A hybrid two-component system protein of aprominent human gut symbiont couples glycansensing in vivo to carbohydrate metabolismErica D. Sonnenburg, Justin L. Sonnenburg, Jill K. Manchester, Elizabeth E. Hansen, Herbert C. Chiang,and Jeffrey I. Gordon*

Center for Genome Sciences, Washington University School of Medicine, St. Louis, MO 63108

Contributed by Jeffrey I. Gordon, April 21, 2006

Bacteroides thetaiotaomicron is a prominent member of our nor-mal adult intestinal microbial community and a useful model forstudying the foundations of human–bacterial mutualism in ourdensely populated distal gut microbiota. A central question is howmembers of this microbiota sense nutrients and implement anappropriate metabolic response. B. thetaiotaomicron contains alarge number of glycoside hydrolases not represented in our ownproteome, plus a markedly expanded collection of hybrid two-component system (HTCS) proteins that incorporate all domainsfound in classical two-component environmental sensors into onepolypeptide. To understand the role of HTCS in nutrient sensing,we used B. thetaiotaomicron GeneChips to characterize their ex-pression in gnotobiotic mice consuming polysaccharide-rich or-deficient diets. One HTCS, BT3172, was selected for further anal-ysis because it is induced in vivo by polysaccharides, and itsabsence reduces B. thetaiotaomicron fitness in polysaccharide-richdiet-fed mice. Functional genomic and biochemical analyses of WTand BT3172-deficient strains in vivo and in vitro disclosed that�-mannosides induce BT3172 expression, which in turn inducesexpression of secreted �-mannosidases. Yeast two-hybrid screensrevealed that the cytoplasmic portion of BT3172’s sensor domainserves as a scaffold for recruiting glucose-6-phosphate isomeraseand dehydrogenase. These interactions are a unique feature ofBT3172 and specific for the cytoplasmic face of its sensor domain.Loss of BT3172 reduces glycolytic pathway activity in vitro and invivo. Thus, this HTCS functions as a metabolic reaction center,coupling nutrient sensing to dynamic regulation of monosaccha-ride metabolism. An expanded repertoire of HTCS proteins withdiversified sensor domains may be one reason for B. thetaiotaomi-cron’s success in our intestinal ecosystem.

Bacteroides thetaiotaomicron � glycoside hydrolases � gut microbialecology � metabolic regulation � signal transduction

The adult human gut is home to tens of trillions of microbes.This community is composed primarily of members of the

domain Bacteria but also contains representatives from Archaeaand Eukarya (1). Most bacteria live in the colon where, amongtheir other functions, they extract energy from dietary polysac-charides that would otherwise be lost because our humangenome does not encode the requisite glycoside hydrolases andpolysaccharide lyases needed for cleaving their glycosidic link-ages (2, 3). Intestinal absorption of the short-chain fatty acid endproducts of microbial polysaccharide fermentation accounts forup to 10% of our daily harvest of calories (4). Two divisions ofbacteria, the Bacteroidetes which include the saccharolytic Bac-teroides, and the Firmicutes, account for �99% of all phyloge-netic types (phylotypes) identified in the colon (5).

Several recent developments have set the stage for definingthe structure and operations of this community. Comprehensive16S rRNA sequence-based enumerations combined with newcomputational methods are allowing community membership tobe defined and compared between healthy individuals (5, 6).

Sequencing individual representatives of the major divisions ofbacteria as well as whole community DNA is revealing the genecontent of the gut ‘‘microbiome’’ and will allow in silico predic-tions to be made of the microbiota’s metabolic potential (7).Germ-free (GF) mice, colonized with one or more sequencedrepresentatives of the human gut microbiota, are providinginsights about the transcriptional and metabolic networks ex-pressed by community members in their habitats (3).

One important question about the operation of this microbialcommunity is how its members respond to changes in theirnutrient environments. Subsumed under this question is the issueof how these microbes sense the nutrient landscape of their guthabitats and adapt to changes in resource availability. Theanswers could help us to attribute niches (professions) to com-munity members, understand how a microbiota is selected,elucidate how community robustness is maintained, and developstrategies for intentionally manipulating the operations of thegut microbiota so as to provide additional benefits to humanhosts. One benefit could be a new era of personalized nutritionwhere diet is matched to the nutrient processing capacity of anindividual’s microbiota.

We have used Bacteroides thetaiotaomicron as a model gutsymbiont to address some of these issues. In the most compre-hensive 16S rRNA enumeration study of the human gut micro-biota published to date, this Gram-negative obligate anaerobecomprised 12% of all Bacteroidetes, and 6% of all colonicbacteria in three healthy adults (5). The B. thetaiotaomicrongenome sequence (8) revealed an unprecedented expansion ofgenes dedicated to polysaccharide metabolism: its ‘‘glycobiome’’(genes involved in carbohydrate acquisition and processing) in-cludes an arsenal of 226 glycoside hydrolases and 15 polysaccharidelyases capable of cleaving many of the glycosidic linkages present inplant glycans that are not digestible by humans (3, 8, 9).

