RESEARCH ARTICLE

GENOMICS The Impact of a Consortium of Fermented Milk Strains on the Gut Microbiome of Gnotobiotic Mice and Monozygotic Twins

Nathan P. McNulty,1* Tanya Yatsunenko,1* Ansel Hsiao,1* Jeremiah J. Faith,1 Brian D. Muegge,1 Andrew L. Goodman,1† Bernard Henrissat,2 Raish Oozeer,3 Stéphanie Cools-Portier,3 Guillaume Gobert,3 Christian Chervaux,3 Dan Knights,4 Catherine A. Lozupone,5 Rob Knight,5,6 Alexis E. Duncan,7,8 James R. Bain,9,10 Michael J. Muehlbauer,9 Christopher B. Newgard,9,10,11 Andrew C. Heath,7 Jeffrey I. Gordon1‡

Understanding how the human gut microbiota and host are affected by probiotic bacterial strains requires carefully controlled studies in humans and in mouse models of the gut ecosystem where potentially confounding variables that are difficult to control in humans can be constrained. Therefore, we characterized the fecal microbiomes and Downloaded from metatranscriptomes of adult female monozygotic twin pairs through repeated sampling 4 weeks before, 7 weeks during, and 4 weeks after consumption of a commercially available fermented milk product (FMP) containing a consortium of Bifidobacterium animalis subsp. lactis, two strains of Lactobacillus delbrueckii subsp. bulgaricus, Lactococcus lactis subsp. cremoris,andStreptococcus thermophilus. In addition, gnotobiotic mice harboring a 15-species model human gut microbiota whose genomes contain 58,399 known or predicted protein-coding genes were studied before and after gavage with all five sequenced FMP strains. No significant changes in bacterial species composition or in the proportional representation of genes encoding known enzymes were observed in the feces of http://stm.sciencemag.org/ humans consuming the FMP. Only minimal changes in microbiota configuration were noted in mice after single or repeated gavage with the FMP consortium. However, RNA-Seq analysis of fecal samples and follow-up mass spec- trometry of urinary metabolites disclosed that introducing the FMP strains into mice results in significant changes in expression of microbiome-encoded enzymes involved in numerous metabolic pathways, most prominently those related to carbohydrate metabolism. B. animalis subsp. lactis, the dominant persistent member of the FMP consor- tium in gnotobiotic mice, up-regulates a locus in vivo that is involved in the catabolism of xylooligosaccharides, a class of glycans widely distributed in fruits, vegetables, and other foods, underscoring the importance of these sugars to this bacterial species. The human fecal metatranscriptome exhibited significant changes, con- fined to the period of FMP consumption, that mirror changes in gnotobiotic mice, including those related to plant polysaccharide metabolism. These experiments illustrate a translational research pipeline for character- on March 29, 2016 izing the effects of FMPs on the human gut microbiome.

INTRODUCTION and types of culture-independent metagenomic studies of intra- and Our physiology and physiological differences are not only manifesta- interpersonal variations in human microbial ecology—as a function of tions of our human genes and epigenomes but also a reflection of the our human life cycle, cultural traditions, and health status (2–7). Long- genes and genetic variations that exist in our resident microbial com- term goals of this quest to understand the genomic and metabolic munities (microbiomes). Our microbiomes contain at least 100 times underpinnings of our mutually beneficial relationships with microbes more genes than our human genomes (1). Marked increases in DNA include using our symbionts as a new class of biosensors and biomark- sequencing capacity have led to an explosive increase in the number ers of wellness, and devising safe and effective ways to deliberately manipulate the structure and functions of our microbiome to optimize 1Center for Genome Sciences and Systems Biology, Washington University School of our health, as well as to treat various diseases. Medicine, St. Louis, MO 63108, USA. 2Architecture et Fonction des Macromolécules A necessary starting point for assessing how the structure and Biologiques, CNRS, 13288 Marseille Cedex 9, France. 3Danone Research, 91 767 Palaiseau functions of the human microbiome are related to our biology is to 4 Cedex, France. Department of Computer Science, University of Colorado, Boulder, CO 80309, characterize the normal variations that occur in these communities, USA. 5Department of Chemistry and Biochemistry, University of Colorado, Boulder, CO 80309, USA. 6Howard Hughes Medical Institute, University of Colorado, Boulder, CO theirgenepools,andtheirgeneexpressionprofilesbothwithinand 80309, USA. 7Department of Psychiatry, Washington University School of Medicine, between individuals. This requires carefully designed studies where St. Louis, MO 63108, USA. 8George Warren Brown School of Social Work, Washington potentially confounding variables such as host genotype, diet, and var- University, St. Louis, MO 63130, USA. 9Sarah W. Stedman Nutrition and Metabolism Center, Duke University Medical Center, Durham, NC 27710, USA. 10Department of Medicine, Duke ious environmental exposures can be controlled and systematically University Medical Center, Durham, NC 27710, USA. 11Department of Pharmacology and manipulated. Monozygotic (MZ) twins represent one way to constrain Cancer Biology, Duke University Medical Center, Durham, NC 27710, USA. some of these variables, given that they have more similar genotypes *These authors contributed equally to this work. and have experienced more similar dietary and other early environ- †Present address: Section of Microbial Pathogenesis and Microbial Diversity Institute, Yale School of Medicine, New Haven, CT 06536, USA. mental exposures than any other combination of individuals. A com- ‡To whom correspondence should be addressed. E-mail: [email protected] plementary approach is to use germ-free mice colonized at various

www.ScienceTranslationalMedicine.org 26 October 2011 Vol 3 Issue 106 106ra106 1 RESEARCH ARTICLE points in their life with defined collections of microbes, with sequenced of colony-forming units (CFU) in a typical cup of the FMP [4.9 × genomes, that represent major phylogenetic lineages encountered in 107 CFU/g (B. animalis subsp. lactis), 8.4 × 107 CFU/g (L. delbrueckii the body habitats of human populations of interest. Gnotobiotic mice subsp. bulgaricus)]. harboring “synthetic” model human microbiomes, where all compo- Three fecal samples were obtained over the course of a 4-week nent organisms and microbial genes are known, can be reared under period before initiation of FMP consumption (“pretreatment phase”; conditions where a number of the variables that confound human see Fig. 1A). Each co-twin then consumed two servings of the FMP studies are extremely well controlled. Insights gleaned from these per day for 7 weeks (breakfast and dinner). Four fecal samples were gnotobiotic animals can be applied back to humans (8). collected at defined intervals during this treatment period, whereas Common intended or unintentional disturbances to our micro- two additional samples were collected during the 4 weeks after cessa- biomes include changes in our diets, consumption of antibiotics, and tion of FMP consumption (“posttreatment phase”;Fig.1A).Participants ingestion of live microbial strains posited to improve health. The latter kept a daily log of their FMP consumption and stool parameters in- include commercially available probiotics that are incorporated into cluding frequency. Statistical analyses of this log indicated that, in this fermented milk products (FMPs). With increasing regulatory pres- population, FMP consumption was associated with significantly softer sure to validate the composition and health claims of probiotics and stools but no significant changes in stool frequency (see the Supple- “functional” foods, it is particularly important to develop informative mentary Material). However, based on existing regulatory criteria, our translational medicine pipelines so that proof-of-concept clinical trials study of this small cohort was insufficiently powered to draw clinical can be performed with validated biomarkers for quantitative phenotyp- conclusions about these stool parameters. Moreover, the MZ twin pop- Downloaded from ing of subjects and of their responses. The present study demonstrates ulation recruited was composed entirely of healthy individuals, so one type of approach. It uses adult MZ twin pairs and metagenomic these data cannot be used to make statements about the impact of methods to first define temporal fluctuations in the organismal and FMP consumption on stool softness in unhealthy patient populations. gene content and gene expression profiles of their fecal microbial com- All fecal samples collected during the three phases of this study munities as a function of administration of a widely used commercial were frozen at −20°C within 30 min of their passage and maintained FMP. It then takes the five sequenced strains present in the FMP and at this temperature during overnight shipment to a biospecimen re- introduces them as a consortium, at a dose analogous to that experi- pository where they were subsequently stored at −80°C before meta- http://stm.sciencemag.org/ enced by humans, into gnotobiotic mice containing a model human genomic analyses. To assess intra- and interpersonal variations in microbiome composed of 15 sequenced human gut symbionts. Quan- microbial community structure, we performed multiplex 454 FLX pyro- titative analyses of temporal changes in the proportional representation sequencing of amplicons generated from variable region 2 of bacte- of microbial species and genes, and of microbiome gene expression and rial 16S rRNA (ribosomal RNA) genes present in fecal DNA. A total metabolism before and after an ecological “invasion” with the five- of 431,700 sequencing reads were obtained from 126 fecal samples member FMP microbial consortium,haveprovidedinsightsintothe (3426 ± 2665 sequences per sample; table S2A). Noise due to PCR ways that FMP strains and the indigenous model gut community re- and pyrosequencing artifacts was removed from this data set using spond to one another. The transcriptional responses were used as bio- software incorporated into the QIIME suite of 16S rRNA analysis markers to interrogate metatranscriptome data sets obtained from the tools (9). Denoised reads were binned into species-level phylogenet- MZ twins’ fecal specimens. ic types (phylotypes), with a species defined as isolates that share on March 29, 2016 ≥97% identity in their 16S rRNA gene sequence. To ensure even coverage across samples, we subsampled each of the 126 data sets RESULTS to 1640 reads per fecal microbiota. A phylogenetic tree was built from one representative sequence from each phylotype using FastTree’s Human studies approximately maximum-likelihood implementation (10) and com- Study design and assessment of intra- and interpersonal var- munities were compared using unweighted UniFrac (11): The UniFrac iations in the fecal microbiota of MZ twin pairs over a 4-month metric measures community similarity based on the degree to which period. Details concerning the seven adult female MZ twin pairs re- their members share branch length on a reference phylogenetic tree cruited for this study are provided in table S1. All had been vaginally of Bacteria. delivered; none consumed antibiotics in the 6 months before and To quantify temporal variation in community composition within during their participation in the present study; none had a history and between MZ twins, we generated a matrix of unweighted UniFrac of gastrointestinal diseases (including irritable bowel syndrome) or distances for all pairwise comparisons of all 126 fecal samples obtained any other acute or chronic medical conditions; none were consuming from the twins in our study. This matrix allowed us to compare any dietary supplements or probiotics at the time of enrollment; and none two fecal communities separated by all possible time intervals between had a history of gluten sensitivity or other food allergies, nor were any samplingforeachindividualineachoftheseventwinpairs(Fig.2A). vegans or lacto-vegetarians. The results indicated that no matter how far apart in time sample col- Fresh lots of an FMP were shipped every 2 weeks to the partici- lection occurred (1 week to 4 months), the phylogenetic distance be- pants’ homes from the same pilot production facility; strain-specific tween communities from the same individual was less than the distance quantitative polymerase chain reaction (qPCR)–based assays indicated between communities between co-twins or unrelated individuals. that, at the time of shipment, each gram of the FMP contained on av- UniFrac distances between samples harvested from a given individual erage 3.2 × 107 genome equivalents (GEs) of Bifidobacterium animalis increased modestly as a function of time during the 4-month period, subsp. lactis (strain CNCM I-2494) and 6.3 × 107 GEs of Lactobacillus although the changes were not statistically significant (Fig. 2A). delbrueckii subsp. bulgaricus (CNCM I-1632, CNCM I-1519). These Each sample contained 163 ± 3 (mean ± SEM) observed species- results were consistent with previous measurements of the number level phylotypes. Four of the total 1673 phylotypes identified in

www.ScienceTranslationalMedicine.org 26 October 2011 Vol 3 Issue 106 106ra106 2 RESEARCH ARTICLE our data set were found in all 126 samples; all belonged to the family determined by Friedman test with post hoc correction). The Spearman Lachnospiraceae (order Clostridiales; phylum Firmicutes) and repre- correlation test revealed no significant effect of human family mem- sented 2.5 ± 0.04% of the 16S rRNA sequences in each sample. We de- bership on the levels of B. animalis subsp. lactis during FMP con- termined that 26.4 ± 0.4% of species-level phylotypes observed in a sumption. Levels fell to below the limits of detection of the assay in given sample were consistently represented in all nine samples from all but four participants within 2 weeks of cessation of FMP consump- that individual (Fig. 2B): The family-level taxa to which these species tion (fig. S1); two of these individuals represented a twin pair, whereas belong consist principally of Lachnospiraceae, Ruminococcaceae, and the other two individuals were unrelated to each other. Veillonellaceae (phylum Firmicutes); the Bacteroidaceae and Rikenellaceae Co-occurrence analysis (see the Supplementary Material) indicated (phylum Bacteroidetes); and Coriobacteriaceae (phylum Actinobacteria). that with the FMP dosing schedule used, no species-level phylotypes Moreover, 13.7 ± 0.2% of the observed phylotypes were represented in present in the pretreatment microbiota exhibited a statistically signif- all samples from both co-twins (Fig. 2B). icant change in their proportional representation in feces in any indi- Impact of FMP consumption on fecal bacterial community vidual, during or after the period of FMP consumption. In addition, composition. A qPCR assay disclosed that 1 week after initiation no species-level taxa that were undetected in the pretreatment period of FMP consumption, the level of representation of B. animalis subsp. appeared and persisted during and/or after treatment in any individ- lactis (CNCM I-2494) was 107 cell equivalents per gram of feces; this ual [paired t test, analysis of variance (ANOVA)]. It is possible that level was sustained in all 14 individuals throughout the ensuing with even deeper sampling, differences might be revealed in feces or

