DOI: 10.1126/science.1224203, 1255 (2012);336 Science

et al.Elizabeth K. Costellothe Human MicrobiomeThe Application of Ecological Theory Toward an Understanding of

This copy is for your personal, non-commercial use only.

clicking here.colleagues, clients, or customers by , you can order high-quality copies for yourIf you wish to distribute this article to others

  here.following the guidelines

can be obtained byPermission to republish or repurpose articles or portions of articles

  ): July 1, 2012 www.sciencemag.org (this information is current as of

The following resources related to this article are available online at

http://www.sciencemag.org/content/336/6086/1255.full.htmlversion of this article at:

including high-resolution figures, can be found in the onlineUpdated information and services,

http://www.sciencemag.org/content/336/6086/1255.full.html#relatedfound at:

can berelated to this article A list of selected additional articles on the Science Web sites

http://www.sciencemag.org/content/336/6086/1255.full.html#ref-list-1, 26 of which can be accessed free:cites 79 articlesThis article

http://www.sciencemag.org/content/336/6086/1255.full.html#related-urls5 articles hosted by HighWire Press; see:cited by This article has been

http://www.sciencemag.org/cgi/collection/microbioMicrobiology

subject collections:This article appears in the following

registered trademark of AAAS. is aScience2012 by the American Association for the Advancement of Science; all rights reserved. The title

CopyrightAmerican Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005. (print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by theScience

on

July

1, 2

012

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

with a quantitative understanding of key envi-ronmental risk factors and their interactions withour human and microbial genomes. To completethis picture, it is important to move beyond large-ly DNA sequencing–based association studiestoward a mechanistic understanding of howmembers of the gut microbiota, either in isola-tion or through mutualistic interactions with eachother, can transform each compound. This willrequire multiple complementary top-down andbottom-up approaches, including detailed in vitroanalyses of culturable microbes; studies in germ-free and intentionally colonized animal models;metagenomic surveys of patients before, during,and after treatment; and large-scale clinical trials.These types of studies will likely lead to newmicrobial therapeutic targets, noninvasive bio-markers for drug toxicity or efficacy, and a broader

understanding of the short- and long-term impactof xenobiotics on host and microbial physiology.Furthermore, the detailed study of pharmaceuti-cally active compounds may be a tractable firststep toward understanding the fundamental rulesthat govern the immense phylogenetic and meta-bolic diversity of our microbial partners and howthey influence our predisposition to and recoveryfrom disease.

References and Notes1. L. S. Goodman, L. L. Brunton, B. Chabner, B. C. Knollmann,

Goodman & Gilman's Pharmacological Basis ofTherapeutics (McGraw-Hill, New York, ed. 12, 2011).

2. Q. Ma, A. Y. Lu, Pharmacol. Rev. 63, 437 (2011).3. T. Sousa et al., Int. J. Pharm. 363, 1 (2008).4. S. P. Claus et al., mBio 2, e00271-10 (2011).5. T. A. Clayton, D. Baker, J. C. Lindon, J. R. Everett,

J. K. Nicholson, Proc. Natl. Acad. Sci. U.S.A. 106, 14728(2009).

6. C. Jernberg, S. Löfmark, C. Edlund, J. K. Jansson,Microbiology 156, 3216 (2010).

7. L. Dethlefsen, D. A. Relman, Proc. Natl. Acad. Sci. U.S.A.108, 4554 (2011).

8. M. Blaser, Nature 476, 393 (2011).9. L. C. Antunes et al., Antimicrob. Agents Chemother. 55,

1494 (2011).10. I. Sekirov et al., Infect. Immun. 76, 4726

(2008).11. J. Y. Chang et al., J. Infect. Dis. 197, 435 (2008).12. J. F. Dobkin, J. R. Saha, V. P. Butler Jr., H. C. Neu,

J. Lindenbaum, Science 220, 325 (1983).13. Z. Wang et al., Nature 472, 57 (2011).14. B. D. Wallace et al., Science 330, 831 (2010).