The B. thetaiotaomicron genome also contains a sizeablecollection of genes involved in environmental sensing and signaltransduction. This collection includes 54 sensor kinases and 29response regulators (8) that form ‘‘classical’’ two-componentsystems, plus a family of 32 unique proteins that incorporate allof the domains found in classical two-component systems into asingle polypeptide. Classical two-component systems transducesignals from the cell’s outside environment (pH, ions, and

Conflict of interest statement: No conflicts declared.

Freely available online through the PNAS open access option.

Abbreviations: GF, germ-free; HTCS, hybrid two-component system; MM, minimal medium;MM-G, MM with 0.5% glucose; PR, polysaccharide-rich; CPS, capsular polysaccharide syn-thesis; TYG, tryptone�yeast extract�glucose; GS, 35% glucose�35% sucrose.

Data deposition: The GeneChip data sets reported in this paper have been depositedin the Gene Expression Omnibus database, www.ncbi.nlm.nih.gov�geo� (accessionnos. GSM40917–GSM40926, GSM40886–GSM40888, GSM40897–GSM40899, GSM40902–GSM40906, and GSE4703).

*To whom correspondence should be addressed. E-mail: jgordon@wustl.edu.

© 2006 by The National Academy of Sciences of the USA

8834–8839 � PNAS � June 6, 2006 � vol. 103 � no. 23 www.pnas.org�cgi�doi�10.1073�pnas.0603249103

Dow

nloa

ded

by g

uest

on

May

22,

202

1

metabolites) to its genome via a multistep phosphorelay. Theprototypic two-component system consists of (i) a membrane-localized sensor protein containing histidine kinase and phos-phoacceptor domains (annotated as HATPase�c and HisKA inthe Pfam database; http:��pfam.wustl.edu�), and (ii) a cytoplas-mic response regulator protein containing a response regulatorreceiver domain (Response�reg) and a signal output domain(10). The majority of response regulators are transcriptionfactors that contain one of three types of DNA-binding domains(trans�reg�c, GerE, or HTH�8). B. thetaiotaomicron’s 32 ‘‘hybridtwo-component system’’ (HTCS) proteins are composed of anN-terminal periplasmic sensor that is interrupted by up to fivepredicted transmembrane segments, plus four conserved cyto-plasmic domains [HisKA, HATPase�c, Response�reg, plus ahelix–turn–helix putative DNA-binding domain (HTH�AraC)that is distinct from the HTH�8 domain common to classicalresponse regulators]. The sensor domains of HTCS proteins areless highly conserved among family members than their intra-cellular signaling domains (19% vs. 32% average amino acidsequence identity, respectively), suggesting that HTCS proteinshave diversified to respond to distinct signals but have conservedtheir method of intracellular signal transduction.

B. thetaiotaomicron has more HTCS proteins than any othersequenced prokaryote: since their initial identification in the B.thetaiotaomicron genome, HTCS family members have beenidentified in other Bacteroidetes, Proteobacteria, and Chlo-roflexi (ref. 11 and Table 1, which is published as supportinginformation on the PNAS web site). Intriguingly, 17 of B.thetaiotaomicron’s HTCS genes are located adjacent to genesencoding paralogs of SusC�SusD polysaccharide-binding pro-teins, or glycoside hydrolases (8). Based on these findings, wehave hypothesized that evolution of a large collection of HTCSparalogs may be a salient feature of the adaptation of B.thetaiotaomicron (and other Bacteroidetes) to their gut habitats(addresses) and niches (12). Below, we present evidence thatsupports this hypothesis.

Results and DiscussionBT3172 Expression Improves the Fitness of B. thetaiotaomicron in theCeca of Mice Fed a Polysaccharide-Rich (PR) Diet. We used customAffymetrix GeneChips containing probesets that recognize98.7% of B. thetaiotaomicron’s 4,779 known or predicted chro-mosomal protein-coding genes to characterize HTCS gene ex-pression in three nutrient environments: (i) during growth fromearly log to stationary phase in a batch culture fermentercontaining a minimal medium (MM) with 0.5% glucose(MM-G) as the sole fermentable carbon source; (ii) in the cecaof adult male GF mice belonging to the NMRI inbred line thatwere fed a standard autoclaved PR rodent chow diet andcolonized with B. thetaiotaomicron for 10 days (a period that en-compasses 2–3 rounds of turnover of the gut epithelium and itsoverlying mucus layer); and (iii) in the ceca of GF mice that wereswitched from a PR diet to a simple sugar diet (35% glucose�35% sucrose; GS) devoid of polysaccharides for 14 days beforebeing colonized with B. thetaiotaomicron for 10 days (n � 3–5mice per treatment group; each cecal sample was analyzedindependently with a GeneChip). The cecum was selectedbecause this anatomically distinct region, located between thesmall intestine and colon, contains a large bacterial populationthat is readily harvested. We have shown that the density ofcolonization of the ceca of mice in groups ii and iii is notsignificantly different (1011–1012 colony forming units�ml lumi-nal contents), and that removing polysaccharides from the dietcauses B. thetaiotaomicron to redirect its glycan foraging to themucus overlying the cecal epithelium, with attendant inductionof genes that degrade host-derived glycans (e.g., sialidases,fucosidases, �-hexosaminidases, and a mucin-desulfating sulfa-tase; see ref. 3)