7 weeks of FMP consumption (that is, there were no statistically sig- may exist in more proximal regions of the gut. Further details of this Downloaded from nificant differences between the 1-, 2-, 4-, and 7-week time points as co-occurrence analysis, including the results of tests at the genus and

A (Human) Fermented milk product consumption period

Pre4 Pre2 Pre1 FMP1 FMP2 FMP4 FMP7 Post2 Post4 = Feces (16S sequencing)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 http://stm.sciencemag.org/ Weeks = Feces (shotgun sequencing) = Feces (RNA-Seq)

B (Mouse) on March 29, 2016 = Feces (COPRO-Seq) Single = Feces (INSeq) 7 1421283542 Days = Feces (RNA-Seq) = Urine (GC/MS) = Cecal contents (COPRO-Seq) Multiple

Synthetic community species Fermented milk product (FMP) species Bacteroides caccae Bacteroides WH2 Eubacterium rectale Bifidobacterium animalis subsp. lactis CNCM I-2494 Bacteroides ovatus Clostridium scindens Faecalibacterium prausnitzii Lactobacillus delbrueckii subsp. bulgaricus CNCM I-1632 Bacteroides thetaiotaomicron Clostridium spiroforme Parabacteroides distasonis Lactobacillus delbrueckii subsp. bulgaricus CNCM I-1519 Bacteroides uniformis Collinsella aerofaciens Ruminococcus obeum Lactococcus lactis subsp. cremoris CNCM I-1631 Bacteroides vulgatus Dorea longicatena Ruminococcus torques Streptococcus thermophilus CNCM I-1630

Fig. 1. Experimental design for human and mouse studies. (A) Human ther manipulations, whereas animals in the multiple-treatment group re- study. Seven healthy lean MZ female twin pairs were sampled before, ceived additional 2-day gavages 1 and 3 weeks after the first gavage. during, and after FMP consumption. (B) Gnotobiotic mouse study. Two Samples were collected at the indicated time points for profiling bacterial groups of five germ-free mice were colonized by oral gavage at 6 to 8 community membership (shotgun and 16S rRNA gene sequencing for weeks of age with a 15-member microbial consortium constituting a human fecal samples, COPRO-Seq for mouse fecal and cecal samples), model human gut microbiota (day of gavage denoted by black arrows). gene expression profiling (microbial RNA-Seq), and metabolite analysis Two weeks later, the five-species FMP strain consortium was admin- (urines, GC-MS). The species that compose the model 15-member hu- istered by oral gavage to each group of mice twice over 2 days (denoted man community and the 5-member FMP consortium are listed in the gray by green arrows). Mice in the single-treatment group underwent no fur- and green boxes, respectively.

www.ScienceTranslationalMedicine.org 26 October 2011 Vol 3 Issue 106 106ra106 3 RESEARCH ARTICLE family level, plus deeper sequencing of a subset of twin pairs are sation of FMP consumption; see Fig. 1A). Two of these twin pairs lived provided in the Supplementary Material. together, whereas two pairs lived 3 and 932 miles apart. A 634-Mb data Effects of FMP consumption on the functional gene repertoire set was generated (60,863 ± 28,775 sequences per sample; average of the fecal microbiome. To determine the effects of FMP consump- length, 238 nucleotides; table S2B). A BLASTX search against version tion on the representation of gene functions in the microbiome, we 54 of the Kyoto Encyclopedia of Genes and Genomes (KEGG) GENES performed shotgun sequencing on 48 of the fecal DNA preparations database (12–14) yielded a total of 2205 ± 26 (mean ± SEM) KEGG generated from four of the MZ twin pairs (n = 6 samples per individ- Orthology identifiers (KOs) per microbiome sample: 64% of the ual; two fecal samples obtained before, two during, and two after ces- KOs in a given sample (1417 ± 46) were consistently represented in

A C Average KOs/sample (2205±26 KOs/sample) Self Co-twin 64% of KOs in a given sample Unrelated (1417±46 KOs/sample) consistently represented in all samples of an individual

0.6 55% of KOs in a given sample (1212±33 KOs/sample) consistently 0.58 represented in all samples of an individual and her co-twin

0.56 Downloaded from 0.54 41% of KOs (892) found in all 48 samples 0.52

0.5 KOs mapped KEGG category to category 0.48 Membrane transport 66 0.46 Carbohydrate metabolism 59 UniFrac distance Replication and repair 57 0.44 Amino acid metabolism 56 http://stm.sciencemag.org/ 0.42 Translation 54 Metabolism of cofactors and vitamins 44 0.4 Nucleotide metabolism 23 12345678910111213141516 Enzyme families 18 Number of weeks between samples Folding, sorting, and degradation 15 Lipid metabolism 15 Energy metabolism 15 Transcription 11 Signal transduction 11 Glycan biosynthesis and metabolism 10 Biosynthesis of secondary metabolites 10 Metabolism of other amino acids 4 Signaling molecules and interaction 3 on March 29, 2016 Xenobiotics biodegradation & metabolism 2

B D P < 0.001 Average OTUs per time point (163±3 OTUs/sample) P < 0.001 0.41 Average individual core (25% of OTUs; 40±2 OTUs/sample) 0.4 Average twin pair core 0.39 NS (14% of OTUs; 22±2 OTUs/sample) 0.38 Completely conserved core 0.37 (2.5% of OTUs found in all 126 samples) 4 OTUs, Firmicutes (family Lachnospiraceae) 0.36 Hellinger distance 0.35 Self Twin-twin Unrelated Fig. 2. Metagenomic studies of human fecal microbiomes sampled of spread provided in parentheses represent ±SEM. (C) KEGG Orthology over time. (A)16S rRNA–based time course study of intra- and inter- identifiers (KOs) consistently present within the fecal microbiome of an personal variations in fecal bacterial community structure during the individual over time (gray), between co-twins over time (orange), and in course of the 4-month study. Unweighted UniFrac measurements of all 48 microbiomes analyzed from the four sets of MZ twins during the community distances, from pairwise comparisons of all samples ob- 4-month study (red). The white box indicates the average number of tained from a given individual, from co-twins, and from unrelated indi- unique KOs (±SEM) identified in a particular sample. All measures of spread viduals are plotted as mean values ± SEM. (B) Colored boxes represent provided in parentheses represent ±SEM. (D) Hellinger distance measure- the proportion of bacterial phylotypes that were consistently present ments of fecal microbiomes based on their KO content. Tests of statistical within an individual over time (gray), between co-twins over time (orange), significance are based on 1000 permutations of a Hellinger distance matrix. and in all 126 fecal samples (red). The white box represents the average Mean values (±SEM) are shown for the three types of comparisons (self-self; number of species-level phylotypes found in a given sample. All measures co-twin–co-twin; unrelated-unrelated individual).

www.ScienceTranslationalMedicine.org 26 October 2011 Vol 3 Issue 106 106ra106 4 RESEARCH ARTICLE all 6 samples from that individual; 55% were consistently repre- Figure 1B presents the study protocol. Two groups of adult 6- to sented in all samples from both co-twins; 892 KOs (41% of the total 8-week-old germ-free C57BL/6J male mice were colonized with a sin- KOs in a given sample) were identified in all 48 samples forming a gle gavage of the 15-member community (6 × 106 CFU per member, core set of shared fecal microbiome-associated functions. Figure 2C total of 9 × 107 CFU). Each group (n = 5 animals) was maintained on provides a visual representation of this conserved set of 892 KOs: a low-fat, plant polysaccharide–rich diet. Fourteen and 15 days after 38% of the 892 belong to six KEGG categories: “membrane trans- gavage with the 15-member community, both groups of mice were port,”“carbohydrate metabolism,”“DNA replication and repair,” inoculated with a mixture of the five FMP strains. One group received “amino acid metabolism,”“translation,” and “metabolism of cofactors a second pair of gavages of the five strains 7 and 8 days later, and a third and vitamins.” pair 21 and 22 days after the first inoculation of the FMP consortium The proportional representation of KEGG pathways and their com- (multiple-treatment group). Each gavage consisted of a community com- ponent KOs was subsequently calculated for each of the 48 microbiomes. posed of 2 × 107 CFU: 25% (5 × 106 CFU) Streptococcus thermophilus, The microbiomes were then subjected to all possible pairwise compar- 25% B. animalis subsp. lactis,and25%Lactococcus lactis subsp. cremoris, isons on the basis of these two classification schemes. The results, with the remaining 25% split between the two L. delbrueckii subsp. quantified using the Hellinger distance metric, disclosed that over bulgaricus strains (12.5% each; 2.5 × 106 CFU per strain). Dosing time, unlike the UniFrac-based 16S rRNA comparisons of community was based on the following considerations: (i) a daily dose of two cups bacterial species composition, there was no significant difference in the of the FMP contains ~1010 CFU of B. animalis subsp. lactis;(ii)as- 14 degree of similarity of microbiome functional profiles for a given co- suming ~10 bacteria in the human gut, the ratio of the number of Downloaded from twin compared to the degree of similarity that existed between co- input B. animalis subsp. lactis CFU to the human gut symbiont popula- − − twins (that is, intrapersonal variation was not significantly different tion is about 10 4; (iii) to maintain this ratio of 10 4 in mice, and from interpersonal variation between co-twins). However, as with the assuming 1011 to 1012 organisms in the mouse gut, we administered a UniFrac results, individual and twin pair microbiomes were signifi- total of 107 B. animalis subsp. lactis CFU per gavage period (one period cantly more similar to one another than those from unrelated indi- equals two gavages within 24 hours); (iv) the difference in CFU between viduals (Fig. 2D). No KEGG pathways or KOs exhibited a statistically the least and the most abundant microbial species in the FMP product significant change in their relative abundance in response to FMP remainslessthanorequaltotwofold during manufacture and storage; http://stm.sciencemag.org/ consumption in any of the subjects at any of the time points (Student’s therefore, each species in the gavage was represented at equivalent paired t test and two-way ANOVA with Bonferroni post hoc testing). levels. At this point in our analysis, the human studies indicated that ex- FMPs contain defined collections of microbes, a dairy matrix, and posure of a healthy individual’s resident gut microbiota to the FMP products of microbial metabolism of this matrix generated during strains did not produce a detectable perturbation in fecal bacterial spe- manufacturing and storage. In principle, the dairy matrix and micro- cies composition, nor did it have a broad effect on the functional pro- bial metabolic products could be responsible for some of the effects file of fecal microbiome genes. To help guide further analysis of the observed when humans consume an FMP. Nonetheless, we chose to human data sets, we turned to a simplified in vivo model of the human administer the strain consortium to mice directly by gavage, rather gut microbiota. We based our selection of model community members than by giving them the corresponding commercial FMP. This al- on several criteria. All members of this model community, or their lowed us to more precisely control dosing. It also allayed concerns on March 29, 2016 close relatives, would be represented in the fecal microbiota of the about possible unintended colonization of the gnotobiotic mice with MZ twins and other sampled human populations. They would en- microbial species (other than strains deliberately introduced during compass the three major bacterial phyla present in this host habitat manufacturing) that could be introduced during handling of an FMP (Firmicutes, Bacteroidetes, and Actinobacteria) and would have deep before gavage. draft or finished genome sequences available. Gnotobiotic mice har- Therepertoireofcarbohydrate-active enzymes in members of boring such a model human microbiome would be used to character- the FMP consortium and model human gut microbial community. ize the impact of FMP strain introduction on the community’s species The genomes of the five FMP strains in this study were sequenced, and microbial gene abundances, as well as the microbiome’stranscrip- either completely (B. animalis subsp. lactis)oratadeepdraftlevel tional profile, and to ascertain the impact of the model community on (other four strains) for subsequent analyses of their representation the abundance and gene expression profiles of the FMP strains whose in the model community after gavage of gnotobiotic mice, and so we genome sequences were also known. The knowledge gleaned would could define their in vivo patterns of gene expression (table S3). then be used to help guide further analysis of the human fecal micro- Analysis of the 48 carbohydrate-active enzyme (CAZyme) families biome data sets, including microbial RNA-Seq data sets generated (16) identified in the five FMP strains and the 126 CAZyme families from a subset of the human fecal samples. identified in the 15-member model human microbiota disclosed that 23 of the 24 CAZyme glycoside hydrolase (GH) families, 11 of the 12 Studies in gnotobiotic mice glycosyltransferase (GT) families, 4 of the 4 carbohydrate esterase (CE) A community of 15 sequenced human gut-derived microbes contain- families,and4ofthe8carbohydrate-binding modules (CBMs) repre- ing a total of 58,399 known or predicted protein-coding genes was sentedintheformerwerealsorepresentedinthelatter.TheFMPcon- constructed (Fig. 1B and table S3). Figure S2 uses assigned KOs to sortium contains only six CAZyme families that were not represented provide evidence of the functional similarity of this model human in the model human gut community. Three of these are associated microbiome to a collection of 127 genomes generated from cultured with L. lactis subsp. cremoris: of these, two are predicted to play roles members of the human gut microbiota, a deeply sampled set of fecal in the binding and metabolism of chitin (fig. S3 and table S4). The microbiomes obtained from 124 unrelated Europeans (1), and deeply other three are from B. animalis subsp. lactis: BALAC2494_01193 en- sampled microbiomes from a pair of obese MZ co-twins (15). codes a GT39 family mannose transferase involved in O-glycosylation

