).

Quality control of reads. Reads that mapped to human genome together with

their mated/paired readswere removed from each sample using BWA

with para-

meter ‘-n 0.2’. Then qualitycontrol used the following criteria: (1) readscontaining

more than 3 Nbases were removed; (2) reads containing more than 50 bases with

low quality (Q2) were removed; (3) no more than 10 bases with low quality (Q2)or

assignedas Nin the tail of reads were trimmed. Sequences thatlost their mated reads

were consideredas single reads and were used in the assemblyprocedure. Resulting

filtered reads were considered for the next step of the analysis.

De novo

assembly of the Illumina short reads. Considering that k-mers with

very low frequencies might arise from sequencing errors, they were not used in

assembly by SOAPdenovo

(version 1.05), whichis based on De Brujin graph con-

struction. SOAPdenovo (version 1.05) was used in Illumina short read assembly

with parameters ‘-d 1 -M 3’. Then we removed ambiguous bases from assembled

scaffolds (this coulddivide one scaffold into multiple ones) and discardedscaffolds

with lengths less than 500bp. Finally we tested series of k-mer values (from 31 to

59), then choseone with the longest N50 value for theremaining scaffolds. For each

sample, we mapped clean data against scaffolds using SOAPalignversion 2.21 (ref. 50)

with parameters ‘-u -2 -m 200’. Unused data from each sample were pooled and split

into four parts (considering memory limit). Unused reads were repeatedly assembled

with the same parameters but only one k-mer value, -K 55, was chosen.

Construction of non-redundant human gut gene set. Total DNA was extracted

from the faecal samples of 98 Chinese patients with liver cirrhosis and 83 healthy

Chinese controls(Supplementary Table 1) and sequenced using an IlluminaHiSeq

2000 (Illumina). This produced an average of 4.74 gigabases (Gb) of high-quality

sequence for each sample, providing a total of 858 Gb of sequence data (Supplementary

Table 15). The reads wereassembled into contigs for all samples using the assembly

software SOAPdenovo

.Unassembled reads from 166 samples were pooled and

the de novo assembly process was performed again for these reads (Extended Data

Fig. 9d). Finally, 61.68% of the totalreads were used to generate 4.4 million contigs

without ambiguous bases (minimum length of 500 bp). These contigs had a total

length of 11.1 Gb, an average N50 length of 8,644 bp and ranged from 1,673 to

48,822 bp (SupplementaryTable 15). To predict microbial genesfor each of the 181

samples, we applied the methodology used in the MetaHIT human gut gene cata-

logue study

. The non-redundant human gut gene set was built by pairwise com-

parison of all thepredicted ORFs using blat and the redundant ORFs were removed

using a criterion of 95% identity over 90% of the shorterORF length, which is con-

sistent with the criterion used for the non-redundant European human gut gene

set

and T2D study

MetaGeneMark

(prokaryoticGeneMark.hmm version 2.8) was used to predict

ORFs in scaffolds without ambiguous bases. The program predicted 13,371,697

ORFs using a 100 bp cut-off for p rediction (Supplementary Table 15). The total length

of the predicted ORFs was 9,495,923,532 bp, represen ting 90.28% of the total length of

the contigs. Among the ORFs, 1,047,885 (54.6%) were complete genes, while 869,808

(45.4%) were incomplete. A non-redundant ‘liver cirrhosis gene set’ was established

by removing redundant ORFs, defined as those sharing 95% identity over 90% of the

shorter ORF lengthin pairwise alignments. The final non-redundantliver cirrhosis

gut gene set contained 2,688,468 ORFs, with an average length of 750 bp and 42%

of reads could be aligned to the gene catalogue.

Then genes from the liver cirrhosis, T2D and MetaHIT catalogues were merged

to create a non-redundant gene set for subsequent analyses. We checked the gaps

and frames in the blat results; if there were gaps or the frames were different in the

alignment result of two ORFs, the shorter one would not be removed as a redund-

ancy.We used MetaGeneMarkto predict genesin assembled contigs originallyfrom

MetaHITand T2D and mergedthese three gene setsinto a single onewith the above

method.

Organism abundanceprofiling.SO APalign2. 21 was used to align paired-end clean

reads against referencegenomes with parameters ‘–r 2 –m 200 –x 1000’. Reads with

alignments on the same referencegenomes could be assigned into two types, as fol-

lows. (1) Unique reads (U): reads having alignments with only one genome. These

reads were denoted asunique reads.(2) Multiple reads (M): reads having alignments

with more than one genome. If these genomes came from one species, we denoted

these reads as unique reads. If they were from more than one species, we denoted

these reads as multiple reads.

For species S, if its abundance is Ab(S), and it might have alignments with U

unique reads and Mmultiple reads, the computation is

Ab SðÞ~Ab(U)zAb(M)

Ab(U)~U=l

Ab(M)~(X

i~1

Co fMg)=l

Ab(U) and Ab(M) are abundance of unique and multiple reads, respectively, and l

is length of relative genome. For each multiple read, there is a species-specific coe-

fficient Co; let us suppose one read in {M} has alignments with Ndifferent species,

then Co was calculated as follows:

Co~U=X

i~1

Ab(U)

For these reads,we add a unique abundance of Nspecies as thedenominator. Before

we calculate the abundance of species S, we calculate Ab(U) for all species as con-

stants; if Ab(U) of species Sis 0, then Co will also be 0, and consecutively the abun-

dance of species Sis 0. Species abundance was added to obtain the genus-level profile

table. For some species that do not have a genus, they are denoted as unclassified

genera for each species.

Gene abundance profiling. Reads were aligned against the gene set by using

SOAPalign

with parameters ‘-r -m 200 -x 1000’. We counted a gene’s abundance

if both paired-end reads could be aligned on the same gene. If onlyone of the paired-

end reads could be aligned on a gene, we aligned both reads against assembled

contigs by checking if the previously non-aligned read were in the non-translated

region or not. If true, both reads were validated for gene count; if not, both reads

were discarded.