Acknowledgments: H.J.H. is supported by a postdoctoralfellowship from the Canadian Institutes of HealthResearch (MFE-112991). P.J.T. is supported by a grantfrom the NIH (P50 GM068763).

10.1126/science.1224396

REVIEW

The Application of Ecological TheoryToward an Understanding of theHuman MicrobiomeElizabeth K. Costello,1 Keaton Stagaman,2 Les Dethlefsen,1,3

Brendan J. M. Bohannan,2 David A. Relman1,3,4*

The human-microbial ecosystem plays a variety of important roles in human health and disease.Each person can be viewed as an island-like “patch” of habitat occupied by microbial assemblagesformed by the fundamental processes of community ecology: dispersal, local diversification,environmental selection, and ecological drift. Community assembly theory, and metacommunitytheory in particular, provides a framework for understanding the ecological dynamics of the humanmicrobiome, such as compositional variability within and between hosts. We explore three corescenarios of human microbiome assembly: development in infants, representing assembly inpreviously unoccupied habitats; recovery from antibiotics, representing assembly after disturbance;and invasion by pathogens, representing assembly in the context of invasive species. Judiciousapplication of ecological theory may lead to improved strategies for restoring and maintaining themicrobiota and the crucial health-associated ecosystem services that it provides.

Each person is an assemblage of not onlyhuman cells but also many symbiotic spe-cies. The abundant and diverse microbial

members of the assemblage play critical roles inthe maintenance of human health by liberatingnutrients and/or energy from otherwise inacces-sible dietary substrates, promoting differentiationof host tissues, stimulating the immune system,and protecting the host from invasion by path-ogens. A number of clinical disorders are asso-

ciated with alterations in host-associated microbialcommunities (the “microbiota”), including obe-sity, malnutrition, and a variety of inflammatorydiseases of the skin, mouth, and intestinal tract.Thus, the human body can be viewed as an eco-system, and human health can be construed as aproduct of ecosystem services delivered in partby the microbiota.

There is growing interest in the use of theo-retical methods to study microbial communityecology in general and host-associated micro-biota in particular (1, 2). Recent discoveries ofunexpected variation in the composition of themicrobiome of healthy individuals (3–5) high-light the importance of identifying the processesthat could possibly give rise to such variation.Ecological theory seeks to explain and predictobservable phenomena, such as temporal and

spatial patterns of diversity. Here, we explorehow community assembly theory can be used tounderstand the human-associated microbiota andits role in health and disease.

Ecological Processes Within HumansThe essential building blocks of community as-sembly theory encompass the processes thatcreate and shape diversity in local assemblages:dispersal, diversification, environmental selection,and ecological drift (6). In addition, coevolutionprovides another lens through which to view thehuman-microbial ecosystem (7), although in thisreview we focus on shorter-term dynamics at thelevel of individual hosts.

Dispersal, or the movement of organismsacross space, is a fundamental process by whichdiversity accumulates in local microbial com-munities. The concept put forth in the late 19thand early 20th centuries that “everything is ev-erywhere, but the environment selects” had apowerful impact on thinking about communityassembly (8), but a more recent appreciation ofmicrobial dispersal limitation suggests that thisconceptualization was overly simplistic. Think-ing in terms of dispersal leads to a view of thehuman body as an “island,” a patch of habitatthat is continually sampling the pool of availablecolonists. The list of available colonists maybe influenced by microbial traits—those affect-ing dispersal efficiency, transmission routes, and“ex-host” survivability—and by patterns of hostcontact and carriage, among other factors. Con-trol of infectious disease transmission dependson accurate models of host-to-host microbial dis-persal (9), and these could guide investigationsinto the dissemination of the human microbiome.Selection favors efficient dispersal in pathogens,but perhaps less so among beneficial bacteria,because the host is harmed by the first and not bythe second; for beneficial microbes, transmissionroutes such as direct or close contact may bemoreimportant. The density and spatial arrangement

1Department of Microbiology and Immunology, Stanford Uni-versity School of Medicine, Stanford, CA 94305, USA. 2Instituteof Ecology and Evolution, University of Oregon, Eugene, OR97403, USA. 3Department of Medicine, Stanford UniversitySchool of Medicine, Stanford, CA 94305, USA. 4Veterans AffairsPalo Alto Health Care System, Palo Alto, CA 94304, USA.