www.ScienceTranslationalMedicine.org 26 October 2011 Vol 3 Issue 106 106ra106 5 RESEARCH ARTICLE of proteins; BALAC2494_01288 specifies a predicted b-mannanase inserted transposon mutant strains covering 3435 of the organism’s carrying a C-terminal CBM10 module predicted to bind cellulose; 4779 genes (72%). As noted in the Supplementary Material and table BALAC2494_01971 gives rise to a protein with a CBM23 module S6, by comparing the representation of mutants in fecal samples be- predicted to bind mannan. foreandafterintroducingtheFMPstrains,wewereabletodetermine Minimal changes in the species representation of the 15-member that their presence did not affect the profile of B. thetaiotaomicron’s model human gut microbiota after introduction of the FMP strain genetic determinants of fitness in the distal gut. consortium. We used COmmunity PROfiling by Sequencing (COPRO- Microbial RNA-Seq analysis of the response of B. animalis subsp. Seq), a generally applicable method based on highly parallel DNA lactis to the gut environment and members of the 15-member com- sequencing (17), to quantify the proportional representation of each munity to the FMP strain consortium. Moving beyond COPRO-Seq– component of the 15-member microbiota and of the FMP consortium based structural analysis, we performed microbial RNA-Seq analysis in our gnotobiotic mice. Sequencing reads generated from fecal DNA to determine the functional impact of exposing the established model samples collected before, during, and after introduction of the FMP human community to the FMP strains and to ascertain which FMP strains were analyzed as described in fig. S4A. Briefly, “informative” consortium genes are most highly expressed in the intestines of these tags (that is, reads that can be mapped uniquely to a single genome) animals. B. animalis subsp. lactis attained sufficient abundance in were first identified. Informative tags were then summed by species to gnotobiotic mice to allow profiling of its transcriptome at late time generate digital “counts” of abundance. In cases where a read could points (days 35, 36, and 42). When its in vivo patterns of gene expression not be assigned with certainty during COPRO-Seq analysis, it was ig- were compared with those documented during mid-log and stationary Downloaded from nored. To account for this fact, we normalized species-specific counts phase in MRS medium and in the commercial FMP (see the Supple- using their “informative genome size” (defined as the percentage of all mentary Material and table S7), we noted that the BALAC2494_00604- possible k-mers a genome can produce that are unique multiplied by BALAC2494_00614 locus, encoding enzymes involved in the catabolism the total genome length). Multiplex sequencing using the Illumina of xylooligosaccharides (18), was strongly up-regulated in vivo (average GA-IIx instrument yielded sufficient numbers of reads per sample so across the locus; 27-fold at the day 42 time point compared to mid-log that an organism comprising ≥0.003% of the community could be de- phase in MRS monoculture; 128-fold compared to the FMP; table S7). tected: For a mouse colonized at 1011 to 1012 CFU/ml cecal contents or Xylose is the main building block of dietary hemicelluloses. Addition http://stm.sciencemag.org/ feces, this represents 106 CFU/ml. of this pentose sugar is also one of the first steps in O-glycosylation of COPRO-Seq produced several notablefindings.First,community host mucins. These results support previous observations suggesting assembly before introduction of the FMP strains occurred in a highly that xylooligosaccharides may serve as potent “bifidogenic factors,” reproducible manner, both within and between the two groups of ani- whose consumption increases the densities of Bifidobacteria in the mals (fig. S5A and table S5A). This reproducibility ensured that ani- gut (19, 20). mals in both treatment groups harbored communities with structures Ordination of samples and B. animalis subsp. lactis CAZyme gene comparable to one another at the time of administration of the five- expression patterns by correspondence analysis identified additional member FMP strain consortium. Second, within 1 week of introducing CAZymes that correlate strongly with the in vivo state (Fig. 3), includ- the FMP strains, either in a single treatment or in multiple treatments, ing members of families expected to play roles in the degradation of B. animalis subsp. lactis and L. lactis were detectable in the fecal dietary plant polysaccharides [GH43 (xylan b-xylosidases), GH77 on March 29, 2016 microbiota (fig. S5B and table S5A). These two species persisted in (4-a-glucanotransferases)]. The analysis revealed sets of B. animalis the gut throughout the study. Importantly, B. animalis subsp. lactis subsp. lactis CAZymes that corresponded well with each growth was the most prominently represented member of the FMP consor- condition (that is, MRS medium, commercial dairy matrix, and mice). tium in the model human gut microbiota, exhibiting a progressive Within each growth condition, the expressed groups of CAZymes increase in its representation during the 28 days after initial introduc- often had related functions (Fig. 3). tion, and reaching comparable levels in both the single- and the multiple- We next examined the impact of the FMP strain consortium on treatment groups (up to 1.1%; see fig. S5B). In contrast, S. thermophilus expression of genes in the 15-member community. In a “top-down” and the two strains of L. delbrueckii subsp. bulgaricus were undetectable analysis, genes were binned by function and the community’smeta- or intermittently just over the limit of detection in the single- or the transcriptome was evaluated in aggregate, ignoring the species from multiple-treatment groups. Third, as with the MZ twin pairs, introduc- whicheachtranscriptarose.Acomplementary“bottom-up” analysis tion of the consortium led to minimal rearrangements in overall com- allowed us to determine how each species in the community re- munity structure, whether or not the consortium was administered sponded to the introduction of the FMP consortium. twice in a 2-day period or on two subsequent occasions (see table Top-down analysis of the impact of the FMP strains on the com- S5B for the results of Mann-Whitney tests of significance for each spe- munity metatranscriptome revealed significant increases in expression cies at each time point surveyed relative to the time point just before of genes falling within the KEGG categories carbohydrate metabolism initial introduction). Collinsella aerofaciens, the lone Actinobacteria and “nucleotide metabolism,” whereas decreases were observed in in the 15-member community, showed a significant reduction in its amino acid metabolism and “lipid metabolism” (Fig. 4A and table S8). abundance in both treatment groups immediately after FMP strain Peak responses in both treatment groups occurred 3 weeks after the introduction (fig. S5C) that persisted through later time points, raising first gavage of the FMP strains, corresponding to the time of highest the possibility of a competitive relationship between this organism and representation of B. animalis subsp. lactis in the community. B. animalis subsp. lactis, the only Actinobacteria in the FMP strain The genes that exhibited the highest fold change in expression were consortium. heavily skewed toward the KEGG categories carbohydrate metabolism The Bacteroides thetaiotaomicron component of the 15-member and membrane transport. The latterincludesanumberofadenosine human gut microbiota was composed of a library of 34,544 randomly triphosphate–binding cassette (ABC)– and PTS (phosphotransferase

www.ScienceTranslationalMedicine.org 26 October 2011 Vol 3 Issue 106 106ra106 6 RESEARCH ARTICLE

BALAC2494_01359 CAZy Gene family EC Activity predicted by EC BALAC2494_01354 BALAC2494_00644 GH2 3.2.1.23 β-Galactosidase 2 BALAC2494_01349 BALAC2494_00004 GT2 NA NA BALAC2494_01223 GT2 2.4.1.- Hexosyltransferase BALAC2494_01233 BALAC2494_01233 GT2 2.4.1.- Hexosyltransferase Product BALAC2494_01349 GT2 2.4.1.- Hexosyltransferase 1 BALAC2494_00004 P2 BALAC2494_01354 GT2 2.4.1.- Hexosyltransferase BALAC2494_01223 P3 BALAC2494_01359 GT2 2.4.1.83 Dolichyl-phosphate P1 In vivo β M3 M2 BALAC2494_00637 GH42 3.2.1.23 -Galactosidase M7M6M β M9 BALAC2494_01067 GH42 3.2.1.23 -Galactosidase 0 M8

CA2 M4 BALAC2494_00605 GH43 3.2.1.- Glycosidase L2 M55 BALAC2494_01173 M10 M1 BALAC2494_00637 BALAC2494_00608 GH43 3.2.1.37 Xylan 1,4-β-xylosidase L1 S1 BALAC2494_00605 BALAC2494_00612 BALAC2494_00612 GH43 3.2.1.37 | Xylan 1,4-β-xylosidase 3.2.1.55 BALAC2494_00608 S2 BALAC2494_01173 GH77 2.4.1.25 4-α-Glucanotransferase –1 BALAC2494_01067 In vitro (MRS) Fig. 3. Correspondence analysis of B. animalis subsp. lactis CAZyme gene expression. RNA-Seq data for all B. animalis Downloaded from subsp. lactis genes encoding known or predicted CAZymes –2 were subjected to unconstrained correspondence analysis BALAC2494_00644 using the “vegan” package in R. Correspondence analysis (CA) allows for the generation of biplots in which samples –3 –2 –1 012 and genes can be plotted in the same ordinate space to CA1 reveal associations/anti-associations between the two.