When calculating the abundance of genes, we used the same strategy as for

the abundance profiling of the organisms. For a given gene G, its abundance is

Ab(G), and it might have alignments with Uunique reads and Mmultiple reads,

as follows:

ARTICLE RESEARCH

Ab GðÞ~Ab(U)zAb(M)

Ab(U)~U=l

Ab(M)~(X

i~1

Co fMg)=l

Ab(U) and Ab(M) are the abundances of unique and multiple reads, respectively,

and lis length of gene G. For each multiple read, we calculate a specific coefficient

Co for this gene. Let us suppose one read with multiple {M} alignments in Ndif-

ferent genes, then Co was calculated as follows.

Co~U=X

i~1

Ab(U)

For these reads, we add a unique abundance of Nspecies as the denominator.

Population stratification. Population stratification involved in our metagenomic

data was corrected with the modified EIGENSTART method as follows, First, sin-

gular value decomposition was carried out to obtain axes of variation, where the

number of significant axes wasdetermined according to a Tracy–Widom test at a

significance level of P,0.05; each axis was then replaced with the residuals of this

axis from a regression to disease state; the corrected data were finally achieved by sub-

tracting from original data set the information associated with the residuals of each axis.

Gene count determination. Gene counts were computed essentially as described

in ref. 27. Briefly, datawere downsized to adjust for sequencingdepth and technical

variability by randomly selecting 6.2 million reads mapped to the merged gene

catalogue for each sample and thencomputing the mean number of genes over 30

random drawings (Supplementary Table 4). This was possible for all but two patients

with liver cirrhosis from thevalidation cohort (with insufficient number of mapped

reads), who were excluded from this analysis. The results are displayed in Extended

Data Fig. 4 top left.

Gene functional classification and orthologue group abundance profiling. Protein

sequences of the predictedgenes were searched using National Center for Biotech-

nology Information blastP against the eggNOG 3.0 database

and the KEGG gene

database(KEGG FTP release21 January 2013) withparameters ‘-num_descriptions

100000, -evalue 1e-5’. Genes that had alignments with a bits score higher than 60

were assignedinto one or more eggNOG or KEGG orthologue groups.We used the

methods introduced in ref. 29 to calculate abundance of proteins archived in the

eggNOGand KEGG databases. To calculate abundances of eggNOGor KEGG orth-

ologue groups, we added abundances of proteins assigned into the same eggNOG or

KEGG orthologuegroups, as abundances of eggNOGor KEGG orthologue groups,

then profiles of eggNOG/KEGG orthologue groups were generated.

Gene biomarker identification. Genes from the gene-profile matrix were used in

an association study aimed at identifying those that were differentially abundant

between the patient and the healthycontrol groups. Wilcoxon tests were employed

to compute the probabilitiesthat frequencyprofilesdid not differbetweenthe patient

and the healthy controlgroups by chance alone. Benjamini and Hochberg multiple

test correction wasapplied to the Pvalues. By performing a selection only based on

a threshold of P,0.01, we found 541,582 genes.For specificity and computational

reasons, we useda very stringent significance thresholdof FDR ,0.0001. This pro-

cess identified 75,245 genes that were differentially abundant between the groups

(49,830 were more abundant in the patients with liver cirrhosis and 25,415 in the

healthy controlgroup). A similar Pvalue and group enrichmentmethod was calcu-

lated for the NOG/KEGG orthologue groups as well.

MGS. We followed the approach described in refs 27 and 30 to cluster genes from

the current study into MGS. Briefly, in a first step the pairwise Spearman’s corre-

lation coefficient (r) of different genes was computed, using gene abundances

across all individuals, and the genes correlated over a given threshold were clus-

tered (single-linkage clustering). To favour clustering specificity (that is, assigning

only the genes of the samespecies to the same cluster) we used a ratherhigh thresh-

old (r.0.7). To correct for the concomitant loss of sensitivity, we performed a

second step wherebythe mean abundance signal of each cluster of atleast 50 genes

was computed, using the 50 most connected genes of a cluster. The clusters that

had r.0.85 were fused. This procedurewas applied separately to the 49,830 genes

enriched in patients with liver cirrhosis and the 25,415 genes enriched in healthy

controls. Of the 25,415‘healthy’ genes, 21,423 fell into 43 clusters composed of 51–

2,702 genes after the first clustering step, and 38 clusters of 51–2,970 genes after the

second step. Of the ‘liver cirrhosis’genes, 31,386 out of 49,830 fell into 60 clusters of

51–3,000 genes afterthe first clustering step, and 28 clusters of 51–5,755 genesafter

the second step.

To verify that the genes from a given cluster belonged to the same genome and

to annotatethe MGS taxonomically,we performed blastNand blastP analysesusing

a collection of 6,006 genomes (the available reference genomes from the National

Center for Biotechnology Information and the set of draft gastrointestinal gen-

omes from the Data Analysis and Coordination Center of the HMP and MetaHIT

(3 August 2012 version)). MGS were assigned to a given genome when more than

80% of its ‘tracer genes’

matched the same genome using blastN,at a threshold of

95% identity over 90%of gene length.Six ‘healthy’ and 24 ‘livercirrhosis’MGS could

thus be assigned to the strainlevel (see Extended Data Fig. 9e, f and Supplementary

Table 4). The remaining MGS wereannotated using blastP analysis and assigned to

a given taxonomical level from genus to superkingdom level if more than 80% of

their 50 tracer genes had the same level of assignment

. All but one of the 36 remain-

ing species could thus be assigned to a given genus, family or order (see Supplemen-

tary Table 4). The quality of the clustering was thus validated by the homogenous

annotation of its markergenes, which also held true for all of the MGS genes (data

not shown). The abundanceof the 66 MGS in each individual was computed using

the 50 tracer genes.

To explore the origin of the species-level annotated MGS, we constructed a ref-

erence catalogue, grouping114 publicly availableStreptococcus (57),Fusobacterium

(26), Lactobacillus (16), Veillonella (12) and Megasphaera (3) genomes, mostly of

oral (50) or gut (28) isolates (Supplementary Table 6). The 16 liver cirrhosis MGS

that were assigned to the corresponding genera were compared with the genomes,

using blastN. A score (T) was computed for each MGS, taking into account (1) the

proportion of genes above 95%identity and 90% coverage (Q), (2) the average iden-

tity (R), (3) the average coverage (S)and(4)T5Q3R3S.