*To whom correspondence should be addressed. E-mail:relman@stanford.edu

www.sciencemag.org SCIENCE VOL 336 8 JUNE 2012 1255

SPECIALSECTION

on

July

1, 2

012

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

The Gut Microbiota

of host habitat patches has been highly dynamicthroughout human history, as a result of humanmigration, urbanization, changes in family livingstructure, air travel, and other factors.

A second process operating in microbialcommunities is local diversification. Unlike inmost plant and animal communities, this pro-cess can take place over short ecological timescales for microbes. Large microbial populationsizes, high growth rates, and strong selectiveregimes, all of which can be found in the hu-man body, result in rapid microbial adaptationvia mutation or recombination. Recombinationvia horizontal gene transfer may be especially

common among members of the human mi-crobiota, especially those sharing the same eco-logical niche (e.g., body site) (10). Microbialdiversification may also be driven by interactionswith phage in the human body. The processesof dispersal and diversification may interact (11);for example, immigration may suppress adapt-ive radiations (12).

Relative abundances in local communitiesare shaped over time by a third process: envi-ronmental selection. When considering environ-mental selection, or niche-based interactions,the human body can be viewed in two ways.First, it can be viewed as a “habitat filter,” a col-

lection of resources and conditions that allowthe growth of some microbes but not others, re-sulting in the selection of microbial traits thatpermit survival and growth in the host. In thisview, the host shapes the microbiota, but not theother way around. Body temperature is an ex-ample of such filtering, because microbes alterbody temperature (causing fever) only when theytransgress host anatomic boundaries. Second, thehuman body and its symbionts can be viewed asa community of interacting cells. This view dif-fers from the habitat filter view in that it assumesstrong feedbacks between hosts and microbes aswell as among microbes. This view assumes that