Circles represent individual CAZymes (genes). The genes http://stm.sciencemag.org/ ordinating furthest from the origin in the direction of one of the sample clusters (treatment groups) are labeled according to their locus number and are colored on the basis of CAZyme family assignment (see table to the right of the figure for details; NA, no designation). Red triangles represent samples and are labeled according to the following nomenclature: LX, logarithmic-phase cells in MRS with X being the technical replicate number (for example, L1 refers to the first technical replicate harvested in log phase); SX, stationary-phase cells in MRS with accompanying replicate number; MX, feces from designated gnotobiotic animals obtained 4 weeks after the initial invasion with the FMP strain consortium; PX, samples obtained after fermentation in the FMP dairy matrix. Each cluster of samples from a particular treatment is associated with a functionally related set of expressed CAZymes. system)–type carbohydrate transporters (table S9). When these KEGG single- and the multiple-treatment groups through the remaining 4 weeks category–level responses were subsequently broken down into KEGG of the experiment (Fig. 5A). The levanase response showed remarkable on March 29, 2016 pathways (Fig. 4B), it was apparent that the most significant responses species specificity: This gene is represented in 8 members of the 15-member in the carbohydrate metabolism category involved increases in “starch community, yet the community’s transcriptional response is driven al- and sucrose metabolism,”“fructose and mannose metabolism,” and most exclusively by the levanase in Bacteroides vulgatus (BVU_1663; “pentose and glucuronate interconversions.” Fig. 4C). In contrast, the pectinesterase response was distributed across Transcript data were subsequently binned by enzyme commission six members of the community (Bacteroides caccae, Bacteroides ovatus, (EC) number. The levels of mRNAs encoding these ECs at each time B. thetaiotaomicron, B. vulgatus, Bacteroides WH2, C. aerofaciens), with point were compared using ShotgunFunctionalizeR, an R-based sta- changes in transcription largely due to pectinesterase genes found tistical and visualization tool originally designed to identify genes in B. ovatus (BACOVA_03576, BACOVA_03581, BACOVA_04902), significantly enriched or depleted in environmental microbiomes B. thetaiotaomicron (BT_4109, BT_4110), B. vulgatus (BVU_1116), and (21, 22). Using this approach, we were able to determine that the B. WH2 (BACWH2_3569, BACWH2_3615). Increases in the propor- starch and sucrose metabolism pathway response to the FMP strains tional abundance of cellobiose phosphorylase transcripts reflected was driven by significant up-regulation of genes encoding three en- the contributions of three community members: Bacteroides uniformis, zymes involved in metabolism of dietary plant polysaccharides: (i) Eubacterium rectale,andRuminococcus obeum (table S8). EC 3.2.1.65 (levanase; Fig. 4C), which cleaves 2,6-b-D-fructofuranosidic The KEGG starch and sucrose metabolism, pentose and glucuronate linkages in 2,6-b-D-fructans (levans); (ii) EC 3.1.1.11 (pectinesterase), interconversions, and “pentose phosphate” pathways process products which de-esterifies pectin to pectate and methanol; and (iii) EC 2.4.1.20 generated by these three enzymes. Figure 5B shows that many of the (cellobiose phosphorylase), which uses cellobiose formed from par- other components of these pathways are up-regulated in the 15- tial hydrolysis of cellulose as its substrate to generate a-D-glucose-1- member community when the FMP strain consortium is introduced. phosphate and D-glucose. The genes encoding these ECs, which catalyze ShotgunFunctionalizeR also identified significant increases in the ex- early steps in three entry points of the starch and sucrose metabo- pression of genes encoding five ECs that participate in the generation lism KEGG pathway, underwent significant increases in their ex- of propionate and succinate: The induction occurred within 24 hours pression within 24 hours after introduction of the FMP consortium after the FMP strains were introduced and involved acetate kinase (Fig. 5A). The levels of expression of these genes either increased (EC 2.7.2.1; catalyzes a bidirectional reaction between propanoyl further (levanase) or were sustained (the other two ECs) in both the phosphate and propionate), phosphate acetyltransferase (EC 2.3.1.8),

www.ScienceTranslationalMedicine.org 26 October 2011 Vol 3 Issue 106 106ra106 7 RESEARCH ARTICLE

A B Day 0.030 Day

0.14 ** 14 14 15 15 35 35 0.025 36 36 0.12 42 42 /all pathways)

/all categories) 0.020 X

X 0.10

0.08 0.015

0.06 0.010 *

0.04 0.005

0.02 Norm. counts ratio (Pathway Norm. counts ratio (Category Downloaded from 0

m m m

metabolism metabolis metabolis Lipid metabolism Sulfur metabolism interconversionsStarch and sucrose Purine metabolis Fatty acid metabolism Nucleotide metabolism Fructose and mannose Alanine, aspartate and Pyrimidine metabolism Amino acid metabolism Glutathione metabolismTaurine and hypotaurine Carbohydrate metabolism Pentose and glucuronate glutamate metabolism KEGG category KEGG pathway http://stm.sciencemag.org/

C 700 Day

14 15 600 35 36 42 500

400 on March 29, 2016

300

Normalized counts 200

100

0

6 1 6 8 9 9 2 4 2 Total BT_175 BT_1765 BT_308 BDI_3373 BVU_1663

BACUNI_0115 BACUNI_0115 BACUNI_0115 BACWH2_246 BACCAC_0272BACCAC_02731BACOVA_0450BACOVA_04507 RUMOBE_0217RUMOBE_02174

Genes encoding levanase (EC 3.2.1.65)

Fig. 4. Top-down analysis of the effects of the FMP strain consortium on the (for example, the increased expression of levanase-encoding genes) can be model 15-member community’s metatranscriptome. RNA-Seq reads were dissected to identify the subset of genes/species driving the response. Boxes mapped to the sequenced genomes of the 15 community members. denote top quartile, median, and bottom quartile. Whisker length represents Transcript counts were normalized [reads per kilobase of gene length per mil- 1.5× interquartile range, except where there are no outliers; in these situa- lion reads (RPKM), see the Supplementary Material] and binned using the hi- tions, whiskers span the range from minimum to maximum values. Box color erarchical levels of functional annotation used by KEGG. (A and B)Foreach denotes the day fecal samples were obtained (day 14 is the pretreatment KEGG category (A) or pathway (B) shown, box plots depict the proportion of time point immediately preceding gavage of the FMP strain consortium). normalized read counts assignable to that annotation out of all reads, which When an asterisk is centered over a box, it indicates that there was a statis- could be assigned annotations for that hierarchical level. Data shown corre- tically significant change after administration of the FMP consortium relative spond to the “multiple-treatment” groupofmice(thegroupforwhichthe to the pretreatment time point (P < 0.05 by paired, two-tailed Student’s t most time points were collected); however, data for all mice are provided test). The positioning of asterisks above versus below a box emphasizes in table S8. (C) Illustration of how a model community’sfunctionalresponse the direction of change (above, up-regulation; below, down-regulation).

www.ScienceTranslationalMedicine.org 26 October 2011 Vol 3 Issue 106 106ra106 8 RESEARCH ARTICLE methylmalonyl-CoA (coenzyme A) decarboxylase (EC 4.1.1.41), outliers were B. WH2 [comprised 39.6 ± 1.6% (mean ± SD) of the propionyl-CoA carboxylase (EC 6.4.1.3), and methylmalonyl-CoA community but only contributed 15.4 ± 2% of the raw reads to the mutase (EC 5.4.99.2, yields succinyl-CoA as its product) (fig. S6). Only total RNA-Seq read pool] and R. obeum (2.1 ± 0.4% of the commu- a single treatment with the FMP consortium was required to produce nity; 18.2 ± 4.4% of the transcript pool)(fig.S7).Theseobservations a sustained response involving the enzymes that can yield propionate indicate that community-level transcriptional responses can be driven (fig. S6). by species representing small fractions of the microbiota. A breakdown of RNA-Seq reads by the community member ge- Our bottom-up analysis is summarized in fig. S8 and table S10, nome to which they mapped revealed that the abundance of a species and disclosed early- and later-responding species. Specifically, there in the 15-member community did not necessarily correlate with its were more significantly highly regulated R. obeum transcripts within contribution to the community transcript pool. At the time point sam- the community metatranscriptome 1 day after gavage than would be pled immediately before invasion (day 14), two of the most extreme expected based on its community representation, and more highly

Fig. 5. Mouse and human com- A munities share transcriptional EC 2.4.1.20 EC 3.1.1.11 EC 3.2.1.65 Cellobiosephosphorylase Pectinesterase Levanase Relative fold change responses to the FMP strain con- − in expression 10 2.5 sortium involving ECs related to −10 Downloaded from carbohydrate metabolism. (A) −3 10 −2 Box plots of the proportion of all − 1 RPKM-normalized reads in mouse 10 3.5 2 and human fecal metatranscrip- −4 tomes represented by three ECs 10 10

− involved in plant biomass deg- 10 4.5

radation. Individual samples are % of normalized counts − shown as black dots (n =2to 10 5 http://stm.sciencemag.org/ 10). Boxes are also colored by fold change, as determined by comparing mean values at a giv- en time point to the value at the d15 treatment d15 treatment pretreatment time point [for gno- d15 treatment 1 week into FMP 1 week into FMP 1 week into FMP 4 weeks into FMP 4 weeks into FMP 4 weeks into FMP d14 pre-treatment d14 pre-treatment tobiotic mice, pretreatment refers d14 pre-treatment 1 week before FMP 1 week before FMP 1 week before FMP d42 single treatment to day 14; in the case of humans, d42 single treatment d42 single treatment d42 multiple treatment d42 multiple treatment d42 multiple treatment 4 weeks after last FMP 4 weeks after last FMP pretreatment refers to the fecal 4 weeks after last FMP sample collected 1 week before Mouse Human Mouse Human Mouse Human initiation of FMP consumption

(sample “Pre1” in Fig. 1A)]. Statisti- B on March 29, 2016 1 day after FMP strain introduction vs pre-FMP Pectin Xylitol cal significance was determined 4 weeks after FMP strain introduction (multiple treatment) vs pre-FMP 1 day after FMP strain introduction vs pre-FMP using the ShotgunFunctionalizeR EC 3.1.1.11 EC 1.1.1.21 package in R and an adjusted P 4 weeks after FMP strain introduction (multiple treatment) vs pre-FMP cutoff of <0.01. Pretreatment time A Cellobiose D-Xylose D-Sorbitol points and subsequent time points α D-Glucose-1P EC 2.4.1.20 EC 5.3.1.5 where expression levels were not EC 1.1.1.21 significantly different from the pre- EC 5.4.2.2 D-Xylulose EC 1.1.1.14 EC 5.3.1.9 treatment mean are colored white. α D-Glucose α β-D-Glucose-6P (B) Components of KEGG starch D-Glucose-6P EC 2.7.1.17 EC 5.3.1.5 and sucrose metabolism, pentose EC 5.3.1.9 and glucuronate interconversions, EC 2.7.1.1 EC 2.7.1.4 and pentose phosphate pathways D-Fructose β-D-Fructose-6P D-Xylulose-5P whose expression in the 15-member D-Erythrose-4P EC 3.1.3.- model community changed com- EC 2.7.1.90 EC 2.7.1.11 pared to pretreatment values when EC 2.2.1.2 EC 2.2.1.1 D-Fructose-2P β-D-Fructose-1,6P the 5-member FMP strain consorti- 2 um was introduced. Gray indicates D-Sedoheptulose-7P EC 2.7.1.69 EC 3.2.1.65 that the fold change was statisti- EC 4.1.2.13 cally significant (adjusted P <0.01). Levan D-Fructose-1P D-Glyceraldehyde-3P EC 2.2.1.1 Ovals highlight the three enzymes shown in (A). Dashed arrows indi- L-Lyxose L-Arabinose L-Ribulose L-Ribulose-5P EC 5.1.3.4 cate that multiple enzymatic reac- EC 5.3.1.4 EC 2.7.1.16 EC 5.-.-.- tions lead from these ECs and EC 5.3.1.- their indicated substrates to the EC 2.7.1.53 products shown. These interme- L-Xylulose-1P EC 2.7.1.5 L-Xylulose L-Xylulose-5P diate reactions have been omitted for clarity or because the omitted ECs did not manifest significant changes in their expression.