A majority of the MGS enriched in patients with liver cirrhosis (15 out of 28)

were of oral origin by this criterion whereas six were from gut or faeces, including

a single species from the ileum (Supplementary Table 4 and Extended Data Fig. 4

top right). To explore further the origin of the liver-cirrhosis-enriched MGS, we

comparedthem by blastN withthe genes from three available ileum metagenomes

and failed to reveal identity beyond that found with sequenced genomes.

Only a small minority of the 38 MGS enriched in healthy individuals (15.8%)

could be assigned speciesphylogenetic information by comparisonwith sequenced

gut genomes using blastN(95% identity and 90% overlap; SupplementaryTable 4).

Annotation to comparabletaxonomic levels was observed for the 58 gut MGS ana-

lysed in the context ofgene richness in a Danish cohort

(Extended Data Fig. 9e, f),

reflecting a paucityof isolated and sequenced gut strains. Furthermore, it is striking

that all 38 MGS enriched in healthy Chinese were found in the Danish cohort

(Extended Data Fig. 3). In sharp contrast with the MGS enriched in healthy sub-

jects, an overwhelming majority of the MGS enriched in patients (24 out of 28)

could be assigned to a species. Such a difference has a vanishingly low probability

of being caused by chance alone (1.3 310

221

by a x

test, Extended Data Fig. 9e, f)

and indicates a highly modified composition of gut microbes.

Co-occurrence network of MGS. The 66 marker profiles of the differentially

abundant MGS betweenpatient and healthy individuals werecorrelated separately

for patients and for healthy individuals, essentially as described in ref.53. For each

of the 2,112 possible edges [(66 366/2) – 66] we computed 1,000 permutations by

renormalizing the data after each step and computed Spearman’s correlation coe-

fficients to obtain the null distributions due to the compositionality effect

. For

each of the edges we also computed the bootstrap distribution of the Spearman’s

correlation coefficients to have the confidence interval and the corresponding var-

iance. We next applied for each edge a z-test with the pooled variance from both

distributions and computeda significance Pvalue.Multiple testingcorrections were

applied to the Pvalues using the Benjamini and Hochberg method, and only those

having FDR ,10

were used to construct the network. This FDR threshold cor-

responds approximately to r.0.4. The network reflects strong correlations that

are not spurious and that are not due to the compositionality effect. The resulting

network is displayed as Fig. 2.

Marker selection by mRMR. Patient discrimination gene markers (23,000 from

healthy controls and 23,000 from patients, selected as most discriminant by the

Wilcoxonrank-sum test upon adjustment for age, performed as described in ref.54;

Supplementary Table 11) were selected with a two-step scheme (using the side

Channel Attack R package). All markers retained were first filtered by the mRMR

algorithm

(using the side Channel Attack R package), and the top 180 best ones

were selected for further analysis. Then, we performed an incremental search to

select theoptimal subset of genes,named as markers. Concisely, genes were sequen-

tiallyadded into the subset with a step of 5, the performance of which was evaluated

on the basis of linear discriminant analysis and leave-one-out cross-validation.

Here, Matthews correlation coefficient is a balanced measure taking into account

true and false positives and negatives; it is superior to accuracy or error rate when

the classes (healthy and diseased, etc.) are of very different sizes. Matthews corre-

lation coefficient (MCC) is defined as

MCC~TP|TN{FP|FN

ﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃﬃ

TPzFPðÞTPzFNðÞTNzFPðÞTNzFNðÞ

where TP, TN, FP and FN are true positive, true negative, false positive and false

negative, respectively. We finally selected a set of 15 gut microbial gene markers as

the optimal selection for patient discrimination.

RESEARCH ARTICLE

Model construction and validation. On the basis of the 15 metagenomic markers

described above, a support vector machine classifier (radial basis function kernel

and default parameters)was constructed for patient discrimination(realized by the

e1071 package of R software), the performance of which was assessed by receiver

operating characteristic analysis. The AUC and corresponding 95% confidence

intervals for trainingand validation data sets, obtained by using the pROC package

of R software (10,000 bootstrap replicates), were 0.97 (0.95–0.99) and 0.889 (0.79–

0.98), respectively.

Definition of PDI. To facilitate clinical application of the selected 15 metage-

nomic markers, we defineda more straightforward index (PDI) for discrimination

of patients. For each individual sample, the PDI of sample jthat was denoted by I

was computed as follows:

j~X

i[N

Aij

j~X

i[M

Aij

Ij~

{

|106

where A

is the relative abundance of marker iin sample j.Nand Mare subsets of

patient- and control-enriched markers in these 15 selected gut metagenomic mar-

kers, respectively. Moreover, jNjand jMjare the sizes of these two sets.

47. Li, M. et al. Symbiotic gut microbes modulate human metabolic phenotypes.

Proc. Natl Acad. Sci. USA 105, 2117–2122 (2008).

48. Li, H. & Durbin, R. Fast and accurate shortread alignment with Burrows-Wheeler

transform. Bioinformatics 25, 1754–1760 (2009).

49. Li, R. et al. De novo assembly of human genomes with massively parallel short

read sequencing. Genome Res. 20, 265–272 (2010).

50. Li, R. et al. SOAP2: an improved ultrafast tool for short read alignment.

Bioinformatics 25, 1966–1967 (2009).

51. Noguchi, H., Park, J. & Takagi, T. MetaGene: prokaryotic gene finding from

environmental genome shotgun sequences. Nucleic Acids Res. 34, 5623–5630

(2006).

52. Powell, S. et al. eggNOG v3.0: orthologous groups covering 1133 organisms

at 41 different taxonomic ranges. Nucleic Acids Res. 40, D284–D289

(2012).

53. Faust, K. et al. Microbial co-occurrence relationships in the human microbiome.

PLOS Comput. Biol. 8, e1002606 (2012).

54. Price, A. L. et al. Principal components analysis corrects for stratification in

genome-wide association studies. Nature Genet. 38, 904–909 (2006).

55. Ding, C. & Peng, H. Minimum redundancy feature selection from microarray

gene expression data. J. Bioinform. Comput. Biol. 3, 185–205 (2005).