Our GeneChip analysis revealed that 18 of the 32 HTCS familymembers have significantly different levels of expression amongthe three growth conditions tested [P � 0.01; empirical falsediscovery rate, 0%]. These differentially regulated HTCS genescluster into four groups: (i) highest expression in MM-G (twoHTCS); (ii) highest expression in the ceca of mice on the GS diet(three); (iii) highest expression in the ceca of animals on the PRdiet (ten); and (iv) highest expression in mice on both dietscompared with MM-G (three) (Fig. 1A). The distal human gut, likethe ceca of our gnotobiotic mice, normally provides an environmentrich in polysaccharides for B. thetaiotaomicron. Therefore, we choseBT3172 for further analysis because it represents the group ofHTCS genes that are most highly expressed in the ceca of mice onthe PR diet (10-fold higher levels than in MM-G at mid-log phase;2-fold higher compared with GS diet-treated animals; P � 0.01;Fig. 1A).

A mixture of 108 colony forming units of WT B. thetaiotaomi-cron and 108 colony forming units of an isogenic mutant con-taining a BT3172-null allele (BT3172::pGERM; erythromycinresistant) was introduced by gavage into 10 12-week-old GFNMRI male recipients. Five animals were maintained on a PRdiet. The second group had been switched from PR to GS chow14 days before colonization. Both groups of gnotobiotic micewere killed 10 days after gavage, and the number of viable WTvs. mutant cells in their cecal contents was defined by selectiveplating. In the PR diet-fed group, WT B. thetaiotaomicronachieved a statistically significant 40-fold higher density than themutant (P � 0.01 using Student’s t test), whereas in the GS-fedgroup, both strains colonized the cecum at equivalent levels (Fig.1B). Colonization of mice consuming one or the other diets withthe WT or the mutant strain alone resulted in equivalentnumbers of colony forming units (P � 0.2 among treatmentgroups). Control experiments confirmed that there was nodetectable reversion of the mutant to WT in vivo under eithernutrient condition (�10�6) as defined by an erythromycin-resistant phenotype, and that the fitness defect observed withBT3172::pGERM was not due to the insertion of the suicidevector (pGERM) itself [co-colonizations with WT B. thetaio-taomicron and an isogenic mutant strain lacking an HTCS whoseexpression is similar in mice fed PR and GS diets(BT4178::pGERM) produced equivalent levels of colonizationof both strains in the ceca of PR diet-treated animals (data notshown)]. Based on these results, we concluded that BT3172 plays arole in successful B. thetaiotaomicron colonization of its cecalhabitat when organisms are competing for dietary polysaccharides.

Loss of BT3172 Affects �-Mannosidase Gene Expression. To obtain acomprehensive view of the impact of BT3172 deficiency on theB. thetaiotaomicron transcriptome, we performed GeneChipprofiling of WT and BT3172::pGERM harvested from the cecaof GF mice that had been colonized for 10 days with either strainwhile consuming a PR or GS diet (n � 5 mice per treatmentgroup per strain). Hierarchical clustering of all genes revealedthat the BT3172::pGERM data set obtained from mice on thePR diet clustered more closely with the WT B. thetaiotaomicronPR diet data set than with the BT3172::pGERM GS data set(Fig. 5A, which is published as supporting information on thePNAS web site). These findings underscore the impact of diet onthe B. thetaiotaomicron transcriptome.

Comparison of WT and mutant B. thetaiotaomicron harvestedfrom the ceca of PR diet-treated animals revealed statisticallysignificant differences in the expression of 686 genes (14% of thegenome): 440 genes were up-regulated in BT3172::pGERMrelative to WT, whereas 246 genes were down-regulated [P �0.01; false discovery rate, �0.4%; see Fig. 5 B and C and Table2, which are published as supporting information on the PNASweb site, for a Clusters of Orthologous Genes (COG)-basedcategorization of regulated genes]. The absence of a BT3172

Sonnenburg et al. PNAS � June 6, 2006 � vol. 103 � no. 23 � 8835

MIC

ROBI

OLO

GY

Dow

nloa

ded

by g

uest

on

May

22,

202

1

transcript did not produce statistically significant difference inthe expression of any other HTCS family member. COG M (cellwall�membrane biogenesis, includes genes involved in capsularpolysaccharide biosynthesis) and COG G (carbohydrate trans-port and metabolism, includes glycoside hydrolases) representedthe dominant functional categories of responses to BT3172deficiency.