www.ScienceTranslationalMedicine.org 26 October 2011 Vol 3 Issue 106 106ra106 9 RESEARCH ARTICLE regulated R. obeum genesinthecomparisonbetweenday14(just alternatively, by the induction of microbiome genes encoding xylan- before gavage) and day 15 metatranscriptomes than between day 14 degrading enzymes (for example, BACOVA_04387 and BACOVA_04390, and day 42 metatranscriptomes. In contrast, B. WH2, Clostridium which were up-regulated 5.2- and 11.0-fold, respectively, after intro- scindens, and B. uniformis were defined as late responders to the duction of the FMP strains; table S10). Changes in other metabolites FMP consortium. such as xanthosine (Fig. 6C), a purine metabolite, suggest that the Identifying predictive features from the model community metabolic consequences of FMP strain introduction extend beyond metatranscriptome data using a random forests classifier. Ma- the processing of carbohydrates. chine learning techniques using the random forests classifier can be Collectively, our transcriptional and metabolite analyses indicated applied to metagenomic data (23) to learn a function that maps a set that introducing FMP strains that constitute a small fraction of a de- of input values or predictors (in this case, relative abundance of fined model human gut microbiota signals the microbiota to change KEGG categories, KEGG pathways, or ECs in a community) to a dis- its metabolic activities, including activities related to carbohydrate me- crete output value (here, the presence/absence of the FMP strains). tabolism. With this information in hand, we returned to the human KEGG categories, KEGG pathways, and ECs were all able to predict fecal samples to determine the extent to which observations made in pretreatment/posttreatment status with low estimated generalization our gnotobiotic mouse model were applicable to humans. error (KEGG categories, 6.7%; ECs, 13.3%; KEGG pathways, 10.0%). Microbiome transcriptional responses to FMP strains that are In all cases, these generalization error rates were less than half of the shared by gnotobiotic mice and humans. Microbial RNA-Seq baseline error rate of 33% (that is, that achieved by always predicting analysis was performed on human fecal samples obtained 1 week be- Downloaded from thelargestcategory).Therewere11predictiveand5highlypredictive fore FMP consumption, 1 and 4 weeks into the consumption period, KEGG categories, 35 moderately predictive ECs, and 27 predictive and and 4 weeks after cessation (both co-twins from family 1; one co-twin 4 highly predictive KEGG pathways (table S11). The predictive ECs fromfamily3;seetableS1).Usingananalysispipelinecomparableto identified using our supervised classification approach include a num- the one used for the mouse data, we first aligned all RNA-Seq reads ber of carbohydrate metabolism–related functions that were also iden- against a reference set of 131 human gut microbial genomes plus the tified using ShotgunFunctionalizeR in our top-down analysis. FMP strain genomes, binned the aligned transcripts on the basis of Metabolomic analyses. To evaluate the impact of invasion with their EC annotations, and used ShotgunFunctionalizeR to identify http://stm.sciencemag.org/ the five-member FMP consortium on microbial-host co-metabolism, ECs whose abundances were significantly changed as a function of we performed untargeted gas chromatography–mass spectrometry FMP exposure (Benjamini-Hochberg adjusted P < 0.01). (GC-MS) on urine samples collected at multiple time points (days 0, Categorical analysis of the responses of the human fecal commu- 14, and 42) from members of the single- and multiple-treatment groups nity to FMP consumption revealed that significantly up-regulated ECs (Fig.1B).Ametaboliteprofilewasconstructed for each urine sample were principally distributed among the KEGG categories carbohydrate (n = 19) using the spectral abundances of all identifiable metab- metabolism, amino acid metabolism, and metabolism of cofactors and olites. A total of 198 metabolites met our reverse match score cutoff vitamins (see table S13 for a complete list of ECs identified from the of 65% and were present in at least 50% of samples at one or more various pairwise comparisons of time points). time points [for an explanation of the reverse match score, see (24) Figure 7 highlights the 86 ECs that were significantly changed and table S12]. Comparing day 0 and 14 samples revealed 39 metab- (P < 0.01) in the same direction in all humans and in all sampled mice on March 29, 2016 olites whose levels were significantly higher or lower after coloniza- as a function of exposure to the FMP strain consortium. Similar to our tion with the defined 15-member community (see table S12A). The findings in mice, the most prominently represented KEGG category changes included decreases in the levels of oligosaccharides we would among up-regulated gene functions in all comparisons of human meta- expect to be consumed by members of the microbiota [melibiose (87% transcriptomes was carbohydrate metabolism (Fig. 7). The three ECs decline); raffinose/maltotriose (98%); note that oligosaccharides are involved in entry points in the KEGG starch and sucrose metabolism by their nature difficult to identify with certainty with the present, pathway [levanase (EC 3.2.1.65), pectinesterase (EC 3.1.1.11), and cel- nontargeted GC-MS technique, and our annotations of these metab- lobiose phosphorylase (EC 2.4.1.20)] were significantly up-regulated olites as melibiose and raffinose/maltotriose are provisional]. The within 1 week after FMP consumption was initiated in the humans observed 3.4-fold increase in pyrogallol, a polyphenol, is consistent surveyed. This transcriptional response was sustained in the case of with the known ability of many gut microbes to cleave these mole- levanase and pectinesterase and ceased (fell to below the limits of cules from polyphenols present in dietary plant material. A 4.4-fold detection) within 4 weeks after FMP administration was stopped increase in taurine after the initial colonization of mice was also (Fig. 5A). noted, probably a result of microbial deconjugation of taurine from ECs involved in succinate and propionate metabolism (EC 2.7.2.1 bile compounds. and EC 6.4.1.3) were also up-regulated in the human fecal metatran- Table S12B lists urinary metabolites that change significantly after scriptome within 1 week of the initiation of FMP consumption (FMP1 introducing the five FMP strains (compare day 14 versus day 42 in versus Pre1, Fig. 7). As with levanase, pectinesterase, and cellobiose table S12B). Fructose and xylose were not significantly affected by phosphorylase, this response was sustained during and subsided after the introduction of the defined 15-member community but increased the period of FMP consumption (see “FMP4 versus Pre1” and “FMP1 significantly after the introduction of the FMP strain consortium (2.3- versus Pre1” in Fig. 7 and table S13). and2.9-fold,respectively;Fig.6,AandB).Increasesinfructosemay Human fecal transcripts were detected that mapped to the B. animalis reflect an enhanced capacity of the community to liberate this mono- subsp. lactis genome (see the Supplementary Material). The presence saccharide from levan and other polyfructans via levanase-catalyzed of these transcripts was limited to the period of FMP consumption, reactions. Increases in xylose might be explained by the additional supporting the notion that they emanated from the FMP strain rather xylanase activity introduced by B. animalis subsp. lactis (Fig. 3) or, than from a related species present within the microbiota (fig. S9).

www.ScienceTranslationalMedicine.org 26 October 2011 Vol 3 Issue 106 106ra106 10 RESEARCH ARTICLE

A Fructose (urine) B Xylose (urine) consumption. Analyses of the fecal gene n/s P = 0.0033 repertoire over the course of the 16 weeks 28 28 of the experiment indicated that (i) varia- n/s P = 0.047 n/s P = 0.016 tions in the functional features of the (fecal) 26 26 microbiome were less than the variations 24 in bacterial species composition; (ii) there 24 was no significant difference in the degree 22 of similarity in representation of KEGG 22 orthology group functions for a given

(spectral abundance) (spectral abundance) 20 2 2 co-twin at each time point compared to the degree of similarity that existed be- Log 20 Log 18 4 2 4 2 tween co-twins, whereas individual and d0 d0 4 d1 d4 d1 d twin pair microbiomes were significantly Time point Time point more similar to one another than those Fig. 6. Select urinary metabolites whose levels from unrelated individuals; and (iii) there C Xanthosine (urine) are altered after the introduction of the FMP strain were no statistically significant changes

n/s consortium into mice harboring a defined model hu- in the representation of these functions Downloaded from 24 man gut microbiota. (A to C) The statistical signifi- n/s P = 0.048 when the FMP strain consortium was be- cance in pairwise comparisons was evaluated using ing consumed. With these findings in 22 ’ a two-tailed Students t test on the log-transformed mind, and with each individual as well as spectral abundance of the metabolite in each sam- 20 each genetically identical co-twin serving ple. Values for the statistical significance of differ- as a control, we concluded that at least at 18 ences between time points as evaluated by one-way ANOVA, followed by false discovery rate correction the depth and frequency of sampling used and a post hoc Tukey’s honestly significant differ- for this small healthy cohort, the bacterial http://stm.sciencemag.org/

(spectral abundance) 16 2 ence test are also provided in table S12. Horizontal species and gene content of their fecal

Log 14 bars represent group means; vertical bars represent microbiota/microbiome were not an in- 4 d0 ±SEM.n/s,notsignificant. formative biomarker for understanding d1 d42 whetherorhowthiscommercialFMPaf- Time point fected microbial community properties. Gnotobiotic mice harboring a model 15-member gut microbial community that This clear linkage to FMP consumption was not evident in the case represented the three principal bacterial phyla present in the human of other members of the consortium, so we could not confidently ana- gut microbiota, and whose 58,399 known or predicted protein-coding lyze their patterns of gene expression in vivo. The highest number of genes encompassed many of the prominent functions present in the on March 29, 2016 mapped reads to the B. animalis subsp. lactis genome was obtained normal adult human fecal microbiome, provided a means for charac- 1 week after FMP administration began: Among the 4000 reads, we terizing the impact of the five-member FMP strain consortium on were able to detect transcripts from all but 1 of the 10 genes in the expressed gut microbial community functions and then applying the BALAC2494_00604-BALAC2494_00614 locus that encodes enzymes resultstothehumanfecalspecimenscollectedforthisstudy.Aswith involved in the catabolism of xylooligosaccharides, leading us to con- the MZ twins, introduction of the five-member strain consortium did clude that this locus is highly expressed in the distal human gut, just not significantly affect the representation of the 15 species that make as it is in our mouse model. up the model human microbiota. As with the MZ twins, B. animalis subsp. lactis exhibited the greatest fitness of the five FMP strains in the gut, as judged by its prominence and persistence. Unlike the human DISCUSSION arm of the study, where all subjects consumed the FMP twice daily, the design of the mouse study, with its single- versus multiple-treatment Repeated sampling of seven healthy MZ adult twin pairs over a 4-month regimens, allowed us to directly compare the persistence of FMP con- period emphasized that intrapersonal variation in bacterial communi- sortium members. Only B. animalis subsp. lactis and L. lactis subsp. ty structure was less than interpersonal variation, with co-twins having cremoris were able to maintain a foothold in the gut ecosystem at significantly more similar phylogenetic and taxonomic structure in their detectable levels for the entire 4-week monitoring period after a single fecal microbiota compared to those from unrelated individuals (9, 25, 26). dose. In addition, colonization levels were not affected by the number The results also showed that (i) consumption of an FMP containing of times the FMP strains were administered to mice. five bacterial strains was not associated with a statistically significant An advantage of constructing the model human gut microbiome change in the proportional representation of resident community mem- was that its entire predicted gene repertoire was known. This allowed bers within and between individuals; (ii) the appearance and disap- us to define the impact of introducing the FMP strain consortium on pearance of strains comprising the FMP consortium did not exhibit the functions expressed by the overall community as well as by its in- familial patterns in the fecal microbiota; and (iii) B. animalis subsp. lactis dividual components. A major theme emanating from our analysis CNCM I-2494 was the most prominent assayed member of the consor- was the effect of introducing the FMP consortium on carbohydrate tium represented in the microbiota during the 7-week period of FMP metabolism by the community, as well as the effect of the community

www.ScienceTranslationalMedicine.org 26 October 2011 Vol 3 Issue 106 106ra106 11 RESEARCH ARTICLE

on a feature of carbohydrate metabolism by B. animalis subsp. lactis. Scale The model 15-member community responded to the FMP consor- tium by inducing genes encoding enzymes involved in catalyzing re- FMP4 vs Pre1 FMP1 vs Pre1 d15 vs d14 d42 multi vs d14 3 2 1 –1–2–3 actions that represent the three entry points into the KEGG “starch