5

6

8

5

6

8

9

1011

12

9

1011 100

12

00

9

10

2

9

2

9

1

1

1 7

5

6

8

6

5

8

6

888

555

7

5

71

4

32

1

4

9

1

2

9

1

9

1

1

3

1

4

2

1

4

1

4

3 13

1

4

32

1

1

1211 1210 11

9 100865431

432 9865 7

5

6

8

9

1011

12

9

1011

12

5

6

8100 7 66

5

8

5

8

100 7

1

4

9

1

2

1

9

1

1

9

1

9

1

3 11

1

4

32

1

4

9

1

12

9

1

9

1

1

3

5

8

7

5

8

5

6

8

16 1

1

4

32

1 1211 1210 11

9 100865431

432 9865 7

1 1211 1210 119 100865431 432 9865 71211 12

10 119 10086543

1432 9865 7

1010

5

3

1 1200

5

3

12 12

11

5

3

5

3

12 1

1

2

1

1

1 1 14

121

1

1

1

1

2 1

1

9

2 2 2

2

5

2

5

8 8

1

5

8

5

2 12

2

5

7 7

1

9 9

2

5

12

9

14 1 22

2

5

7

9

12 2

3

32

2

3

3

2

3

2

2

3

2

3

33 2

3

3

3

2

2

3

2

3

3 33 3

9

103

3

11

3

3

3

9

11

12

9

1011 00

12

9

1

12

9

10

2

00 7

888

9

11

12

9

10

A

C

B

D

iiii ii iii iiiiiii ii iii iii

iiii ii iii iii ii ii iiiii iii

γ-diversityEnvironmental selection

Species pool

Historical contingency

Species pool

Invasion sequence

α-diversity

Localcommunity

Habitat typeA A B B C C A A B B C C

β-diversity

Dispersal limitationRandom sampling

Species pool Species pool

Invasion sequence

Localcommunity

Habitat typeA A B B C C A A B B C C

Fig. 1. Alternative community assembly scenarios could give rise to the com-positional variations observed in the human microbiota. Each panel shows theassembly of local communities in different habitat types from a pool of availablespecies. (A to C) Each local community has access to all available colonists, butthe order of invasion varies. In (A), local species composition is determinedprimarily by environmental selection: Regardless of invasion order, habitats withinitially similar conditions select for similar assemblages. In (B), the opposite istrue: Regardless of initial habitat conditions, historical contingencies (i.e., differ-ences in the timing and order of species invasions) determine assemblage com-position. In (C), neither habitat nor history matter: Local communities assemblevia random draws from the species pool. (D) Dispersal barriers result in localcommunities that assemble from different species pools. For each of the pools,

local communities may assemble as in (A), (B), or (C). The meaning of threedifferent diversity measures is shown in (A): g-diversity refers to the “regional”species pool (i.e., the total diversity of the local communities connected viadispersal); b-diversity refers to the differences between local communities (spe-cies turnover); and a-diversity refers to the diversity within a local community.Although multiple scenarios are likely to apply to any real-world setting, onemay dominate. For example, differences between body habitats may be bestexplained by environmental selection, differences between siblings for the samehabitat may be best explained by historical contingency, differences betweenmonozygotic twins prior to weaning highlight the role of stochasticity, anddifferences between neonates born by cesarean section versus vaginal deliveryare likely to be explained by dispersal limitation. [Adapted from (81, 82)]

8 JUNE 2012 VOL 336 SCIENCE www.sciencemag.org1256

on

July

1, 2

012

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

the host shapes the microbiota, and vice versa.Interactions between the host immune systemand the microbiota might be best represented bythis view [see (13)]. The overall patterns thatarise from dispersal and environmental selectioncan vary as a function of the spatial scale overwhich these processes occur (14).

In addition to selection-driven changes, spe-cies abundances may fluctuate as the resultof a fourth process, known as ecological driftor demographic stochasticity. As a result of thisprocess, low-abundance species (e.g., recent im-migrants, strains nearly eliminated by antibiotics,or strains occupying niches with low carryingcapacity) are more likely to proceed toward localextinction and become lost from the system,unless they have (or can gain) a competitive ad-vantage, access a different niche, or become re-plenished by dispersal from outside the community.Thus, dispersal can effectively “rescue” speciesfrom the brink of local extinction.

Finally, the human habitat can be understoodas a host-symbiont “holobiont”—an ecologicalsystem under selection to minimize conflict be-tween individual members. This view emphasizesa dominant role of coevolution in the assemblyand dynamics of the human ecosystem and re-minds us that long- and short-term selective pres-sures on the human microbiota are not necessarilyaligned. Any mutualistic trait that imposes acost on the microbes that express it—such asproducing dedicated molecules to interfere withpathogens or modulate host immune activity—represents a trade-off between the immediate se-lection against that cost and the long-term selectionin favor of mutualism (7).

In summary, different views of humans asmicrobial habitats make different assumptionsabout the processes most important to theassembly and dynamics of the human micro-biome. Community assembly can be conceptual-ized as being niche-based, dispersal-limited,historically contingent, or random, dependingon the relative contributions of habitat condi-tions, colonist availability, arrival order (and tim-ing), or chance-driven events, respectively, inshaping observed patterns (Fig. 1). Metacom-munity theory integrates the four processes de-scribed above and provides a useful frameworkfor considering community assembly in the hu-man body (6).

Metacommunity Theory and theHuman MicrobiomeVarious theoretical frameworks are used to studycommunity assembly; one key framework isneutral theory (15), in which it is assumed thatdispersal, diversification, and ecological drift arepurely chance-driven processes. It is a neutralmodel because it invokes neither environmentalselection nor inherent differences in species’ability to disperse or diversify. Although neutraltheory on its own is quite valuable in testing this

null hypothesis, a broader framework for theassembly of the human microbiome might ac-commodate alternative theories and combine thestrengths of transmission dynamic models (e.g.,inclusion of host contact and carriage dynamics)with those of community ecology (e.g., focus oncommunities rather than individual pathogens).One such approach is metacommunity theory(16), which could be especially useful for model-ing host-associated communities.