Mannans (polymannose containing glycans) and mannosides(D-mannose connected to other sugars by various glycosidiclinkages) are abundantly represented in dietary plant glycans

(mainly as �-linked mannose residues) and in fungal and mam-malian glycans (primarily as �-linked mannose). Our genomecontains no mannanases and 16 mannosidases (15 �-mannosi-dases and 1 �-mannosidase). The carbohydrate-active enzymes(CAZy) database (www.cazy.org; see ref. 2) divides B. thetaiotaomi-cron’s 241 glycoside hydrolases and polysaccharide lyases into 45families. B. thetaiotaomicron is well equipped to liberate both �- and�-anomers of mannose: it possesses 25 �-mannosidases, 10 �-man-nanases, and 5 �-mannosidases.

Glycoside hydrolase family 92 was most affected by loss ofBT3172 (see Fig. 6A, which is published as supporting informa-tion on the PNAS web site). This family contains 23 members,all of which are predicted to be �-mannosidases capable ofcleaving disaccharides or more complex glycans with terminalmannose residues. Loss of BT3172 significantly reduced expres-sion of five family 92 �-mannosidases in vivo: two of the genesencoding these enzymes are in a single operon (BT3962 andBT3963; see Fig. 6 B and C). Loss of BT3172 did not affectexpression of B. thetaiotaomicron’s 10 mannanases.

The Effect of BT3172 Deficiency on Capsular Polysaccharide LocusGene Expression in Vivo Indicates a Defect in Sensing Polysaccharides.The B. thetaiotaomicron genome contains eight capsular poly-saccharide synthesis (CPS) loci, each composed of genes in-volved in the decoration of the microbe’s surface with polysac-charides. Elegant work by Comstock and coworkers (13) hasshown that by regulating CPS locus expression Bacteroides fragiliscan match its surface glycan features with the glycan landscapeof its habitat. This molecular mimicry may help Bacteroidesspecies evade a host immune response.

Transcription profiling of WT B. thetaiotaomicron revealedthat CPS3 (BT0595–BT0614) is normally expressed at higherlevels than the other CPS loci when polysaccharides are scarce(i.e., during growth in MM-G) and is down-regulated when cellsare (i) exposed to tryptone�yeast extract�glucose (TYG) me-dium, which is rich in �-linked mannans as well as otherSaccharomyces cerevisiae-derived polysaccharides, or (ii) residein the ceca of mice consuming a PR diet. In contrast, CPS4(BT1336–BT1358) is selectively up-regulated under conditionsin which polysaccharides are abundant (i.e., in the ceca of micefed a PR diet and in TYG medium) and is down-regulated inMM-G (ref. 3 and data not shown).

B. thetaiotaomicron containing the BT3172::pGERM allelehas the opposite pattern of expression of CPS3: it up-regulatesthis locus relative to WT during growth in TYG (a 3.1-foldincrease in the aggregate signal produced by all probesets thatrecognize the products of CPS3; P � 0.001) and in the ceca ofmice fed a PR diet (a 4.6-fold increase in the aggregate signal;P � 0.001; see Fig. 7, which is published as supporting infor-mation on the PNAS web site). This aberrant pattern of CPS3expression suggests that removal of BT3172 causes B. thetaio-taomicron to ‘‘misinterpret’’ its nutrient landscape as beingpolysaccharide-deficient when it is replete with glycans.

Loss of BT3172 Produces a Change in Prioritized Utilization of Glycans.GeneChip profiling of WT B. thetaiotaomicron during growthfrom log to stationary phase in TYG medium disclosed aprioritized pattern of glycan utilization; it first expresses glyco-side hydrolases that liberate fructose from fructans, followed byenzymes that degrade mannans, then starch, and finally hex-osamine-containing glycans (Fig. 2). These findings indicate thatWT B. thetaiotaomicron prioritizes expression of genes involvedin the acquisition of carbohydrates that are most easily shuntedinto the glycolytic pathway (i.e., fructose and mannose). Con-sistent with this notion, the WT strain up-regulates genesinvolved in glycolysis in the early log phase and down-regulatesthese genes as it approaches stationary phase (Fig. 3A).

BT3172::pGERM shows a similar pattern of nutrient utiliza-

Fig. 1. Effects of nutrient environment on expression of B. thetaiotaomicrongenes encoding hybrid two-component proteins in vitro and in the ceca ofgnotobiotic mice. (A) GeneChip profiling of HTCS gene expression duringgrowth of WT B. thetaiotaomicron in a batch culture fermenter containingMM-G or collected from the cecal contents of monoassociated mice fed a GSdiet or a PR diet (n � 3 and 5 animals, respectively; each analyzed separately).Cells were harvested at five time points during growth from log to stationaryphase in MM-G (n � 2 samples per time point). Four patterns of HTCSexpression are observed; those that are either most highly expressed in MM-G,in vivo on a GS diet, in vivo on a PR diet, or in vivo on both diets compared withMM-G. Colors indicate standard deviation above (red) and below (green) themean level of expression (black) of a gene across all conditions. (B) Cocoloni-zation of GF mice gavaged once with equivalent numbers of isogenic WT andBT3172::pGERM strains and killed 10 days later. Loss of BT3172 reduces levelsof cecal colonization in animals fed a PR but not a GS diet. Mean values � SEMare plotted. **, P � 0.01.

