1、REPORTSMICROBIOMEPopulation-based metagenomicsanalysis reveals markers for gutmicrobiome composition and diversityAlexandra Zhernakova,1,2* Alexander Kurilshikov,3,4 Marc Jan Bonder,1Ettje F. Tigchelaar,1,2 Melanie Schirmer,5,6Tommi Vatanen,5,7Zlatan Mujagic,2,8Arnau Vich Vila,9Gwen Falony,10,11Sara
2、 Vieira-Silva,10,11Jun Wang,10,11Floris Imhann,9Eelke Brandsma,12Soesma A. Jankipersadsing,1Marie Joossens,10,11,13Maria Carmen Cenit,1,14,15Patrick Deelen,1,16Morris A. Swertz,1,16LifeLines cohort study, Rinse K. Weersma,9Edith J. M. Feskens,2,17Mihai G. Netea,18Dirk Gevers,5 Daisy Jonkers,8Lude Fr
3、anke,1Yurii S. Aulchenko,4,19,20,21Curtis Huttenhower,5,6Jeroen Raes,10,11,13Marten H. Hofker,12Ramnik J. Xavier,5,22,23,24Cisca Wijmenga,1* Jingyuan Fu1,12*Deep sequencing of the gut microbiomes of 1135 participants from a Dutch population-basedcohort shows relations between the microbiome and 126
4、exogenous and intrinsic host factors,including 31 intrinsic factors, 12 diseases, 19 drug groups, 4 smoking categories, and 60dietary factors.These factors collectively explain 18.7% of the variation seen in theinterindividual distance of microbial composition. We could associate 110 factors to 125s
5、pecies and observed that fecal chromogranin A (CgA), a protein secreted by enteroendocrinecells, was exclusively associated with 61 microbial species whose abundance collectivelyaccounted for 53% of microbial composition. Low CgA concentrations were seen in individualswith a more diverse microbiome.
6、These results are an important step toward a betterunderstanding of environment-diet-microbe-host interactions.The human gut microbiome plays a majorrole in the production of vitamins, enzymes,and other compounds that digest and me-tabolize food and regulate our immune sys-tem (1). It can be conside
7、red as an extraorgan, with remarkable dynamics and a majorimpact on our physiology. The composition of thegut microbiome can be considered as a complextrait, with the quantitative variation in the micro-biome affected by a large number of host andenvironmental factors, each of which may haveonly a s
8、mall additive effect, making it difficult toidentifytheassociationforeachseparateitem.Inthis study, we present a systematic metagenomicassociation analysis of 207 intrinsic and exoge-nous factors from the LifeLines-DEEP cohort, aDutch populationbased study (2, 3). Our studyreveals covariates in the
9、microbiome and, moreimportantly, provides a list of factors that corre-late with shifts in the microbiome compositionand functionality.This study includes stool samples from 1179LifeLines-DEEP participants from the generalpopulation of the northern part of the Netherlands(2). The cohort comprised pr
10、edominantly Dutchparticipants; 93.7% had both parents born in theNetherlands. The gut microbiome was analyzedwithpaired-endmetagenomicshotgunsequencing(MGS)onaHiSeq2000,generatinganaverageof3.0 Gb of data (about 32.3 million reads) per sam-ple (4). After excluding 44 samples with low readcounts, 113
11、5 participants (474 males and 661 fe-males) remained for further analysis. We tested207 factors with respect to the microbiomes ofthese participants: 41 intrinsic factors of variousphysiological and biomedical measures, 39 self-reported diseases, 44 categories of drugs, 5 cate-gories of smoking stat
12、us, and 78 dietary factors(fig. S1 and table S1). These factors cover dietaryhabits, lifestyle, medication use, and health pa-rameters. Most of the factors showed a low ormodest intercorrelation (table S2, A to C, and fig.S2, A to D); many are highly variable, including,as expected in the Dutch popu