Metacommunity theory views the world as acollection of patches—spatially distinct areas ofsuitable habitat surrounded by a matrix of unsuit-able habitat. These patches each contain a com-munity of organisms, and these spatially distinctcommunities are connected together to form ametacommunity by the dispersal of organismsfrom patch to patch. Human populations can beviewed likewise, with host-to-host dispersallinking microbial communities. Metacommunitytheory is especially helpful for understanding therelative importance of dispersal and environmen-tal selection in shaping host-associated commu-nities (1), an issue that has received relatively littleattention in studies of the human microbiome.

The predictions of metacommunity theorydepend on the frequency and extent of dispersal,differences in the traits of individual organisms,and the degree to which patches vary in theirenvironmental conditions (16, 17). Dispersal canbe infrequent and localized, or widespread andfrequent, as discussed above. In some metacom-munity models, patches are assumed to be envi-ronmentally identical, with movement amongpatches determining variation in communitymembership. Such models might be especiallyappropriate for populations of closely relatedhosts. Other models assume that patches varystrongly in their available niches, and that var-iation in community membership results at leastin part from environmental selection (e.g., under-pinned by host genetics or diet).

Metacommunity theory enables one to pre-dict the conditions under which community dy-namics within a patch are driven by immigrationfrom outside versus local adaptation (11). Lowdispersal rates favor adaptation within a patch,whereas high dispersal favors immigration. Thisconcept could be useful, for example, in under-standing responses to antibiotic use. If acquisitionof antibiotic resistance were primarily a result ofimmigration, then interventions focused on quar-antine and hygiene would be more effective thanthose focused on altering antibiotic duration ordose (see below).

Metacommunity theory has been used to elu-cidate the drivers of non–host-associated mi-crobial community membership and dynamics(18–20), but it has rarely been used to study host-associated communities (21, 22)—for example,to explore the stringency of host selection and itsdependence on the microbial group or the ageof the host. Ultimately this information will result

in a better understanding of how microbes are“filtered” by the host and, conversely, how mi-crobes evade this filtering. This information iscrucial if clinicians are to directly manipulatehost-associated communities—for example, bydesigning probiotics that can evade host filteringand establish within a host.

The effective application of metacommunitytheory (and assembly theory in general) to thehuman microbiome requires a preliminary un-derstanding of how the microbiome varies acrosshosts and over time. We next review our currentunderstanding of this variation, focusing on thedynamics of communities in newly created hab-itats (e.g., neonatal colonization), after distur-bance (e.g., after antibiotic treatment), and afterinvasion (e.g., by a pathogen). We chose thesescenarios because they represent and reveal thefundamental types of assembly relevant tohuman health.

Postnatal Acquisition and Developmentof the Human MicrobiomeBabies are born essentially sterile and acquiretheir microbiome from their surroundings. Thepostnatal assembly of the human microbiotaplays an important role in infant health, pro-viding resistance to pathogen invasion, immunestimulation, and other important developmentalcues early in life (23). Acute and chronic disor-ders, such as necrotizing enterocolitis, antibiotic-associated diarrhea, malnutrition, inflammatorybowel disease, and asthma have been linked toinadequate, inappropriate, or disrupted post-natal microbiome acquisition and development(24). Mechanisms controlling the appearanceof bacteria in healthy infants have been studiedfor well over a century (25), with microbiomedevelopment having been likened to ecologicalsuccession (23, 26). The view of succession as amode of community assembly has largely empha-sized niche-based processes, but the importanceof stochastic and/or historical events has alsolong been recognized.