8836 � www.pnas.org�cgi�doi�10.1073�pnas.0603249103 Sonnenburg et al.

Dow

nloa

ded

by g

uest

on

May

22,

202

1

tion in TYG medium, with the exception that an operon(BT3962–BT3965), containing three CAZy family 92 �-manno-sidases, is up-regulated during the early log phase rather than thelate log�stationary phase (Fig. 2). The lack of a discerniblegrowth defect of the mutant in vitro and in monoassociationexperiments contrasts with its reduced fitness in the face of com-petition with WT in vivo (Fig. 1A). These data are consistent withthe notion that there is functional redundancy in B. thetaiotaomi-cron’s nutrient sensing and acquisition machinery that effectivelycompensates for loss of BT3172 except in highly competitivesituations.

The mutant exhibits a constant level of expression of glycolyticgenes so that compared with WT, their expression is lower in theearly log phase and higher in the late log and stationary phase(Fig. 3A). Biochemical assays disclosed that glucose 6-phosphateand fructose 6-phosphate levels are significantly higher in themutant compared with the WT strain during mid-log-phasegrowth in TYG medium and in the ceca of mice fed a PR diet(Fig. 3B), consistent with reduced glycolytic pathway activity inthe absence of BT3172. (The presence of these metabolites incecal contents could be attributed to B. thetaiotaomicron because

they were below the limit of detection in cecal contents harvestedfrom GF controls fed the same diet; data not shown).

Mannose enters the glycolytic pathway via the action of twoenzymes, a hexokinase (BT2430), which phosphorylates man-nose to mannose 6-phosphate, and mannose 6-phosphateisomerase (BT0373), which converts mannose 6-phosphate tofructose 6-phosphate. GeneChip profiling of BT3172::pGERMand WT B. thetaiotaomicron disclosed no statistically significantdifferences in the expression of genes encoding either enzyme, bothin vitro and in vivo. However, mannose 6-phosphate levels weresignificantly elevated in the mutant strain, in vitro (mid-log-phasecells) and in vivo (Fig. 3B). This elevation is consistent with theobserved increase in fructose 6-phosphate and�or reduction inglycolysis.

BT3172 Expression Is Activated by Mannosides but Not by MonomericD-Mannose. These findings indicated that loss of BT3172 disruptsB. thetaiotaomicron’s pattern of mannoside utilization and raisedthe question of whether expression of this HTCS is regulated bymannosides. Therefore, we used quantitative RT-PCR (seeTable 3, which is published as supporting information on thePNAS web site, for a list of all primers used for quantitativeRT-PCR) to measure BT3172 mRNA levels during mid-log-phase growth in MM containing yeast extract (rich in �-linkedmannans) or 0.5% mannan. BT3172 expression is induced tosimilar degrees in MM–yeast extract and MM–yeast mannancompared with MM-G (4.5 � 0.5- and 5.8 � 1.1-fold). Although,MM containing 0.5% D-mannose was not able to induce BT3172,MM plus 0.5% �1,3-mannobiose up-regulated BT3172 4.0 �

Fig. 2. The prioritized pattern of glycan utilization by B. thetaiotaomicronis disrupted by BT3172::pGERM. GeneChip analysis of glycoside hydrolase-containing operons that display differential expression in WT B. thetaiotaomi-cron during growth from log to stationary phase in rich (TYG) medium isshown. The orderly pattern of induction and suppression of expression pro-vides insights about the prioritization of glycan consumption: fructans areconsumed initially, followed by mannans, starch, and finally hexosamine-containing glycans. Note that an operon involved in the degradation ofmannose-containing glycans and import of liberated mannose (BT3962–BT3965) is induced prematurely during early log phase in the BT3172::pGERMstrain.

Fig. 3. The absence of BT3172 impairs glycolysis. (A) GeneChip analysis of theexpression of genes in the glycolytic pathway during growth of isogenic WTand BT3172::pGERM strains in batch culture fermenters containing TYG me-dium. (B) Concentration of metabolites in WT and mutant B. thetaiotaomicronstrains harvested from the cecal contents of gnotobiotic mice fed a PR diet orduring mid-log phase growth in a fermenter containing TYG medium. *, P �0.05; **, P � 0.01

Sonnenburg et al. PNAS � June 6, 2006 � vol. 103 � no. 23 � 8837

MIC

ROBI

OLO

GY

Dow

nloa

ded

by g

uest

on

May

22,

202

1

0.6-fold above MM-G controls. Together, these data reveal that�-linked mannosides, but not monomeric mannose, modulatesBT3172 expression.