EC 2.7.1.31 Glycerate kinase and sucrose metabolic pathway,” as well as enzymes that catalyze fer- EC 1.2.1.11 Aspartate-semialdehyde dehydrogenase EC 1.4.1.16 Diaminopimelate dehydrogenase EC 3.5.4.19 Phosphoribosyl-AMP cyclohydrolase mentation of carbohydrates to propionate. The mechanism by which EC 3.6.1.31 Phosphoribosyl-ATP diphosphatase EC 4.2.1.20 Tryptophan synthase the FMP strains elicit this response is unclear at present, but the ef- EC 4.2.1.52 Dihydrodipicolinate synthase EC 6.3.1.1 Aspartate--ammonia ligase fect is rapid (occurring within the first 24 hours after invasion) and EC 4.1.3.- Oxo-acid-lyase EC 2.6.1.9 Histidinol-phosphate transaminase EC 2.4.2.- Pentosyltransferase was persistent whether the consortium was introduced in a single set EC 6.3.5.5 Carbamoyl-phosphate synthase (glutamine-hydrolyzing) EC 4.2.3.1 Threonine synthase of gavages during a 1-day period or with subsequent repeated gavage EC 6.3.4.4 Adenylosuccinate synthase EC 6.1.1.9 Valine--tRNA ligase EC 2.6.1.1 Aspartate transaminase over a several-week period. The persistence of both the carbohydrate EC 1.1.1.86 Ketol-acid reductoisomerase EC 4.1.2.14 2-Dehydro-3-deoxy-phosphogluconate aldolase pathway response and of B. animalis subsp. lactis suggests but does EC 4.1.3.16 4-Hydroxy-2-oxoglutarate aldolase EC 6.4.1.3 Propionyl-CoA carboxylase EC 6.3.5.4 Asparagine synthase (glutamine-hydrolyzing) not prove that the latter may be instrumental in instigating and main- EC 6.3.4.5 Argininosuccinate synthase EC 2.2.1.1 Transketolase taining the former. EC 2.5.1.10 Geranyltranstransferase EC 1.17.4.1 Ribonucleoside-diphosphate reductase “ EC 2.7.1.16 Ribulokinase Intriguingly, the carbohydrate response showed features of differ- EC 2.7.1.6 Galactokinase EC 3.1.1.11 Pectinesterase entiation.” As noted in the Results section, the levanase response was EC 3.2.1.65 Levanase EC 3.2.1.8 Endo-1,4-beta-xylanase driven almost entirely by changes in transcription in just a single spe- Downloaded from EC 3.5.99.6 Glucosamine-6-phosphate deaminase EC 4.2.1.44 Myo-inosose-2 dehydratase EC 4.2.1.47 GDP-mannose 4,6-dehydratase cies (B. vulgatus), the pectinesterase response by six community mem- EC 4.2.2.2 Pectate lyase EC 4.2.3.3 Methylglyoxal synthase bers (B. caccae, B. ovatus, B. thetaiotaomicron, B. vulgatus, B. WH2,and EC 5.3.1.14 L-rhamnose isomerase EC 5.3.1.17 4-Deoxy-L-threo-5-hexosulose-uronate ketol-isomerase EC 5.3.1.4 L-arabinose isomerase C. aerofaciens), and the cellobiose phosphorylase response by three EC 5.3.1.5 Xylose isomerase EC 5.3.1.8 Mannose-6-phosphate isomerase components of the defined model human gut microbiota (B. uniformis, EC 6.4.1.1 Pyruvate carboxylase EC 1.2.7.3 2-Oxoglutarate synthase EC 2.7.2.3 Phosphoglycerate kinase E. rectale, and R. obeum). Of the 50 genes with predicted xylan- EC 2.7.9.1 Pyruvate, phosphate dikinase EC 4.1.1.49 Phosphoenolpyruvate carboxykinase (ATP) degrading capacity in the model microbiome (that is, those encod- EC 4.1.2.13 Fructose-bisphosphate aldolase EC 5.3.1.1 Triose-phosphate isomerase http://stm.sciencemag.org/ EC 5.3.1.6 Ribose-5-phosphate isomerase ing enzymes in ECs 3.2.1.37 and 3.2.1.8), only BACOVA_04387 and EC 1.1.1.158 UDP-N-acetylmuramate dehydrogenase EC 5.1.3.14 UDP-N-acetylglucosamine 2-epimerase BACOVA_04390 (both from B. ovatus) were significantly up-regulated EC 3.2.1.- Glycosidase EC 3.5.1.10 Formyltetrahydrofolate deformylase EC 2.7.2.1 Acetate kinase after FMP strain introduction (this is ignoring xylanase genes encoded EC 1.2.1.12 Glyceraldehyde-3-phosphate dehydrogenase (phosphorylating) EC 2.7.6.1 Ribose-phosphate diphosphokinase by FMP strains like B. animalis subsp. lactis). This up-regulation in a EC 3.2.1.20 Alpha-glucosidase EC 6.3.2.6 Phosphoribosylaminoimidazolesuccinocarboxamide synthase limited subset of the model community coincides with an increase in EC 2.3.1.79 Maltose O-acetyltransferase EC 2.5.1.55 3-Deoxy-8-phosphooctulonate synthase EC 3.4.24.- Metalloendopeptidase urinary xylose. EC 5.2.1.8 Peptidylprolyl isomerase EC 2.4.2.1 Purine-nucleoside phosphorylase The ability to attribute EC-level changes to individual genes in spe- EC 2.7.4.3 Adenylate kinase EC 3.5.4.16 GTP cyclohydrolase I EC 4.2.1.75 Uroporphyrinogen-III synthase cific bacterial species was not possible with our RNA-Seq analysis of EC 2.1.2.3 Phosphoribosylaminoimidazolecarboxamide formyltransferase EC 5.3.1.- Intramolecular oxidoreductase the human fecal samples. The differentiation of carbohydrate re- EC 1.1.1.18 Inositol 2-dehydrogenase] EC 2.7.10.1 Receptor protein-tyrosine kinase] EC 1.1.1.69 Gluconate 5-dehydrogenase] sponses among bacterial species documented in gnotobiotic mice em- EC 1.1.1.58 Tagaturonate reductase on March 29, 2016 EC 1.2.7.- Oxidoreductase phasizes a challenge and opportunity that can be addressed in these EC 1.6.5.- Oxidoreductase EC 3.6.3.17 Monosaccharide-transporting ATPase EC 3.6.5.3 Protein-synthesizing GTPase models, namely, to further delineate the niches, interactions, and EC 3.4.11.18 Methionyl aminopeptidase EC 3.4.21.92 Endopeptidase Clp adaptive resource switching behaviors of community members by EC 6.3.5.2 GMP synthase (glutamine-hydrolyzing) EC 2.7.7.7 DNA-directed DNA polymerase EC 2.3.1.179 Beta-ketoacyl-acyl-carrier-protein synthase II intentional addition, removal or substitution of taxa, and/or their EC 2.7.3.9 Phosphoenolpyruvate--protein phosphotransferase EC 2.4.2.19 Nicotinate-nucleotide diphosphorylase (carboxylating) modification through genetic manipulation. Although requiring signif- EC 3.2.1.101 Mannan endo-1,6-alpha-mannosidase EC 4.2.99.18 DNA-(apurinic or apyrimidinic site) lyase icantly more animals and loss of the ability to use an animal as its own EC 3.1.26.- Endoribonuclease EC 6.1.1.3 Threonine--tRNA ligase control, future studies could be expanded to include sampling of com- Amino acid metabolism Carbohydrate metabolism Nucleotide metabolism munity gene expression in different segments of the small intestine. Biosynthesis of secondary metabolites Metabolism of cofactors and vitamins The increased expression of genes encoding enzymes involved in the interconversion of propionate and succinate is intriguing given the Fig. 7. Shared transcriptional responses to FMP strain exposure in mice and humans. The heat map shows ECs that exhibit a statistically significant fact that this short-chain fatty acid has been linked in some reports to change in their expression (ShotgunFunctionalizeR, adjusted P <0.01)and effects on gastrointestinal transit time. However, work in this area has manifest a consistent direction of change in their expression in all four com- yielded varying results and conclusions, perhaps because of the diver- parisons shown. Comparisons include those where the pretreatment time sity of models and methodological approaches used (27–30). Propio- point was compared with a time point shortly after FMP strains were intro- nate may also link the gut microbiota and human physiology through duced (mouse: “d15 vs d14,” human: “FMP1 vs Pre1”) and those where the itseffectsonhepaticandadiposetissuemetabolism(31). Notably, an- − − − − pretreatment period was compared to a time point several weeks after strain other group has reported that in the T-bet / Rag2 / mouse model of “ ” “ ” “ ” introduction (mouse: d42multi vs d14, human: FMP4 vs Pre1 ). d42multi colitis, consumption of an FMP containing a dairy matrix plus the indicates the multiple-treatment group at day 42 of the mouse experiment. same strains used in this study led to increased cecal propionate levels The colored boxes correspond to the KEGG categories that contain the ECs and a reduction in intestinal inflammation (32). shown to the right of the heat map. The scale refers to fold difference in the mean of relative abundance of each EC between treatment and pre- The extent of translatability of data from gnotobiotic mouse models treatment groups based on the mean number of normalized reads (RPKM) harboring collections of sequenced representatives of the human gut of transcripts assigned to a given EC. The 18 ECs shown at the bottom of the microbiota to humans themselves needs to be tested further, not only figure are not associated with the five prominent KEGG categories listed. at the transcriptional level but also at the level of community-host Their assigned categories are provided in table S13. co-metabolism. Although current models can and should be evolved