In the absence of microbial invasion of theamniotic cavity, which is thought to be a rare,pathologic condition, rupture of membranes sig-nals the moment when microbes, most likely ofmaternal vaginal origin, first gain access to theneonate. Vaginally delivered infants clearly re-ceive a strong input of vaginal and possibly otherurogenital or fecal microbiota as they pass throughand exit the birth canal (27, 28). Vaginal micro-biome composition in nonpregnant, reproductive-age women is highly dynamic and is characterizedby at least five compositional classes delineatedby different, dominant Lactobacillus species ora lack of Lactobacillus dominance. There isfrequent class switching over time, includingswitching to and from compositions indicative ofbacterial vaginosis, even in the absence of symp-toms (4, 29, 30). Whether these dynamics occursimilarly in pregnant and postpartum women

www.sciencemag.org SCIENCE VOL 336 8 JUNE 2012 1257

SPECIALSECTION

on

July

1, 2

012

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

The Gut Microbiotahas important implications for the initial col-onization of vaginally delivered infants; if theydo, infant-to-infant variation in the compositionof initial colonists may be imposed in some casesby maternal vaginal microbiome class at the timeof delivery. Likewise, maternal gut microbiometypes (5, 31) may also determine the pool ofcolonists available to vaginally delivered infantsat birth. Thus, variation among neonate micro-biomes may reflect variation in maternal micro-biomes, but this has not been widely tested formaternal habitats other than the vagina. At thetime of delivery, microbiomes do not differ con-sistently among infant body sites (27), whichimplies that sampling is driving initial commu-nity assembly, with minimal filtering by the in-fant host.

Delivery mode also determines microbial ex-posure at the time of birth. For example, infantsdelivered by cesarean section do not receivecontributions from the vaginal microbiota, andinstead are exposed initially to what appears to beambient skin microbiota (27). Incidental expo-sures to maternal (or other) gut or vaginal micro-biota may occur later in cesarean section infants,at low density or low frequency, but may beinadequate for outcompeting already establishedstrains. For example, cesarean section infantsdisplay reduced abundances and/or incidencesof colonization by the genera Bacteroides andBifidobacterium early in development relativeto infants born by vaginal delivery (32, 33). Theeffects of delivery mode can persist for monthsand may have consequences for infant health; ce-sarean section infants have a higher risk for someimmune-mediated diseases (34, 35). The ambientenvironment may also play a role in colonizationat delivery; infants delivered at home versus thehospital were colonized differently at 1 monthof age (33). Thus, dispersal limitation imposedby certain medical interventions may contributeto interindividual variation early in life.

Over the first few months—roughly up untilthe first solid foods are introduced—a well-constrained range of stereotypical bacteria appearin the feces (distal gut), diversity generally in-creases, and aerobes are thought to be supplantedby facultative and then strict anaerobes (23). Ex-clusive breast-feeding selects for increased abun-dance of particular Bifidobacterium species whosegenome sequences reflect specialized use of hu-man milk oligosaccharides and similar host-derived substrates (36), or for other bacteria suchas Bacteroides that could compete for the sameecological niche (37). Strikingly, during this earlyphase, microbiota composition is highly dynam-ic within and between infants (3, 23, 38–40),with temporal variation characterized by periodsof relative stability (for varying lengths of time)punctuated by abrupt shifts in composition andstructure. In some cases, these shifts can be linkedwith life events that likely impose environmentalselection, such as fever, formula feeding, or anti-

biotic therapy (3, 39, 41). Extraordinarily paralleltransitions observed in a pair of dizygotic twinssuggest that stochastic exposures (shared expo-sures in their case) can also play an important roleduring this phase, driving within- and between-infant variation (3). This finding underscores theneed to better understand how infants sample theirenvironment over time (e.g., whether outside-of-host environmental reservoirs, or host exchangescenarios—either direct or indirect, such as viafomites—prevail) and with regard to the frequencyand extent of dispersal (as discussed above). Abruptshifts might reflect opportunistic invasions bybetter-adapted species or changes in filtering bythe host. An infant’s unique developmental paththrough this early unstable phase may have longer-term health implications. For example, recent workhas shown that colonization during the neonatalperiod has a particularly important effect on mu-cosal immune development (42).