13、lation, the high con-sumption of milk products and low use of anti-biotics.AntibioticuseintheNetherlandsisthelowest in Europe, at a level half that of the UKand one-third that of Belgium. To cover health-domain factors relevant to the host immune sys-tem and gut health, we collected cell counts fore
14、ight different blood cell types, measured bloodcytokine concentrations, assessed stool frequencyandstooltypebyBristolstoolscore,andmeasuredfecallevelsofseveralsecretedproteins,includingcalprotectin as a marker for the immune systemactivation,humanb-defensin-2(HBD-2)asamark-er for defense against inv
15、ading microbes, andchromogranin A (CgA) as a marker for neuro-endocrine system activation.After quality control and removal of sequencereadsmappingto thehuman genome,themicro-biome sequence reads were mapped to 1 millionmicrobial-taxonomyspecific marker genes withMetaPhlAn 2.0 (5) to predict the abu
16、ndance ofmicroorganisms (fig. S3A). For each participant,we predicted the abundances for 1649 microbialtaxonomic clades ranging from four different do-mains to 632 species (Fig. 1A). Most of the reads(97.6%) came from Bacteria; 2.2% were fromArchaea, 0.2% from Viruses, and 0.01% of total microbial c
17、omposition and pres-ent in at least 10 individuals) and 215 MetaCycpathways(fig.S3C). When corrected for age, gen-der, and sequence depth, we found 485 associa-tions at FDR=0.3Color key for correlationSmokingDietsmk_historysmk_currentsmk_fathersmk_mother0 0.001 0.002carbohydrates.totalprotein.planth
18、ow_often_fruitsbeerbreadskcalsoda_with_sugarcoffeenonalc_drinkshow_often_coffeehow_often_sodahow_often_vegetablesred_winefruitspastrysavoury_snackslow_carb_dietprotein.animalhow_often_chocomilk_sweetened_milk_drinkspastahow_often_pastaprotein.totalcerealsricehow_often_crisps_savory_crackershow_often
19、_alcoholalcohol_productssaucesfat.totalready_mealmeathow_often_mueslivegetableswholefat_milkhow_often_ricedairyhow_often_breakfasthow_often_yoghurt_milk_based_puddingscheesebuttermilkvegetarianteagluten_free_dietsoda_no_sugarhow_often_pulseshow_often_teahalffat_milkalcohol.gweight_related_diethow_of
20、ten_meathow_often_nutshow_often_juicehow_often_milk_or_sourmilkspreadshow_often_breadhow_often_boiled_potatoshow_often_fishlegumeseggspotatosShannons indexGene richnessCOG richness0 0.002 0.004Fig. 3. Factors associated with interindividual variation of gut microbiome. A total of 126 factors(FDR0.1)
21、 were associated with interindividual variation of the gut microbiome.The bar plot indicates theexplained variation of each factor in the interindividual variation of microbial composition Bray-Curtis (BC)distance.The heatmap next to the bar plot shows the correlation coefficients ofeach factor with
22、 Shannonsindex of diversity, gene richness, and COG richness, respectively. Color key for correlation is shown.RESEARCH | REPORTSon April 28, 2016http:/science.sciencemag.org/Downloaded from carbohydrates in the diet was associated withlower microbiome diversity. Total carbohydrateintake was positiv
23、ely associated with Bifidobac-teria but negatively associatedwith Lactobacillus,Streptococcus,andRoseburia species. A low-carbohydrate diet consistently showed oppositedirectionsofassociationforthesespecies.Wedidnot observe an association of carbohydrate in-take to prevotella species, as has been de
24、scribedpreviously (24).As expected, the use of antibiotics was signif-icantlyassociatedwithmicrobiomecomposition,in particular with strong and significant decreasesin two species from the genus Bifidobacterium(Actinobacteria phylum) (table S11), in line withprevious studies (25). Several other drug
25、catego-ries, such as proton-pump inhibitors (PPIs) (95users), metformin (15 users), statins (56 users), andlaxatives (21 users), also had a strong effect onthe gut microbiome. PPI users were found to haveprofound changes in 33 bacterial pathways (tableS12). The most significant positive correlationo