ARTICLE RESEARCH

Extended Data Figure 1

Venn diagram comparing the current major

human microbiome gene set and the results of a principal component

analysis of biomarkers distributed between patients with liver cirrhosis and

healthy controls. a, Venn diagram of the four currently available major human

microbiome gene sets. The total gene number in each gene set and the

overlapping areas are indicated.b, Venn diagram of the three major human gut

gene sets (LC, liver cirrhosis gene set; T2D, type 2 diabetes gene set; MetaHIT,

MetaHIT gene set; HMP, HMP gene set). c, Visualization of the principal

component analysis results for the liver-cirrhosis-associated genes that differed

significantly in the discovery cohort (FDR ,0.0001, Wilcoxon rank-sum test

adjusted for multiple testing). The principal component analysis is built here

using these genes in the validation cohort (25 patients with liver cirrhosis in red,

31 healthy controls in green).

RESEARCH ARTICLE

Extended Data Figure 2

Phylogenetic abundance at the phylum, genus

and species levels from liver cirrhosis and healthy control samples.

a, Phylogenetic abundance variation box plot at the phylum level and the 30

most abundant phylotypes at the genus and species levels in the healthy

controls are shown. Red, green, blue, turquoise and purple represent

Bacteroidetes, Firmicutes, Proteobacteria, Actinobacteria and other phyla,

respectively. The colour of each genus and species corresponds with the colour

of its respective phylum. b, Phylogenetic abundance variation box plot at the

phylum level and the 30 most abundant phylotypes at the genus and species

levels in the liver cirrhosis are shown (see Methods for the calculations). The

boxes represent the interquartile range, from the first and third quartiles, and

the inside line represents the median. The whiskers denote the lowest and

highest values within an interquartile range of 1.53 from the first and third

quartiles. The circles represent outliers beyond the whiskers.

ARTICLE RESEARCH

Extended Data Figure 3

MGS enriched in healthy Chinese individuals

(

5114) are present in Danish individuals (

5292). Presence and

abundance of 50 ‘tracer’ genes for each species; genes are in rows; abundance

is indicated by colour gradient (white, not detected; red, most abundant).

Individuals, ordered by increasing gene count, are in columns. Significance of

correlation of species abundance (computed as mean abundance of the tracer

genes) and gene count (qvalue, FDR adjusted) is given. Species in the Chinese

cohort that were identical to those previously found, as correlated with the

gene diversity in the Danish cohort

, are highlighted in red. Left, the Chinese

healthy cohort. Right, the Danish obesity cohort.

RESEARCH ARTICLE

Extended Data Figure 4

Massive changes in the gut microbiome in liver

cirrhosis. Top left, healthy individuals have more gut microbial genes than

patients with liver cirrhosis. Gene count was computed after downsizing the

mapped reads to a level of 6.2 million(ref. 27). The significance of the difference

was computed using a Student’s t-test. Bottom, abundance of patient-enriched

species (n528) in patients with liver cirrhosis (n598) and healthy controls

(n583). The relative abundance of each patient-enriched species was

computed as a sum of the abundances of all the genes assigned to it divided by

the sum of the abundances of all gut microbial genes in each patient, which is

equal to 1 in the normalized data set. Bar length indicates the relative

abundance of a given species depicted by a different colour. Patients were

ordered by the total patient-enriched species abundance; LPA and HPA

quartiles (n524) are separated by red vertical lines. Top right, oral species are

frequent in patients with liver cirrhosis. MGS enriched in healthy controls

are largely not assigned to a species level, while those enriched in patients with

liver cirrhosis are largelyassigned to a species level and are mostly of oral origin

(see Methods for species assignment).

ARTICLE RESEARCH

Extended Data Figure 5

The distribution of eggNOG orthologue group

and KEGG functional categories for liver-cirrhosis-related markers.

a, Comparison between the liver-cirrhosis-enriched and control-enriched

eggNOG orthologue group markers for 24 eggNOG orthologue group

functional categories shown by number. b, Comparison between the liver-

cirrhosis-enriched and control-enriched eggNOG orthologue group markers

for 24 eggNOG orthologue group functional categories shown by percentage.

c, Comparison between the liver-cirrhosis-enriched and control-enriched

KEGG orthologue groupmarkers for each KEGG functional categoryshown by

number. d, Comparison between the liver-cirrhosis-enriched and control-

enriched KEGG orthologue group markers for each KEGG functional category

shown by percentage.

RESEARCH ARTICLE

Extended Data Figure 6

A comparison of the gene markers for the

different groups. a, Venn diagram showing a gene marker comparison of

case-enriched gene markers from the liver cirrhosis, T2D and IBD studies.

b, Venn diagram showing a gene marker comparison of control-enriched gene

markers from the liver cirrhosis, T2D and IBD studies. c, The length of the bar

(yaxis) represents the numberof genes; the Pvalue in the relatedrange is shown

on the xaxis. The pink and light green bars show genes involved in type 2

diabetes and liver cirrhosis, respectively. Inset, the log Pvalue of the gene

markers between the two studies.

ARTICLE RESEARCH

Extended Data Figure 7

The distribution of eggNOG functional categories

for case-enriched and control-enriched gene markers in liver cirrhosis

only, T2D only and the liver cirrhosis/T2D groups. a, Comparison of the

eggNOG orthologue group functional categories for case-enriched gene

markers shown by number. b, Comparison of the eggNOG orthologue group

functional categories for case-enriched gene markers shown by percentage.

c, Comparison of the eggNOG orthologue group functional categories for the

control-enriched gene markers shown by number. d, Comparison of the

eggNOG orthologue group functional categories for the control-enriched gene

markers shown by percentage.

RESEARCH ARTICLE

Extended Data Figure 8

The distribution of the KEGG functional

categories for case-enriched and control-enriched gene markers in liver

cirrhosis only, T2D only or the liver cirrhosis/T2D group. a, Comparison of

the KEGG pathway categories for the case-enriched gene markers shown by

number. b, Comparison of the KEGG pathwaycategories for the case-enriched

gene markers shown by percentage. c, Comparison of the KEGG pathway

categories for the control-enriched gene markers shown by number.

d, Comparison of the KEGG pathway categories for the control-enriched gene

markers shown by percentage.