The introduction of solid foods and wean-ing are associated with the onset of a transitiontoward an adult-like gut microbiome. Differencesattributable to early exposures such as deliverymode fade as microbiota compositions becomemore canalized. Life events such as illness, dietmodification, and antibiotic therapy can stillimpose disturbances, although specific composi-tions appear to recover. Taxa characteristic of theadult eventually establish, but the process ofmicrobial community assembly appears to ex-tend past the first year of life and into childhood(3, 39). If there is an imprint of microbial flowfrom parents to children, it is difficult to detect atearly ages and/or emerges gradually later in life.In one study, fecal patterns of bacterial taxo-nomic diversity in 1-year-olds were not found tobe more similar to those of their parents than tothose of unrelated adults (3); but in another study,patterns of microbial diversity in adult twinswere slightly more similar to those of their mother(43). These findings suggest that we acquiremicrobes from sources other than, or in additionto, our family members. Further, there may bestrong selection for an individualized microbiota.Describing the adult state as “stable” may notsuffice when stability is defined as the permanentcoexistence of locally occurring species, becauseeven adult gut composition appears to changeover time (44). In summary, microbiome as-sembly in newly created habitats likely involvesa gradual shift from conditions under the stronginfluence of dispersal limitation (as well as sto-chastic and/or historical factors) toward condi-tions increasingly influenced by environmentalselection (e.g., by diet), with weaning as a strongcatalyst, and with development toward adult-likecomposition continuing into childhood.

Community Assembly After Disturbance:Antibiotics as a ParadigmThe assembly of human-associatedmicrobial com-munities does not, in general, proceed smoothly to

a stable climax state, which then resists furtherchanges in composition. Disturbances often re-move or kill some fraction of the community,providing an opportunity for remaining commu-nity members or new colonists to increase inabundance. For example, personal oral hygieneremoves bacterial biofilm from teeth, and anti-biotics affect not only the targeted pathogen butalso members of the normal microbiota (Fig.2A). The former case represents a deliberateattempt to interrupt the development of microbialcommunities that might be associated withperiodontitis; the latter case is an inadvertentconsequence of medical intervention. In additionto directly inducing a shift in the communitystate, disturbance may also involve a change inthe community’s habitat—for example, as theresult of a change in host diet (Fig. 2B). In manycases, a crucial unknown is resilience—that is,the degree to which the post-disturbance com-munity returns to its former state (45). Althoughmost work on community resilience has consideredresilience in terms of taxonomic composition, as-sessment of function and ecosystem services maybe even more important.

The effect of antibiotics on the gut microbiotaserves as a paradigm for disturbances in human-associated communities. Antibiotics are now oneof the most common and important forms ofdisturbance of the human microbiota; on anygiven day, approximately 1 to 3% of people inthe developed world are exposed to pharmaco-logic doses of antibiotics (46). Over the past sev-eral decades there has been increasing concernabout the spread of antibiotic resistance amongpathogens, as well as growing concern that anti-biotic use may disrupt the host-microbe inter-actions that contribute to human health.

Antibiotic therapy is meant to achieve asufficient concentration of the drug for a suf-ficient duration in a particular body site so thatthe targeted pathogen is eliminated. Even if thisaim were always attained, the antibiotic will befound at a range of concentrations at many lo-cations in the body, depending on the mode ofadministration and its pharmacodynamic prop-erties. Where members of the indigenous mi-crobiota are exposed to antibiotics that affecttheir growth without killing them, there is se-lection for resistance. Human gut and oral com-munities are recognized as reservoirs for theevolution and horizontal transfer of antibioticresistance determinants, including to pathogens(10, 47, 48). However, antibiotic resistance amongthemicrobiota is one of several mechanisms thatmay act to enhance the resilience of indigenouscommunities, hence preserving their beneficialecosystem services. Others may include population-level resistance via stress-response signaling (49)and the existence of dormant persister cells (50)or refugium-like locations (e.g., mucus layer).