26、f PPIs was observed with the pathway of 2,3-butanediol biosynthesis (q =5.31014). We alsoobserved overlap between species and pathwaysassociated to PPI and with calprotectin levels, par-ticularly for bacteria typical of the oral microbiome(tableS2,AtoC;tableS11;andfig.S12).Thisisinlinewiththecorrela
27、tionsofPPIswithcalprotectinlevels reported in the literature (26). Even after ex-cluding the 95 PPI users from our analysis, thepositive correlation ofcalprotectin tomostoralbacteria remained significant, indicating that thisassociation is not due to the confounding effectof PPIs (fig. S12). Further
28、more, the amounts ofcalprotectinwerepositivelycorrelatedwithageandmetabolic phenotypes body mass index (BMI),diabetes, use of statins and metformin, glycatedhemoglobin (HbA1c), and systolic blood pressure,but negatively correlated with the consumptionofvegetables,plantproteins,chocolate,andbreads.Mu
29、ltivariate analysis correcting for all factorsrevealed 14 species (table S13) and 114bacterialmetabolic pathways (table S14) exclusively associ-ated with calprotectin, suggesting that calprotectinis robustly associated with the gut microbiome.Metformin is commonly used to control bloodsugar concentr
30、ationsfor treating type 2diabetes,but can cause gastrointestinal intolerance (27). In15 metformin users, we observed an increasedabundance of Escherichia coli and a positive cor-relation with specific pathways, including thedegradation and utilization of D-glucarate and D-galactarate and pyruvate fe
31、rmentation pathways.Previous studies in Caenorhabditis elegans indi-catedthespecificdrug-bacteriainteractionofmet-forminandE.coli(28).Ourresultsareinlinewithrecent observations in humans (29) that suggestthatmetformincanaffectthemicrobiomethroughshort-chainfattyacid(SCFA)production.Tocon-firm this o
32、bservation, we profiled acetate, propi-onate,andbutyratein24 type2diabetespatientsin our cohort9 nonmetformin users and 15 users(4)and found that SCFA concentrations wereconsistently higher in metformin users, especiallyforpropionate (Wilcoxon test, P=0.035)(fig.S13).Weassessedtheeffectofcurrentsmok
33、ingstatus,smokinghistory,parentalsmoking,andmaternalsmokingduringpregnancyonthegutmicrobiome.TheseparameterswereassociatedwithBray-Curtisdistance, albeit with very modest effect. We didnot detect significant associations for individualspeciesoratpathways.Inthisstudy,weincluded39 self-reported diseas
34、es, for which participantshad reported at least five cases.IBS was reportedby9.9%ofparticipants(n=112,tableS1)andwasassociated with changes in the gut microbiomeandalowermicrobialdiversity(adjustedP=0.05)(table S6). Species from the Eggerthella and Copro-bacillus genera were positively associated wi
35、thmedication and food allergies, respectively. Indi-viduals who had suffered a heart attack (n =10)in the past had a significantly lower abundanceof Eubacterium eligens bacterium, even after cor-recting for all other factors (q =4.6104).Linking the deep-sequenced MGS data to var-ious intrinsic and e
36、xogenous factors from thesame individual not only allowed us to detect as-sociations at species level, but also provided newinsights into the interaction between the host, mi-crobiota,andenvironmentalfactors,includingdiet.Forinstance,wehavereplicatedandexpandedourassociation of BMI and blood lipid c
37、oncentrationswith the gut microbiota based on 16SrRNA genesequencingdata(30)byshowingassociationswithfour specific species of the family Rikenellaceae.WepreviouslyassociatedthisfamilywithBMIandtriglycerides in 16S rRNA data. In this study, weobserved that a higher BMI was associated with568 29 APRIL