ARTICLE RESEARCH

Extended Data Figure 9

Estimating the optimum number of markers and

establishing the taxonomic assignment of MGS. a, Comparison of the case-

enriched gene markers. b, Comparison of the control-enriched gene markers.

c, The mRMR method was used to identify the liver-cirrhosis-associated

markers. Sequential subsets were generated at five-marker intervals. For each

subset, the error rate was estimated using a leave-one-out cross-validation of a

linear discrimination classifier. The optimum (highest value of the Matthews

correlation coefficient) subset contains 15 gene markers. d, The study included

a discovery and a validation phase. Volunteers for both phases were recruited in

the same hospital. Both direct read mapping and de novo assembly were

performed for each sample. A taxonomy profiling table was established for

taxonomy analysis. A novel gut gene set was established, and annotated.

Identification of the MGS, finding markers and validating markers is also

shown. e, MGS enriched in Chinese patients with liver cirrhosis and healthy

individuals. Species-level assignment was deduced from the best BlastN hits of

genes from a given MGS at thresholds of the average of more than 95% identity

and more than 90% overlap with genes from a sequenced genome. For MGS

where these thresholds were not reached, an assignment was attributed at the

lowest taxonomy level where at least 80% of the genes had the same best hit

BlastP taxonomy; in all cases these criteria held true at higher taxonomic levels.

f, Taxonomic assignments of 58 species related to gut gene richness in a

Danish cohort

RESEARCH ARTICLE

Extended Data Figure 10

Phylogenetic abundance of healthy controls in

the discovery stage in the liver cirrhosis and T2D studies. The relative

abundance of top bacterial phylotypes at the phylum, genus and species levels,

respectively, in the livercirrhosis study (top three panels) and in the T2D study

(bottom three panels).

ARTICLE RESEARCH

MSFT-transformer: a multistage fusion tabular transformer for disease prediction using metagenomic data

Article

Full-text available

May 2025
BRIEF BIOINFORM

More and more recent studies highlight the crucial role of the human microbiome in maintaining health, while modern advancements in metagenomic sequencing technologies have been accumulating data that are associated with human diseases. Although metagenomic data offer rich, multifaceted information, including taxonomic and functional abundance profiles, their full potential remains underutilized, as most approaches rely only on one type of information to discover and understand their related correlations with respect to disease occurrences. To address this limitation, we propose a multistage fusion tabular transformer architecture (MSFT-Transformer), aiming to effectively integrate various types of high-dimensional tabular information extracted from metagenomic data. Its multistage fusion strategy consists of three modules: a fusion-aware feature extraction module in the early stage to improve the extracted information from inputs, an alignment-enhanced fusion module in the mid stage to enforce the retainment of desired information in cross-modal learning, and an integrated feature decision layer in the late stage to incorporate desired cross-modal information. We conduct extensive experiments to evaluate the performance of MSFT-Transformer over state-of-the-art models on five standard datasets. Our results indicate that MSFT-Transformer provides stable performance gains with reduced computational costs. An ablation study illustrates the contributions of all three models compared with a reference multistage fusion transformer without these novel strategies. The result analysis implies the significant potential of the proposed model in future disease prediction with metagenomic data.

Study of gut microbiota in cholangiocarcinoma patients

Article

Apr 2025

Aim . To analyze the taxonomic composition of the intestinal microbiota in patients with cholangiocarcinoma (CCA) and compare it to individuals without oncopathology. Materials and methods . The study included patients with histologically verified cholangiocarcinoma (n = 30) and a control group (n = 27). An integrated approach was used, including clinical and anamnestic, laboratory, and instrumental methods. The intestinal microbiota was studied through amplicon sequencing of the bacterial 16S rRNA gene. Results . The assessment of alpha- and beta-diversity of the microbiota in patients with CCA did not show any significant differences compared to the control group. However, a comparative analysis revealed changes in the representation of a number of microorganisms at different taxonomic levels, including a higher content of Bacteroides and Lachnospiraceae_NK4A136_group in patients with CCA. Additionally, bacteria that influence the change in the global balance of microorganisms were identified in both groups, such as [Ruminococcus]_torques_group , Subdoligranulum , Parasutterella , unclassified Firmicutes in samples of patients with CCA and Oscillospiraceae and Erysipelotrichaceae UCG-006 in the control group. Conclusion . The study found a number of significant differences in bacterial representation between patients with cholangiocarcinoma and control group participants. Further research on the intestinal microbiota has the potential to develop non-invasive tools for early diagnosis of CCA.

Whole genome comparisons reveal gut-to-lung translocation of Escherichia coli and Burkholderia cenocepacia in two cases of ventilator-associated pneumonia in ICU patients

Article

Full-text available

May 2025
RESP RES

Background Identifying the sources of pathogenic bacteria causing ventilator-associated pneumonia (VAP) in intensive care unit (ICU) patients is crucial for developing effective prevention and treatment strategies. However, the scarcity of reported cases with confirmed sources limits the ability to evaluate and manage VAP, which remains a major challenge for healthcare systems globally. Methods Pathogens were isolated from endotracheal aspirate (ETA) samples of VAP patients using conventional culture techniques. Whole-genome comparisons, based on average nucleotide identity (ANI), were performed to identify genetically identical strains by comparing pulmonary isolate genomes with gut metagenome-derived bacterial genomes. Mouse models of pneumonia and colitis were used to validate the translocation of pathogenic bacteria from the gut to the lungs. Metagenomic analysis was performed to characterize the gut microbiome and resistome. Results Pathogenic isolates were obtained from the ETA samples of seven VAP patients, with one isolate per sample. Among these, Escherichia coli (Ec1) and Burkholderia cenocepacia (Bc1) from two patients were genetically identical to strains in their respective gut microbiota, with ANI values above 99%, indicating gut-to-lung translocation. The Ec1 strain demonstrated increased resistance to cefazolin while remaining susceptible to gentamicin, amikacin, and kanamycin, compared to previously reported pneumonia-associated E. coli strains. The Bc1 strain showed elevated resistance to macrolides, chloramphenicols, and tetracyclines relative to pneumonia-associated B. cenocepacia strains. Metagenomic analysis revealed a highly individualized gut microbiota composition among VAP patients. Notably, the translocated bacteria were not dominant within their gut microbiota. Additionally, these patients showed a marked increase in the total abundance of antibiotic resistance genes (ARGs) in their gut microbiota. The translocation ability of the Ec1 strain was validated in a mouse pneumonia model, where it caused more severe lung damage. Furthermore, elevated levels of Escherichia-Shigella were detected in the lung tissues of colitis mice, suggesting that gut-to-lung bacterial translocation may occur in a severely inflamed host, potentially leading to pneumonia. Conclusions This study demonstrates the gut-to-lung translocation of E. coli and B. cenocepacia, highlighting their role in the development and progression of VAP in ICU patients. These findings provide valuable insights for implementing targeted prevention and treatment strategies for VAP in ICU settings. Graphical abstract