www.ScienceTranslationalMedicine.org 26 October 2011 Vol 3 Issue 106 106ra106 12 RESEARCH ARTICLE to embrace more of the diversity present in our gut communities, even SUPPLEMENTARY MATERIAL with current limitations, they can serve as part of a preclinical discovery pipeline designed to identify candidate biomarkers and mediators of www.sciencetranslationalmedicine.org/cgi/content/full/3/106/106ra106/DC1 Materials and Methods the effects of existing or new probiotic strains on the properties of mi- Results crobial communities and their hosts. They also represent an analytical Fig. S1. Levels of B. animalis subsp. lactis (CNCM I-2494) in human fecal samples collected tool for characterizing the effects of specified dietary components on before, during, and after consumption of an FMP. the indigenous gut community and on probiotic species that are de- Fig. S2. KEGG pathway coverage ratios suggest that the model human gut microbiome encodes many of the functions present in more complex human fecal communities. liberately consumed. The results could yield new candidate prebiotics Fig. S3. CAZyme profiles of the 20 bacterial strains introduced into gnotobiotic mice. that may affect the representation and metabolic properties of probio- Fig. S4. Summary of COPRO-Seq and RNA-Seq analysis pipelines used in this study. tic species or entrenched members of our gut microbiota and provide Fig. S5. COPRO-Seq–based time series analysis of the abundance of members of the model the proof-of-mechanism and proof-of-principle observations needed human microbiota and of the FMP strain consortium in the feces of gnotobiotic mice. ’ to justify, direct, and interpret human studies. Fig. S6. Top-down analysis of the model community s transcriptional response to the FMP strain consortium reveals up-regulation of genes involved in interconversion of propionate and succinate. Fig. S7. A species’ contribution to the metatranscriptome is not necessarily proportional to its MATERIALS AND METHODS abundance in the 15-member community. Fig. S8. Bottom-up analysis of genes whose expression changes significantly after introduction of the FMP strain consortium. Human studies Downloaded from Fig. S9. The number of RNA-Seq reads, obtained from human fecal samples that map to Subject recruitment. Seven MZ female twin pairs aged 21 to genomes in the FMP strain consortium, peaks shortly after FMP consumption begins. 2 32 years with body mass indices ranging from 20 to 25 kg/m were Table S1. Characteristics of adult female monozygotic (MZ) twins enrolled in study. recruited for this study. These twins were long-standing partici- Table S2. Summary of human fecal metagenomic data sets. pants in the Missouri Adolescent Female Twin Study (MOAFTS) Table S3. Features of the microbial genomes in the 5-member FMP strain consortium and the (26, 33). Procedures for obtaining consent, for providing fecal sam- 15-member model human gut microbiota. Table S4. Carbohydrate-active enzyme (CAZy) annotation data. ples, and for maintaining diaries of FMP consumption, and stool Table S5. COPRO-Seq analysis of bacterial species abundance in mouse fecal samples. frequency and consistency were approved by the Washington Uni- Table S6. INSeq analysis. http://stm.sciencemag.org/ versity Human Studies Committee. Table S7. Differentially expressed B. animalis subsp. lactis (CNCM I-2494) genes. Other procedures. Methods used for the production and distri- Table S8. Top-down function-level analysis of the impact of the FMP strain consortium on the model human gut microbiota’s metatranscriptome. bution of the FMP to study participants, analysis of the effects of FMP Table S9. Model human gut microbiota membrane transport genes demonstrating more than consumption on stool parameters, qPCR analysis of fecal levels of FMP or equal to fourfold increases or decreases in their expression after introduction of the FMP strains (table S14), multiplex pyrosequencing of 16S rRNA genes in strain consortium. fecal samples and the FMP, co-occurrence analysis, and shotgun se- Table S10. Bottom-up (gene-level) analysis of the impact of the FMP strain consortium on the model community’s metatranscriptome. quencing of human fecal microbiomes are described in the Materials Table S11. Results of random forests–supervised classification analysis. and Methods section of the Supplementary Material. Sequenced hu- Table S12. Urine metabolites whose levels change significantly in transitions between man gut–associated microbial genomes used to analyze the human fe- colonization states. cal RNA-Seq data are listed in table S15. Table S13. ShotgunFunctionalizeR analysis of EC-level changes in the metatranscriptome as a on March 29, 2016 function of FMP strain introduction into mice and humans. Table S14. Primers and amplification conditions used for quantitative PCR assays of FMP Studies in gnotobiotic mice consortium strains in fecal DNA. Colonization of germ-free mice. The justification for using mice Table S15. List of 131 human gut microbial genomes used to analyze human fecal RNA-Seq data and the protocols used for treating them were approved by the Wash- sets. ington University Animal Studies Committee. Animals belonging References to the C57BL/6J inbred strain were maintained in plastic flexible film gnotobiotic isolators and fed a standard autoclaved chow diet (B&K rat and mouse autoclavable chow #7378000, Zeigler Bros Inc.) REFERENCES AND NOTES ad libitum. Two groups of 6- to 8-week-old germ-free male animals (n = 5 per group) were colonized with a single gavage of 500 mlof 1. J. Qin, R. Li, J. Raes, M. Arumugam, K. S. Burgdorf, C. Manichanh, T. Nielsen, N. Pons, supplemented TYG medium (TYGs)(34) containing 15 sequenced F.Levenez,T.Yamada,D.R.Mende,J.Li,J.Xu,S.Li,D.Li,J.Cao,B.Wang,H.Liang, human gut-derived bacterial symbionts (6 × 106 CFU per strain; total H. Zheng, Y. Xie, J. Tap, P. Lepage, M. Bertalan, J. M. Batto, T. Hansen, D. Le Paslier, of 9 × 107 CFU for the community). The B. thetaiotaomicron compo- A. Linneberg, H. B. Nielsen, E. Pelletier, P. Renault, T. Sicheritz-Ponten, K. Turner, H.Zhu,C.Yu,S.Li,M.Jian,Y.Zhou,Y.Li,X.Zhang,S.Li,N.Qin,H.Yang,J.Wang, nent of this community was composed of a library of 34,544 transposon S. Brunak, J. Doré, F. Guarner, K. Kristiansen, O. Pedersen, J. Parkhill, J. Weissen- mutants prepared as described (34). Fourteen and 15 days later, both bach; MetaHIT Consortium, P. Bork, S. D. Ehrlich, J. Wang, A human gut microbial groups of mice were gavaged with a mixture of the five FMP strains gene catalogue established by metagenomic sequencing. Nature 464,59–65 (2010). 6 2. C. F. Favier, E. E. Vaughan, W. M. De Vos, A. D. Akkermans, Molecular monitoring of (each species at 5 × 10 CFU) in 300 mlofTYGs.Onegroupofmice succession of bacterial communities in human neonates. Appl. Environ. Microbiol. 68, received a second pair of gavages 7 and 8 days later, and a third pair of 219–226 (2002). gavages 21 and 22 days after the initial FMP strain introduction. 3.K.Kurokawa,T.Itoh,T.Kuwahara,K.Oshima,H.Toh,A.Toyoda,H.Takami,H.Morita, Other procedures. Methods used for sampling animals, COPRO- V. K. Sharma, T. P. Srivastava, T. D. Taylor, H. Noguchi, H. Mori, Y. Ogura, D. S. Ehrlich, Seq, INSeq, microbial RNA-Seq, and nontargeted metabolomics via K. Itoh, T. Takagi, Y. Sakaki, T. Hayashi, M. Hattori, Comparative metagenomics revealed commonly enriched gene sets in human gut microbiomes. DNA Res. 14, 169–181 (2007). GC-MS, are described in the Methods section of the Supplementary 4.M.Arumugam,J.Raes,E.Pelletier,D.LePaslier,T.Yamada,D.R.Mende,G.R.Fernandes, Material, as are methods for sequencing and annotating FMP strain J. Tap, T. Bruls, J. M. Batto, M. Bertalan, N. Borruel, F. Casellas, L. Fernandez, L. Gautier, genomes. T. Hansen, M. Hattori, T. Hayashi, M. Kleerebezem, K. Kurokawa, M. Leclerc, F. Levenez,

www.ScienceTranslationalMedicine.org 26 October 2011 Vol 3 Issue 106 106ra106 13 RESEARCH ARTICLE

C. Manichanh, H. B. Nielsen, T. Nielsen, N. Pons, J. Poulain, J. Qin, T. Sicheritz-Ponten, 26. P. J. Turnbaugh, M. Hamady, T. Yatsunenko, B. L. Cantarel, A. Duncan, R. E. Ley, M. L. Sogin, S.Tims,D.Torrents,E.Ugarte,E.G.Zoetendal,J.Wang,F.Guarner,O.Pedersen, W. J. Jones, B. A. Roe, J. P. Affourtit, M. Egholm, B. Henrissat, A. C. Heath, R. Knight, W. M. de Vos, S. Brunak, J. Doré; MetaHIT Consortium, M. Antolín, F. Artiguenave, J. I. Gordon, A core gut microbiome in obese and lean twins. Nature 457,480–484 H. M. Blottiere, M. Almeida, C. Brechot, C. Cara, C. Chervaux, A. Cultrone, C. Delorme, (2009). G. Denariaz, R. Dervyn, K. U. Foerstner, C. Friss, M. van de Guchte, E. Guedon, F. Haimet, 27. C. Cherbut, L. Ferrier, C. Rozé, Y. Anini, H. Blottière, G. Lecannu, J. P. Galmiche, Short-chain W. Huber, J. van Hylckama-Vlieg, A. Jamet, C. Juste, G. Kaci, J. Knol, O. Lakhdari, fatty acids modify colonic motility through nerves and polypeptide YY release in the rat. S. Layec, K. Le Roux, E. Maguin, A. Mérieux, R. Melo Minardi, C. M’Rini, J. Muller, Am. J. Physiol. 275, G1415–G1422 (1998). R. Oozeer, J. Parkhill, P. Renault, M. Rescigno, N. Sanchez, S. Sunagawa, A. Torrejon, 28. P. S. Kamath, M. T. Hoepfner, S. F. Phillips, Short-chain fatty acids stimulate motility of the K.Turner,G.Vandemeulebrouck,E.Varela,Y.Winogradsky,G.Zeller,J.Weissenbach, canine ileum. Am. J. Physiol. 253, G427–G433 (1987). S. D. Ehrlich, P. Bork, Enterotypes of the human gut microbiome. Nature 473, 174–180 29. P. Marteau, E. Cuillerier, S. Meance, M. F. Gerhardt, A. Myara, M. Bouvier, C. Bouley, F. Tondu, (2011). G. Bommelaer, J. C. Grimaud, Bifidobacterium animalis strain DN-173 010 shortens the 5. C. Palmer, E. M. Bik, D. B. DiGiulio, D. A. Relman, P. O. Brown, Development of the human colonic transit time in healthy women: A double-blind, randomized, controlled study. infant intestinal microbiota. PLoS Biol. 5, e177 (2007). Aliment. Pharmacol. Ther. 16,587–593 (2002). 6. J. E. Koenig, A. Spor, N. Scalfone, A. D. Fricker, J. Stombaugh, R. Knight, L. T. Angenent, 30. P. E. Squires, R. D. Rumsey, C. A. Edwards, N. W. Read, Effect of short-chain fatty acids on R. E. Ley, Succession of microbial consortia in the developing infant gut microbiome. contractile activity and fluid flow in rat colon in vitro. Am. J. Physiol. 262,G813–G817 Proc. Natl. Acad. Sci. U.S.A. 108 (Suppl. 1), 4578–4585 (2011). (1992). 7. M. G. Dominguez-Bello, E. K. Costello, M. Contreras, M. Magris, G. Hidalgo, N. Fierer, 31. S. H. Al-Lahham, M. P. Peppelenbosch, H. Roelofsen, R. J. Vonk, K. Venema, Biological R. Knight, Delivery mode shapes the acquisition and structure of the initial microbiota effects of propionic acid in humans; metabolism, potential applications and underlying across multiple body habitats in newborns. Proc. Natl. Acad. Sci. U.S.A. 107, 11971–11975 mechanisms. Biochim. Biophys. Acta 1801, 1175–1183 (2010). (2010). 32. P. Veiga, C. A. Gallini, C. Beal, M. Michaud, M. L. Delaney, A. DuBois, A. Khlebnikov, 8. J. J. Faith, F. E. Rey, D. O’Donnell, M. Karlsson, N. P. McNulty, G. Kallstrom, A. L. Goodman, J.E.vanHylckamaVlieg,S.Punit,J.N.Glickman,A.Onderdonk,L.H.Glimcher,W.S.Garrett, Downloaded from J. I. Gordon, Creating and characterizing communities of human gut microbes in Bifidobacterium animalis subsp. lactis fermented milk product reduces inflammation by gnotobiotic mice. ISME J. 4, 1094–1098 (2010). altering a niche for colitogenic microbes. Proc. Natl. Acad. Sci. U.S.A. 107, 18132–18137 9.J.G.Caporaso,J.Kuczynski,J.Stombaugh,K.Bittinger,F.D.Bushman,E.K.Costello, (2010). N.Fierer,A.G.Peña,J.K.Goodrich,J.I.Gordon,G.A.Huttley,S.T.Kelley,D.Knights, 33. A. C. Heath, W. Howells, K. K. Bucholz, A. L. Glowinski, E. C. Nelson, P. A. Madden, Ascertain- J. E. Koenig, R. E. Ley, C. A. Lozupone, D. McDonald, B. D. Muegge, M. Pirrung, J. Reeder, ment of a mid-western US female adolescent twin cohort for alcohol studies: Assessment of J.R.Sevinsky,P.J.Turnbaugh,W.A.Walters,J.Widmann,T.Yatsunenko,J.Zaneveld, sample representativeness using birth record data. Twin Res. 5,107–112 (2002). R. Knight, QIIME allows analysis of high-throughput community sequencing data. 34. A. L. Goodman, N. P. McNulty, Y. Zhao, D. Leip, R. D. Mitra, C. A. Lozupone, R. Knight,