Comparison of mid-log-phase WT and BT3172:pGERMstrains grown in MM–�1,3-mannobiose disclosed that lossof BT3172 reduces expression of the two CAZy-group 92�-mannosidases contained in a single operon (BT3962 andBT3963) by 3.3 � 0.5- and 5.6 � 0.9-fold, respectively (P �0.001). These two genes also show reduced expression in the cecaof mice fed a PR diet when BT3172 is missing. Thus, our resultsreveal a positive feedback loop in which mannose present inglycosidic linkages induces BT3172 expression. BT3172, in turn,induces expression of the �-mannosidases that degrade thesemannosides. Glycosidic linkage-dependent regulation of geneexpression is known to occur in B. thetaiotaomicron’s starchutilization system (Sus) operon, which is induced by maltose andhigher order starches, but not by glucose alone (14). Linkage-specific activation is consistent with the idea that at least someB. thetaiotaomicron HTCS incorporate all of their input andoutput domains into one polypeptide so that they can sense veryspecific nutrients in their habitat and mobilize appropriatecellular transcriptional and metabolic responses without theneed for signal amplification. One way to coordinate suchresponses would be to assemble enzymes involved in processingthe products of HTCS-regulated glycoside hydrolases on theHTCS itself: i.e., the HTCS would serve as a scaffold for bringingenzymes and their substrates together to form a multicomponentmetabolic machine.

The Sensor Domain of BT3172 Interacts with Enzymes Involved inGlucose Metabolism. To search for protein partners of BT3172, weperformed a yeast two-hybrid screen using the N-terminal sensordomain (Met-1-Glu-745) of BT3172 as bait and the proteinproducts of a B. thetaiotaomicron cDNA library as preys. ThiscDNA library, containing 330,000 clones, was prepared from RNAisolated from B. thetaiotaomicron harvested during growth in TYGand MM-G at all of the time points shown in Figs. 1 and 2. ThecDNA library of preys was screened at 6-fold coverage by using thestrict selection criteria described in Supporting Text, which is pub-lished as supporting information on the PNAS web site.

Our screen yielded two cytoplasmic proteins that interact withBT3172’s sensor domain: (i) glucose 6-phosphate isomerase(BT2124, Gly-228-Gly-352), which produces fructose 6-phosphate, the substrate for phosphofructokinase (rate-limitingenzyme in glycolysis) and (ii) glucose 6-phosphate dehydroge-nase (BT1221, Thr-309-Gly-464), which catalyzes the firstcommitted step of the pentose phosphate pathway. A thirdinteracting protein was identified (BT0924, Trp-77-Gly-152) thatshares homology to ATP-binding cassette (ABC) transportersubstrate-binding proteins. BT0924 is part of an operon that alsoencodes an ABC transporter ATP-binding protein and an ABCtransporter permease. Loss of BT3172 does not affect theexpression of this operon in vitro or in vivo. Directed yeasttwo-hybrid screens revealed that BT0924 also interacts withthe sensor domains of two other HTCS proteins (BT026and BT3302). Based on these findings, we focused on oursubsequent analyses on the significance of the interactionsbetween BT3172 and the two cytoplasmic enzymes involved inglucose metabolism.

Although loss of BT3172 affects glycolytic activity, it does notappear to affect the pentose phosphate pathway. Enzymescatalyzing the first three steps in this pathway (glucose 6-phosphate 3 glucono 1,5-lactone 6-phosphate 3 6-phospho-gluconate 3 ribulose 5-phosphate) are all encoded by genescontained in a single operon (BT1220–BT1222). None of thesegenes are differentially expressed in the WT compared withBT3172::pGERM strains, in vitro or in vivo, nor are there

statistically significant differences in the levels of 6-phosphoglu-conate, an intermediate in this pathway (data not shown).

Interactions Between BT3172 and These Two Proteins Are Specific forThis HTCS and Limited to a Discrete Cytoplasmic Region of Its SensorDomain. Follow-up, directed, yeast two-hybrid assays establishedthat the BT3172 sensor domain interacted with the intactisomerase and dehydrogenase. BT2124 and BT1221 were sub-sequently screened against the sensor domains of 29 other HTCSproteins (Table 4, which is published as supporting informationon the PNAS web site). The results demonstrated that theinteractions between the isomerase and dehydrogenase werespecific for the BT3172 sensor domain.

To identify the region of BT3172’s sensor domain responsiblefor these protein–protein interactions, four truncation mutantswere constructed: Met-1-Asn-601, Met-1-Pro-451, Met-1-Ser-299, and Met-1-Lys-145. The isomerase and dehydrogenaseinteracted with Met-1-Ser-299 but not Met-1-Lys-145, implyingthat Asp-146-Ser-299 is the critical BT3172 domain. This con-clusion was further validated by replacement of Asp-146-Ser-299in BT3172 with the analogous region of another HTCS (Lys-145-Ser-298 of BT4137): the substitution resulted in the loss ofBT3172’s ability to interact with all three protein partners.