Thomasclavelia ramosa and alcohol-related hepatocellular carcinoma: a microbial culturomics study

Article

Full-text available

May 2025

Background Gut microbiota alteration is implicated in the pathogenesis of alcoholic liver disease (ALD) and associated hepatocellular carcinoma (HCC). No study has characterized the dysbiosis associated with ALD by microbial culturomics, which certifies viability and allows pathobiont strain candidates to be characterized. Methods A case-control study (n = 59) was conducted on patients with ALD without HCC (ALD-NoHCC, n = 16), ALD with HCC (ALD-HCC, n = 19) and controls (n = 24) groups. 16 S rRNA amplicon sequencing and microbial culturomics were used as complementary methods for gut microbiome profiling. Results Compared to the control group, Thomasclavelia ramosa and Gemmiger formicilis were significantly increased in the ALD-HCC group and Mediterraneibacter gnavus was significantly increased in the ALD-NoHCC group using 16 S rRNA sequencing. By microbial culturomics, T. ramosa was detected in all ALD samples (100%), and the most enriched since cultivated in only a small proportion of controls (20%, p < 0.001). Conclusions T. ramosa, identified by culturomics and 16 rRNA sequencing, may be associated with ALD and ALD-HCC. These results highlight the potential role of T. ramosa in liver cancer, in line with its genotoxic properties and its tumor growth-promoting effect in gnotobiotic mice recently reported.

Gut Microbiota as a Mediator Between Intestinal Fibrosis and Creeping Fat in Crohn's Disease

Article

Full-text available

May 2025

Intestinal stricture remains one of the most challenging complications in Crohn's disease, and its underlying mechanisms are poorly understood. Accumulating evidence suggests that gut microbiota is significantly altered in stenotic intestines and may play a key role in the development of fibrogenesis in Crohn's disease. Additionally, the presence of hypertrophic mesenteric adipose tissue, also known as creeping fat, is closely correlated with intestinal stricture and fibrosis. Recent findings have revealed that bacterial translocation to creeping fat might exacerbate colitis and promote intestinal fibrosis. However, there is still a gap in determining whether gut microbiota links the formation of creeping fat to intestinal fibrosis. Hence, this review aims to summarize the known microbial influences on intestinal fibrosis, describes the microbial characteristics of creeping fat in Crohn's disease, and discusses the crosstalk between creeping fat‐associated dysbiosis and the development of intestinal fibrosis.

Quantification of Streptococcus salivarius using the digital polymerase chain reaction as a liver fibrosis marker

Article

Full-text available

Apr 2025

BACKGROUND The Streptococcus salivarius (S. salivarius ) group, which produces the enzyme urease has been identified as a potential contributor to ammonia production in the gut. Researchers have reported that patients with minimal HE had an increased abundance of the S. salivarius group, which is a specific change in the gut microbiota that distinguishes them from healthy individuals. The correlation between the aggregation of specific bacterial species and fibrosis progression in chronic liver disease (CLD) is yet to be fully elucidated. AIM To quantify S. salivarius using digital PCR (dPCR) as a liver fibrosis marker of CLD. METHODS This study retrospectively analysed 52 patients with CLD. To quantify S. salivarius in patients with CLD using dPCR, we evaluated the specificity and sensitivity of S. salivarius bacterial load using dPCR for a type strain. Next, we evaluated the clinical usefulness of dPCR for S. salivarius load quantification for detecting liver fibrosis in patients with CLD. The liver fibrosis stage was categorized into mild and advanced fibrosis based on pathological findings. RESULTS The dPCR assay revealed that S. salivarius was highly positive for the tnpA gene. The lower limit of quantification for dPCR using the tnpA gene with a 1 μL template comprising 1.28 × 102 CFU/mL was 4.3 copies. After considering the detection range in dPCR, we adjusted the extracted DNA concentration to 5.0 × 10-4 ng/μL from 200 mg stool samples. The median bacterial loads of S. salivarius in stool sample from patients with mild and advanced fibrosis were 1.9 and 7.4 copies/μL, respectively. The quantification of S. salivarius load was observed more frequently in patients with advanced fibrosis than in those with mild fibrosis (P = 0.032). CONCLUSION Quantifying of S. salivarius load using digital PCR is a useful biomarker for liver fibrosis in patients with CLD.

Gut microbiota and associated metabolites: key players in high-fat diet-induced chronic diseases

Article

Full-text available

Apr 2025

Excessive intake of dietary fats is strongly associated with an increased risk of various chronic diseases, such as obesity, diabetes, hepatic metabolic disorders, cardiovascular disease, chronic intestinal inflammation, and certain cancers. A significant portion of the adverse effects of high-fat diet on disease risk is mediated through modifications in the gut microbiota. Specifically, high-fat diets are linked to reduced microbial diversity, an overgrowth of gram-negative bacteria, an elevated Firmicutes-to-Bacteroidetes ratio, and alterations at various taxonomic levels. These microbial alterations influence the intestinal metabolism of small molecules, which subsequently increases intestinal permeability, exacerbates inflammatory responses, disrupts metabolic functions, and impairs signal transduction pathways in the host. Consequently, diet-induced changes in the gut microbiota play a crucial role in the initiation and progression of chronic diseases. This review explores the relationship between high-fat diets and gut microbiota, highlighting their roles and underlying mechanisms in the development of chronic metabolic diseases. Additionally, we propose probiotic interventions may serve as a promising adjunctive therapy to counteract the negative effects of high-fat diet-induced alterations in gut microbiota composition.