Nat. Methods 7, 335–336 (2010). J. I. Gordon, Identifying genetic determinants needed to establish a human gut symbiont http://stm.sciencemag.org/ 10. M. N. Price, P. S. Dehal, A. P. Arkin, FastTree: Computing large minimum evolution trees in its habitat. Cell Host Microbe 6, 279–289 (2009). with profiles instead of a distance matrix. Mol. Biol. Evol. 26, 1641–1650 (2009). 35. Acknowledgments: We thank J. Manchester and J. Hoisington-López for assistance with 11.C.Lozupone,M.Hamady,R.Knight,UniFrac—An online tool for comparing microbial DNA sequencing; M. Karlsson, D. O’Donnell, and S. Wagoner for help with gnotobiotic community diversity in a phylogenetic context. BMC Bioinformatics 7, 371 (2006). mouse husbandry; S. Deng for assistance in preparing Illumina DNA libraries; S. Marion 12. M. Kanehisa, S. Goto, KEGG: Kyoto Encyclopedia of Genes and Genomes. Nucleic Acids Res. and D. Hooper for their contributions to the human study; D. Carlsen for coordination 28,27–30 (2000). of FMP production and logistics; S. Baumann and S. Fischer (Agilent Corp.) for providing 13. M. Kanehisa, S. Goto, M. Hattori, K. F. Aoki-Kinoshita, M. Itoh, S. Kawashima, T. Katayama, the Fiehn GC/MS Metabolomics RTL library used for metabolomics analyses; members of M. Araki, M. Hirakawa, From genomics to chemical genomics: New developments in KEGG. the Gordon lab for valuable suggestions during the course of this work; and G. Denariaz Nucleic Acids Res. 34, D354–D357 (2006). for his continued support. We are also grateful to Integrated Genomics for generating the 14. M. Kanehisa, S. Goto, M. Furumichi, M. Tanabe, M. Hirakawa, KEGG for representation and draft genome sequences of B. animalis subsp. lactis (CNCM I-2494) and S. thermophilus analysis of molecular networks involving diseases and drugs. Nucleic Acids Res. 38, (CNCM I-1630). Funding: Supported by grants from the NIH (DK30292, DK70977, and

D355–D360 (2010). DK078669) and Danone Research. Maintenance of the MOAFTS twin cohort is supported on March 29, 2016 15.P.J.Turnbaugh,C.Quince,J.J.Faith,A.C.McHardy,T.Yatsunenko,F.Niazi,J.Affourtit, by NIH grants AA09022, AA11998, AA17915, and HD49024. Author contributions: N.P.M., M. Egholm, B. Henrissat, R. Knight, J. I. Gordon, Organismal, genetic, and transcriptional T.Y., A.C.H., and J.I.G. designed the experiments; N.P.M., A.H., J.J.F., A.L.G., J.R.B., and M.J. variation in the deeply sequenced gut microbiomes of identical twins. Proc. Natl. Acad. M. performed experiments involving mice, whereas T.Y., J.J.F., R.O., S.C.-P., and G.G. con- Sci. U.S.A. 107, 7503–7508 (2010). ducted analyses of human biospecimens; N.P.M., T.Y., A.H., B.D.M., A.L.G., B.H., C.C., D.K., C.A.L., 16.B.L.Cantarel,P.M.Coutinho,C.Rancurel,T.Bernard,V.Lombard,B.Henrissat,The R.K., A.E.D., and C.B.N. analyzed the data; N.P.M., A.H., T.Y., and J.I.G. wrote the paper. Competing Carbohydrate-Active EnZymes database (CAZy): An expert resource for glycogenomics. interests: The authors declare that they have no competing interests. Accession numbers: Nucleic Acids Res. 37, D233–D238 (2009). ThecompletegenomesequenceofBifidobacterium animalis subsp. lactis (CNCM I-2494) has 17.J.J.Faith,N.P.McNulty,F.E.Rey,J.I.Gordon,Predictingahumangutmicrobiota’s been deposited at GenBank under accession number CP002915. The Whole Genome Shotgun response to diet in gnotobiotic mice. Science 333,101–104 (2011). projects for Lactobacillus delbrueckii subsp. bulgaricus (CNCM I-1632, CNCM I-1519), Streptococ- 18. O. Gilad, S. Jacobsen, B. Stuer-Lauridsen, M. B. Pedersen, C. Garrigues, B. Svensson, cus thermophilis CNCM I-1630, and Lactococcus lactis subsp. cremoris CNCM I-1631 have been Combined transcriptome and proteome analysis of Bifidobacterium animalis subsp. deposited at DDBJ/EMBL/GenBank under accession numbers AGFO00000000, AGHW00000000, lactis BB-12 grown on xylo-oligosaccharides and a model of their utilization. Appl. AGFN00000000, and AGHX00000000, respectively. The version of each of these genomes Environ. Microbiol. 76, 7285–7291 (2010). described in this paper is the first version (CP002915.1, AGFO01000000, AGHW01000000, 19.C.K.Hsu,J.W.Liao,Y.C.Chung,C.P.Hsieh,Y.C.Chan,Xylooligosaccharidesand AGFN01000000, and AGHX01000000, respectively). COPRO-Seq and RNA-Seq data are fructooligosaccharides affect the intestinal microbiota and precancerous colonic lesion deposited in the Gene Expression Omnibus (GEO) database (accession numbers GSE31943 development in rats. J. Nutr. 134, 1523–1528 (2004). and GSE31670, respectively), whereas 16S rRNA pyrosequencing reads and shotgun 20. M. Okazaki, S. Fujikawa, N. Matsumoto, Effect of xylooligosaccharide on the growth of pyrosequencing reads of human fecal community DNA are deposited in MG-RAST (accession bifidobacteria. Bifidobacteria Microflora 9,77–86 (1990). numbers qiime:803 and 4473933-4473980, respectively). 21. E. Kristiansson, P. Hugenholtz, D. Dalevi, ShotgunFunctionalizeR: An R-package for functional comparison of metagenomes. Bioinformatics 25,2737–2738 (2009). Submitted 24 May 2011 22. RDC Team, R: A Language and Environment for Statistical Computing (R Foundation for Accepted 7 October 2011 Statistical Computing, Vienna, Austria, 2009). Published 26 October 2011 23. D. Knights, E. K. Costello, R. Knight, Supervised classification of human microbiota. 10.1126/scitranslmed.3002701 FEMS Microbiol. Rev. 35, 343–359 (2011). 24. S. E. Stein, An integrated method for spectrum extraction and compound identification Citation: N. P. McNulty, T. Yatsunenko, A. Hsiao, J. J. Faith, B. D. Muegge, A. L. Goodman, from gas chromatography/mass spectrometry data. J. Am. Soc. Mass Spectrom. 10, 770–781 B. Henrissat, R. Oozeer, S. Cools-Portier, G. Gobert, C. Chervaux, D. Knights, C. A. Lozupone, (1999). R. Knight, A. E. Duncan, J. R. Bain, M. J. Muehlbauer, C. B. Newgard, A. C. Heath, J. I. Gordon, 25. R. E. Ley, P. J. Turnbaugh, S. Klein, J. I. Gordon, Microbial ecology: Human gut microbes The impact of a consortium of fermented milk strains on the gut microbiome of gnotobiotic associated with obesity. Nature 444, 1022–1023 (2006). mice and monozygotic twins. Sci. Transl. Med. 3, 106ra106 (2011).

www.ScienceTranslationalMedicine.org 26 October 2011 Vol 3 Issue 106 106ra106 14 The Impact of a Consortium of Fermented Milk Strains on the Gut Microbiome of Gnotobiotic Mice and Monozygotic Twins Nathan P. McNulty, Tanya Yatsunenko, Ansel Hsiao, Jeremiah J. Faith, Brian D. Muegge, Andrew L. Goodman, Bernard Henrissat, Raish Oozeer, Stéphanie Cools-Portier, Guillaume Gobert, Christian Chervaux, Dan Knights, Catherine A. Lozupone, Rob Knight, Alexis E. Duncan, James R. Bain, Michael J. Muehlbauer, Christopher B. Newgard, Andrew C. Heath and Jeffrey I. Gordon (October 26, 2011) Science Translational Medicine 3 (106), 106ra106. [doi: 10.1126/scitranslmed.3002701] Downloaded from http://stm.sciencemag.org/ on March 29, 2016

Science Translational Medicine (print ISSN 1946-6234; online ISSN 1946-6242) is published weekly, except the last week in December, by the American Association for the Advancement of Science, 1200 New York Avenue, NW, Washington, DC 20005. Copyright 2016 by the American Association for the Advancement of Science; all rights reserved. The title Science Translational Medicine is a registered trademark of AAAS. Article Tools Visit the online version of this article to access the personalization and article tools: http://stm.sciencemag.org/content/3/106/106ra106 Supplemental "Supplementary Materials" Materials http://stm.sciencemag.org/content/suppl/2011/10/24/3.106.106ra106.DC1 Related Content The editors suggest related resources on Science's sites: http://stm.sciencemag.org/content/scitransmed/6/220/220ra11.full http://science.sciencemag.org/content/sci/346/6210/687.full http://stm.sciencemag.org/content/scitransmed/7/286/286ra68.full http://science.sciencemag.org/content/sci/351/6275/802.full http://science.sciencemag.org/content/sci/351/6275/aad3311.full http://science.sciencemag.org/content/sci/351/6275/854.full Permissions Obtain information about reproducing this article: http://www.sciencemag.org/about/permissions.dtl Downloaded from http://stm.sciencemag.org/ on March 29, 2016

Science Translational Medicine (print ISSN 1946-6234; online ISSN 1946-6242) is published weekly, except the last week in December, by the American Association for the Advancement of Science, 1200 New York Avenue, NW, Washington, DC 20005. Copyright 2016 by the American Association for the Advancement of Science; all rights reserved. The title Science Translational Medicine is a registered trademark of AAAS. Editor's Summary

A Yogurt a Day…

We all enjoy a tasty yogurt and believe that the bacterial species contained in this type of fermented milk product will keep us healthy. But how much influence do the microbes in these products have on our gut microbiomes and consequently our health, and are these effects generalizable to different human populations consuming different diets? These questions are of concern to regulatory agencies who are increasing pressure on manufacturers to validate the health claims of various foods, including yogurts. McNulty and his colleagues, in an exciting new study, describe a way to evaluate their effects on the human gut microbiome. First, they studied the effects of consuming a popular yogurt on the gut microbiomes of seven healthy adult female identical twin pairs. The bacterial and gene composition, as well as the gene expression patterns, of their gut microbial communities were analyzed before, during, and after

consumption of the yogurt. These results were compared to those obtained in gnotobiotic mice that were Downloaded from first reared under conditions where the only microbes they harbored were 15 prominent, sequenced human gut bacterial symbionts, after which time they were exposed to the same 5 bacterial strains as those contained in the yogurt.

McNulty and colleagues found during repeated sampling of the gut microbiomes of the twins over a 4-month period that the species and gene content of their gut microbial communities remained stable and were not appreciably perturbed by consuming the yogurt. After exposure of the humanized mice to the five bacterial strains in the fermented milk product, the researchers showed that the mice did http://stm.sciencemag.org/ not exhibit marked changes in the proportional representation of their human symbiotic bacterial species or genes, mirroring the results seen in the twins. However, analysis of gut bacterial gene expression profiles and of urinary metabolites in these mice disclosed that introducing the fermented milk product strains resulted in marked changes in a number of metabolic pathways, most prominently those related to carbohydrate processing. These latter findings helped direct follow-up studies of the twins' gut samples where they found similar changes in metabolism as those observed in mice.

These findings show that mice containing a sequenced model human gut microbiome can serve as part of a preclinical discovery pipeline designed to identify the effects of existing or new bacterial

species with purported health benefits on the properties of the human gut microbiome. Although it on March 29, 2016 remains unclear whether eating a yogurt a day will keep the doctor away, the study by McNulty and his colleagues paves the way for future work to analyze in more detail the direct effects of consuming foods containing bacterial species with potential health benefits on the gut microbiomes of various human populations.

The following resources related to this article are available online at http://stm.sciencemag.org. This information is current as of March 29, 2016.

Science Translational Medicine (print ISSN 1946-6234; online ISSN 1946-6242) is published weekly, except the last week in December, by the American Association for the Advancement of Science, 1200 New York Avenue, NW, Washington, DC 20005. Copyright 2016 by the American Association for the Advancement of Science; all rights reserved. The title Science Translational Medicine is a registered trademark of AAAS.