Glucose 6-phosphate isomerase and glucose 6-phosphate de-

Fig. 4. Model of BT3172-directed metabolic regulation. B. thetaiotaomicronhas 32 HTCS proteins that incorporate the elements of classical two-component signaling proteins [a sensor, phosphoacceptor (HisKA), histidinekinase (HATPase�c), response regulator receiver (Response�reg), and signaloutput domain] in one polypeptide. Structural diversity among HTCS proteinsis greatest among their sensor domains. This domain in BT3172 (Met-1-Glu-745) contains three predicted transmembrane spanning elements, producingperiplasmic- and cytoplasmic-facing elements. Expression of BT3172 is in-duced by mannosides; the HTCS, in turn, induces expression of �-mannosi-dases required for breakdown of mannosides. Asp-146-Ser-299 of the cyto-plasmic portion of BT3172’s sensor domain serves as a site for binding twoproteins: glucose 6-phosphate isomerase (EC 5.3.1.9; key enzyme in the gly-colytic pathway) and glucose 6-phosphate dehydrogenase (EC 1.1.1.49; cata-lyzes the first step in the pentose phosphate shunt). Loss of BT3172 diminishesglycolytic pathway activity in vivo and in vitro. The organization of BT3172 andits interacting partners links sensing the availability of carbohydrates in theenvironment with regulation of bacterial carbohydrate metabolism.

8838 � www.pnas.org�cgi�doi�10.1073�pnas.0603249103 Sonnenburg et al.

Dow

nloa

ded

by g

uest

on

May

22,

202

1

hydrogenase are cytoplasmic proteins. The program TMPRED(www.ch.embnet.org�software�TMPRED�form.html) predictsthree transmembrane segments in the sensor domain of BT3172(Leu-5-Ala-25, Val-393-Ala-41, and Thr-755-Ile-771), resultingin its subdivision into periplasmic and cytoplasmic components.The region of BT3172 sensor domain that interacts with theisomerase and dehydrogenase (Arg-146-Ser-299) is predicted to belocated on the cytoplasmic face of the inner membrane (Fig. 4).

Yeast two-hybrid screens using the predicted periplasmic faceof BT3172’s sensor domain (Gly-421-Glu-745) as bait did notidentify any interacting proteins. Screens using the entire intra-cellular portion of BT3172 (Lys-827-Ile-1310) or just its histidinekinase, phosphotransfer, response regulator, or helix-turn-helixdomains, also failed to identify any interacting proteins thatsatisfied our stringent selection criteria.

Prospectus. GeneChip studies of B. thetaiotaomicron grown invitro in MM-G and TYG media or harvested from the ceca ofmice fed a PR-diet, have shown that B. thetaiotaomicron con-stitutively expresses, at least at low levels, members of all of itsglycoside hydrolase families (secreted and cytoplasmic), as wellas a majority of its SusC�D paralogs (3). These products of theB. thetaiotaomicron glycobiome presumably allow the organismto constantly survey and import samples of glycans present in itsgut habitat into its periplasmic space where metabolic responsesare initiated. Our analysis of the operations of BT3172, sum-marized in Fig. 4, supports this view. The presence of mannosidesinduces BT3172 expression, either through their direct interac-tion with this HTCS or with other cellular signaling proteins.BT3172 induction by mannosides, in turn, induces expression ofsecreted �-mannosidases that aid in the harvest of the perceivednutrient. The cytoplasmic portion of BT3172’s sensor domainserves as a scaffold for recruiting two key enzymes that regulatemetabolism of monosaccharides generated from harvested car-bohydrates. Interaction with these enzymes is a unique charac-teristic of this B. thetaiotaomicron HTCS and specific for acommon region of the cytoplasmic face of its sensor domain.Regulation of these site-specific protein–protein interactionswould provide a means for B. thetaiotaomicron to dynamicallycouple carbohydrate availability with utilization. The role of thephosphorelay system incorporated into BT3172 in modulatingthese protein–protein interactions, or in communicating withother signaling molecules remains to be defined, as does the roleof its C-terminal helix–turn–helix (HTH�AraC) domain. Meta-bolic intermediates may themselves be important mediators ofBT3172–protein interactions.

B. thetaiotaomicron contains the largest collection of HTCSamong sequenced human gut-associated Bacteroides, both inturns of absolute number and when corrected for genome size.The B. thetaiotaomicron genome also encodes the largest col-

lection of glycoside hydrolases and polysaccharide lyases amongsequenced Bacteroides, suggesting that its evolved collection ofHTCS paralogs, with their structurally diversified sensor do-mains, provides a means for controlling its elaborate glycanprocessing apparatus in an integrated and responsive manner.These features could explain B. thetaiotaomicron’s abundantrepresentation in our densely populated, competitive distal gutecosystem. Identifying the ligands recognized by HTCS sensordomains may provide a starting point for developing strategiesfor intentionally manipulating the biological activities of B.thetaiotaomicron and perhaps other Bacteroides in vivo and formatching diet to B. thetaiotaomicron’s niche.

Materials and MethodsGeneration and Growth an Isogenic BT3172::pGERM Mutant. Themethods used for targeted disruption of BT3172 and growth ofWT and mutant strains in a BioFlo-110 batch culture fermentorunder defined nutrient conditions are similar to those reportedin earlier publications for other B. thetaiotaomicron genes andstrains (3, 15). Further details can be found in Supporting Text.