Hepatic encephalopathy in cirrhosis: therapies and developments

Article

Full-text available

May 2025
METAB BRAIN DIS

Hepatic encephalopathy (HE) is a frequent decompensation in patients with cirrhosis, which significantly affects morbidity and mortality. Ammonia is a major neurotoxin implicated in the pathogenesis, progression, and severity of HE, and various organs including the gut, muscle, kidney, and brain are involved in its metabolism. Therefore, therapeutic management involves reducing ammonia production and increasing its elimination from the blood and the brain. Prevention of HE in patients at high risk of first and recurrent episodes is important for prolonging survival. Various anti-ammonia therapies with synergistic and complementary actions have been attempted for overt HE and for prophylaxis of the first and recurrent episodes of HE. In the current review, we summarize the currently used and under-development pharmacotherapies/procedure(s) for HE in cirrhosis and their mechanism of action. Primary and secondary prophylaxis with monotherapies and combination therapies are also discussed.

Global genetic structure of human gut microbiome species is related to geographic location and host health

Article

Full-text available

Apr 2025

Small amounts of misassembly can have disproportionate effects on pangenome-based metagenomic analyses

Article

Apr 2025

Individual genes from microbiomes can drive host-level phenotypes. To help identify such candidate genes, several recent tools estimate microbial gene copy numbers directly from metagenomes. These tools rely on alignments to pangenomes, which, in turn, are derived from the set of all individual genomes from one species. While large-scale metagenomic assembly efforts have made pangenome estimates more complete, mixed communities can also introduce contamination into assemblies, and it is unknown how robust pangenome-based metagenomic analyses are to these errors. To gain insight into this problem, we re-analyzed a case-control study of the gut microbiome in cirrhosis, focusing on commensal Clostridia previously implicated in this disease. We tested for differentially prevalent genes in the Lachnospiraceae and then investigated which were likely to be contaminants using sequence similarity searches. Out of 86 differentially prevalent genes, we found that 33 (38%) were probably contaminants originating in taxa such as Veillonella and Haemophilus , unrelated genera that were independently correlated with disease status. Our results demonstrate that even small amounts of contamination in metagenome assemblies, below typical quality thresholds, can threaten to overwhelm gene-level metagenomic analyses. However, we also show that such contaminants can be accurately identified using a method based on gene-to-species correlation. After removing these contaminants, we observe that several flagellar motility gene clusters in the Lachnospira eligens pangenome are associated with cirrhosis status. We have integrated our analyses into an analysis and visualization pipeline, PanSweep, that can automatically identify cases where pangenome contamination may bias the results of gene-resolved analyses. IMPORTANCE Metagenome-assembled genomes, or MAGs, can be constructed without pure cultures of microbes. Large-scale efforts to build MAGs have yielded more complete pangenomes (i.e., sets of all genes found in one species). Pangenomes allow us to measure strain variation in gene content, which can strongly affect phenotype. However, because MAGs come from mixed communities, they can contaminate pangenomes with unrelated DNA; how much this impacts downstream analyses has not been studied. Using a metagenomic study of gut microbes in cirrhosis as our test case, we investigate how contamination affects analyses of microbial gene content. Surprisingly, even small, typical amounts of MAG contamination (<5%) result in disproportionately high levels of false positive associations (38%). Fortunately, we show that most contaminants can be automatically flagged and provide a simple method for doing so. Furthermore, applying this method reveals a new association between cirrhosis and gut microbial motility.

The Human Oral Microbiome Database: a web accessible resource for investigating oral microbe taxonomic and genomic information

Article

Full-text available

Jun 2010

The human oral microbiome is the most studied human microflora, but 53% of the species have not yet been validly named and 35% remain uncultivated. The uncultivated taxa are known primarily from 16S rRNA sequence information. Sequence information tied solely to obscure isolate or clone numbers, and usually lacking accurate phylogenetic placement, is a major impediment to working with human oral microbiome data. The goal of creating the Human Oral Microbiome Database (HOMD) is to provide the scientific community with a body site-specific comprehensive database for the more than 600 prokaryote species that are present in the human oral cavity based on a curated 16S rRNA gene-based provisional naming scheme. Currently, two primary types of information are provided in HOMD—taxonomic and genomic. Named oral species and taxa identified from 16S rRNA gene sequence analysis of oral isolates and cloning studies were placed into defined 16S rRNA phylotypes and each given unique Human Oral Taxon (HOT) number. The HOT interlinks phenotypic, phylogenetic, genomic, clinical and bibliographic information for each taxon. A BLAST search tool is provided to match user 16S rRNA gene sequences to a curated, full length, 16S rRNA gene reference data set. For genomic analysis, HOMD provides comprehensive set of analysis tools and maintains frequently updated annotations for all the human oral microbial genomes that have been sequenced and publicly released. Oral bacterial genome sequences, determined as part of the Human Microbiome Project, are being added to the HOMD as they become available. We provide HOMD as a conceptual model for the presentation of microbiome data for other human body sites.

Corrigendum: Dietary intervention impact on gut microbial gene richness

Article

Full-text available

Oct 2013

Complex gene-environment interactions are considered important in the development of obesity(1). The composition of the gut microbiota can determine the efficacy of energy harvest from food(2-4) and changes in dietary composition have been associated with changes in the composition of gut microbial populations(5,6). The capacity to explore microbiota composition was markedly improved by the development of metagenomic approaches(7,8), which have already allowed production of the first human gut microbial gene catalogue(9) and stratifying individuals by their gut genomic profile into different enterotypes(10), but the analyses were carried out mainly in nonintervention settings. To investigate the temporal relationships between food intake, gut microbiota and metabolic and inflammatory phenotypes, we conducted diet-induced weight-loss and weight-stabilization interventions in a study sample of 38 obese and 11 overweight individuals. Here we report that individuals with reduced microbial gene richness (40%) present more pronounced dys-metabolism and low-grade inflammation, as observed concomitantly in the accompanying paper(11). Dietary intervention improves low gene richness and clinical phenotypes, but seems to be less efficient for inflammation variables in individuals with lower gene richness. Low gene richness may therefore have predictive potential for the efficacy of intervention.