38、 2016 VOL 352 ISSUE 6285 sciencemag.org SCIENCEFig. 4.The association of fecal level of chromogranin A. (A) Principal coordinate plots of Bray-Curtisdistance of microbial composition. Each dot represents one individual, and its color is based on theabundancelevelofCgA:Warmcolorsindicatehighabundance
39、andcoolcolors,lowabundance.Theredarrow indicates the association direction of CgA, while the directions of the CgA-associated phyla areshownasblackarrows.(B)CorrelationbetweenCgAandotherfactorsatFDR0.1.(C)Taxonomictreeof 170 species, of which 61 species were exclusively associated with CgA level. Ea
40、ch dot represents ataxonomic entity. Red dots indicate positively associated species. Blue dots indicate negatively as-sociated species. (D) Taxonomic tree of the 61 species exclusively associated with CgA level. Thebranches are colored to show phylum levels as shown in the color key. Species in red
41、 show increasedabundance associated with higher CgA levels. Species in blue show lower abundance associated withhigher CgA levels.RESEARCH | REPORTSon April 28, 2016http:/science.sciencemag.org/Downloaded from a lower abundance of two species from the familyRikenellaceae, Alistipes finegoldii,andAli
42、stipessenegalensis, whereas blood lipids were associatedwithtwootherspecies,Alistipes shahii and Alistipesputredinis (table S11). Notably, these species werealsoassociatedtocertaindietaryfactorsanddrugs.For instance, a high level of A. shahii, which wasassociatedtolowtriglyceride(TG)levels,waslinked
43、to higher fruit intake (q = 0.00027). Individualswithahigherabundance ofA.shahiihadahighernumber of different species in the gut (species rich-ness) (Spearman r= 0.2,adjusted P=3.96x1011),suggesting a beneficial effect on the microbialecosystem (table S18). Correlations with the num-ber of different
44、 species were also found for otherbacteria, including Roseburia hominis,Coprococ-cus catus,andBarnesiella intestinihominis andunclassified species from genus Anaerotruncusthat also showed correlation both with fruit, veg-etable, and nut consumption and with intrinsicphenotypeslikeHDL,triglycerides,
45、andqualityof life. On the basis of these data, it would beinteresting to explore the potential to modulatedisease-associated species through medicationor diet, although we still need to address thecausality and underlying mechanism.Our study revealed significant associations be-tween the gut microbi
46、ome and various intrinsic,environmental, dietary and medication parame-ters, and diseasephenotypes, witha high replica-tion rate between MGS and 16S rRNA genesequencingdatafromthesameindividuals.More-over, our study providesmanynewintrinsicandexogenous factors that correlate with shifts inthe microb
47、iome composition and functionalitythat potentially can be manipulated to improvemicrobiome-related health, and we hope our re-sults will inspire further experiments to explorethe biological relevance of associated factors. Al-thoughmostofthefactorsthatweassessedexerteda very modest effect, fecal lev
48、els of CgA showed ahigh potential as a biomarker for gut health.REFERENCES AND NOTES1. J. C. Clemente, L. K. Ursell, L. W. Parfrey, R. Knight, Cell 148,12581270 (2012).2. E. F. Tigchelaar et al., BMJ Open 5, e006772 (2015).3. S. Scholtens et al., Int. J. Epidemiol. 44, 11721180 (2015).4. Information
49、 on materials and methods is available at ScienceOnline.5. N. Segata et al., Nat. Methods 9, 811814 (2012).6. R. Sinha, C. C. Abnet, O. White, R. Knight, C. Huttenhower,Genome Biol. 16, 276 (2015).7. J. K. Goodrich et al., Cell 159, 789799 (2014).8. G. Falony et al., Science 352,560564 (2016).9. C. Huttenhower et al., Nature 486, 207214 (2012).10. F. Hildebrand et al., Genome Biol. 14, R4 (2013).11. C. Hedin et al., Gut 10.1136/gutjnl-2014-308896 (2015).12. T. L. Miller, M. J. Wolin, E. Conway de Macario, A. J. Macario,Appl. Environ. Microbiol. 43, 227232 (1982)