GF Mice. All experiments employing mice were performed usingprotocols approved by the Washington University Animal StudiesCommittee. NMRI-KI mice were maintained in plastic gnotobioticisolators (16) under a strict 12-h light cycle. Animals were fed astandard autoclaved chow diet (B & K Universal, Hull, U.K.) or anirradiated diet composed of 35% (wt�wt) glucose, 35% sucrose, and20% protein (Bio-Serv, Frenchtown, NJ). For further details aboutcolonization and purification of RNA from and analyses of me-tabolites in cecal contents, see Supporting Text.

GeneChip Analysis. cDNA targets were prepared from both in vivoand in vitro RNA samples by using protocols described in the E.coli Antisense Genome Array manual (Affymetrix). The cDNAproduct was isolated (QiaQuick Spin columns; Qiagen, Valencia,CA), fragmented (DNase-I; Amersham Pharmacia Biosciences),biotinylated (bioArray Terminal Labeling Kit; Enzo) and hy-bridized to custom B. thetaiotaomicron GeneChips (3).

Yeast Two-Hybrid Screens. A B. thetaiotaomicron cDNA librarywas produced by using the Matchmaker Library Construction kitand the GAL4 activation domain fusion vector (BD Biosciences).RNA was prepared from samples obtained during growth of WTB. thetaiotaomicron from log to stationary phase in batch culturefermenter vessels containing TYG or MM-G. For further detailsabout the screen, see Table 4 and Supporting Text.

We thank David O’Donnell, Maria Karlsson, Janaki Guruge, andSabrina Wagoner for invaluable assistance with experiments involvinggnotobiotic mice. This work was supported by National Institutes ofHealth Grants DK30292 and DK52574.

1. Ley, R. E., Peterson, D. A & Gordon, J. I. (2006) Cell 124, 837–848.2. Coutinho, P. M. & Henrissat, B. (1999) in Recent Advances in Carbohydrate

Bioengineering, eds. Gilbert, H. J., Davies, G., Henrissat, B. & Svensson, B. (R.Soc. Chem., Cambridge, U.K.), pp. 3–12.

3. Sonnenburg, J. L., Xu, J., Leip, D. D., Chen, C. H., Westover, B. P.,Weatherford, J., Buhler J. D. & Gordon, J. I. (2005) Science 307, 1955–1959.

4. Backhed, F., Ley, R. E., Sonnenburg, J. L., Peterson, D. A. & Gordon, J. I.(2005) Science 307, 1915–1920.

5. Eckburg, P. B., Bik, E. M., Bernstein, C. N., Purdom, E., Dethlefsen, L.,Sargent, M., Gill S. R., Nelson, K. E. & Relman, D. A. (2005) Science 308,1635–1638.

6. Ley, R. E., Backhed, F., Turnbaugh, P, Lozupone C. A., Knight, R. D.& Gordon, J. I. (2005) Proc. Natl. Acad. Sci. USA 102, 11070 –11075.

7. Gill, S. R., Pop, M., DeBoy, R. T., Eckburg, P. B., Turnbaugh, P., Samuel, B.S., Gordon, J. I., Relman, D. A., Fraser-Liggett, C. M. & Nelson, K. E. (2006)Science, in press.

8. Xu, J., Bjursell, M. K., Himrod, J., Deng, S., Carmichael, L. K., Chiang, H. C.,Hooper, L. V. & Gordon J. I. (2003) Science 299, 2074–2076.

9. Salyers, A. A., Vercellotti, J. R., West, S. E. & Wilkins, T. D. (1977) Appl.Environ. Microbiol. 33, 319–322.

10. Stock, A. M., Robinson, V. L. & Goudreau, P. N. (2000) Annu. Rev. Biochem.69, 183–215.

11. Miyazaki, K., Miyamoto, H., Mercer, D. K., Hirase, T., Martin, J. C., Kojima,Y. & Flint, H. J. (2003) J. Bacteriol. 185, 2219–2226.

12. Xu, J., Chiang H. C., Bjursell, M. K. & Gordon J. I. (2004) Trends Microbiol.12, 21–28.

13. Coyne, M. J, Reinap, B., Lee, M. M. & Comstock, L. E. (2005) Science 307,1778–1781.

14. D’Elia, J. N. & Salyers, A. A. (1996) J. Bacteriol. 178, 7180–7186.15. Hooper, L. V., Xu, J., Falk, P. G., Midtvedt, T. & Gordon, J. I. (1999) Proc.

Natl. Acad. Sci. USA 96, 9833–9838.16. Hooper, L. V., Mills, J. C., Roth, K. A., Stappenbeck, T. S., Wong, M. H. &

Gordon, J. I. (2002) in Molecular Cellular Microbiology, eds. Sansonetti, P. &Zychlinsky, A. (Academic, San Diego), Vol. 31, pp. 559–589.

Sonnenburg et al. PNAS � June 6, 2006 � vol. 103 � no. 23 � 8839

MIC

ROBI

OLO

GY

Dow

nloa

ded

by g

uest

on

May

22,

202

1