Identification and assembly of genomes and genetic elements in complex metagenomic samples without using reference genomes

Article

Full-text available

Aug 2014
NAT BIOTECHNOL

Most current approaches for analyzing metagenomic data rely on comparisons to reference genomes, but the microbial diversity of many environments extends far beyond what is covered by reference databases. De novo segregation of complex metagenomic data into specific biological entities, such as particular bacterial strains or viruses, remains a largely unsolved problem. Here we present a method, based on binning co-abundant genes across a series of metagenomic samples, that enables comprehensive discovery of new microbial organisms, viruses and co-inherited genetic entities and aids assembly of microbial genomes without the need for reference sequences. We demonstrate the method on data from 396 human gut microbiome samples and identify 7,381 co-abundance gene groups (CAGs), including 741 metagenomic species (MGS). We use these to assemble 238 high-quality microbial genomes and identify affiliations between MGS and hundreds of viruses or genetic entities. Our method provides the means for comprehensive profiling of the diversity within complex metagenomic samples.

Richness of human gut microbiome correlates with metabolic markers

Article

Full-text available

Aug 2013
NATURE

We are facing a global metabolic health crisis provoked by an obesity epidemic. Here we report the human gut microbial composition in a population sample of 123 non-obese and 169 obese Danish individuals. We find two groups of individuals that differ by the number of gut microbial genes and thus gut bacterial richness. They contain known and previously unknown bacterial species at different proportions; individuals with a low bacterial richness (23% of the population) are characterized by more marked overall adiposity, insulin resistance and dyslipidaemia and a more pronounced inflammatory phenotype when compared with high bacterial richness individuals. The obese individuals among the lower bacterial richness group also gain more weight over time. Only a few bacterial species are sufficient to distinguish between individuals with high and low bacterial richness, and even between lean and obese participants. Our classifications based on variation in the gut microbiome identify subsets of individuals in the general white adult population who may be at increased risk of progressing to adiposity-associated co-morbidities.

Dietary intervention impact on gut microbial gene richness

Article

Full-text available

Aug 2013

Complex gene–environment interactions are considered important in the development of obesity. The composition of the gut microbiota can determine the efficacy of energy harvest from food and changes in dietary composition have been associated with changes in the composition of gut microbial populations. The capacity to explore microbiota composition was markedly improved by the development of metagenomic approaches, which have already allowed production of the first human gut microbial gene catalogue and stratifying individuals by their gut genomic profile into different enterotypes, but the analyses were carried out mainly in non-intervention settings. To investigate the temporal relationships between food intake, gut microbiota and metabolic and inflammatory phenotypes, we conducted diet-induced weight-loss and weight-stabilization interventions in a study sample of 38 obese and 11 overweight individuals. Here we report that individuals with reduced microbial gene richness (40%) present more pronounced dys-metabolism and low-grade inflammation, as observed concomitantly in the accompanying paper. Dietary intervention improves low gene richness and clinical phenotypes, but seems to be less efficient for inflammation variables in individuals with lower gene richness. Low gene richness may therefore have predictive potential for the efficacy of intervention.

Genome sequences of rare, uncultured bacteria obtained by differential coverage binning of multiple metagenomes

Article

Full-text available

May 2013
NAT BIOTECHNOL

Reference genomes are required to understand the diverse roles of microorganisms in ecology, evolution, human and animal health, but most species remain uncultured. Here we present a sequence composition-independent approach to recover high-quality microbial genomes from deeply sequenced metagenomes. Multiple metagenomes of the same community, which differ in relative population abundances, were used to assemble 31 bacterial genomes, including rare (<1% relative abundance) species, from an activated sludge bioreactor. Twelve genomes were assembled into complete or near-complete chromosomes. Four belong to the candidate bacterial phylum TM7 and represent the most complete genomes for this phylum to date (relative abundances, 0.06-1.58%). Reanalysis of published metagenomes reveals that differential coverage binning facilitates recovery of more complete and higher fidelity genome bins than other currently used methods, which are primarily based on sequence composition. This approach will be an important addition to the standard metagenome toolbox and greatly improve access to genomes of uncultured microorganisms.

Minimum redundancy feature selection from microarray gene expression data

Article

Jan 2003

The role of manganese in chronic hepatic encephalopathy

Article

Oct 1995

D Krieger

Small intestinal bacterial overgrowth in human cirrhosis is associated with systemic endotoxemia

Article

Sep 2002

Tilman M. Bauer

OBJECTIVES:Systemic endotoxemia has been implicated in various pathophysiological sequelae of chronic liver disease. One of its potential causes is increased intestinal absorption of endotoxin. We therefore examined the association of small intestinal bacterial overgrowth with systemic endotoxemia in patients with cirrhosis.METHODS:Fifty-three consecutive patients with cirrhosis (Child-Pugh group A, 23; group B, 18; group C, 12) were included. Jejunal secretions were cultivated quantitatively and systemic endotoxemia determined by the chromogenic Limulus amoebocyte assay. Patients were followed up for 1 yr.RESULTS:Small intestinal bacterial overgrowth, defined as ≥105 total colony forming units per milliliter of jejunal secretions, was present in 59% of patients and strongly associated with acid suppressive therapy. The mean plasma endotoxin level was 0.86 ± 0.48 endotoxin units/ml (range = 0.03–1.44) and was significantly associated with small intestinal bacterial overgrowth (0.99 vs 0.60 endotoxin units/ml, p = 0.03). During the 1-yr follow-up, seven patients were lost to follow up or underwent liver transplantation and 12 patients died. Multivariate Cox regression showed Child-Pugh group to be the only predictor for survival.CONCLUSIONS:Small intestinal bacterial overgrowth in cirrhotic patients is common and associated with systemic endotoxemia. The clinical relevance of this association remains to be defined.

A core gut microbiom in obese and lean twins.

Article