Only a handful of studies have used cultivation-independent surveys to examine the consequences

8 JUNE 2012 VOL 336 SCIENCE www.sciencemag.org1258

on

July

1, 2

012

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

of therapeutic doses of antibiotics on the humangut microbiota [e.g., (51–55)]. These studies—although they examined different antibiotics bymeans of various sampling strategies, treatmentdurations, and analytical approaches—all havefound that antibiotics alter the composition of thegut microbiota, and that the abundance of mosttaxa begins to return to prior levels within severalweeks. However, the studies are also consistentin showing that various taxa recover to differ-ent extents and that some do not recover overthe duration of the study. The antibiotic effectis greater than the routine temporal variabilityof community composition (51, 52, 54). Some

studies have revealed that the composition ofstrains within a taxon is sometimes altered, evenif the relative abundance of the taxon as a wholehas returned to pre-antibiotic levels. In both ofthe studies that involved measurements of theprevalence of antibiotic-resistant strains, elevatedlevels of resistance persisted to the end of thestudy (53, 54).

Overall, research suggests that the human gutmicrobiota of generally healthy adults is largely,but not entirely, resilient to short courses of anti-biotic therapy, whereas clinical evidence indi-cates that extended or repeated courses are morelikely to result in serious complications such as

the invasion and bloom of Clostridium difficile(56). Perhaps over short courses of antibiotics, asufficient, although possibly quite small, numberof residual cells from most of the large, preexist-ing populations survives to recolonize the gut.An increasing number of these residual cells maybe lost with longer or repeated courses of anti-biotics. Thus, reassembly of the microbial com-munity after extended antibiotic treatment mayrequire colonization from outside the host—aprocess that would likely be more variable andrequire a longer period of time than reassemblyvia the filtering of existing populations in thehost. In addition, the microbiome may be highly

vulnerable to invasion by (and/orblooms of) pathogens during re-covery after disturbance, becauseresources are in high abundanceand resident populations are di-minished. The longer recoverytime required after extended anti-biotic treatment could lead to ahigher probability of invasion bypathogenic strains. One can en-vision a more enlightened strategyfor the clinical use of antibioticsthat includes pretreatment esti-mates of a patient’s microbial com-munity resilience, based on theuse of a standardized disturbanceand monitoring of key communityproducts, mapping of the commu-nity stability landscape, and assess-ment of the likelihood for communitydisplacement and adoption of adisadvantageous, altered state. As-sessments of elevated risk, or of lossof resilience, might then promptefforts at restoration [see (57)].

Little is known regarding theresponse of the microbiome tofrequent antibiotic use. When dis-turbances take place with a mag-nitude or frequency beyondwhat acommunity has had an opportunityto adapt to, ecological surprisesmayoccur (58). Such frequent distur-bances may allow the persistenceof microbial taxa that are inferiorcompetitors within a given host butare nonetheless maintained acrosshosts because they have traits thatresult in widespread and frequentdispersal (i.e., “fugitive” taxa).Such a scenario is analogous to thepatch dynamics paradigm of meta-community theory (16).

Assembly of the HumanMicrobiome in the Contextof Invaders (Pathogens)It is naïve to consider only hostand pathogen when predicting the

Antibiotics,oral hygiene

Community statelandscape

Community statelandscape

A

B

Alternative state 1

Diet intervention,immunosuppressive drug

Alternative state 1

Alternative state 2

Alternative state 2

Shift in “state” variables alters the community directly

Shift in environmental “parameters” alters the community indirectly

Fig. 2. Disturbance can be illustrated using a stability landscape schematic. The ball represents the community; thechanging horizontal position of the ball represents the changing community state. The depth of a basin indicates thelikelihood of a community remaining in that basin despite frequent “buffeting” by minor disturbance (45) and hencethe relative stability of the community state. Disturbance can alter the community directly (A) by changing itscomposition or activity, or indirectly (B) by changing the environmental parameters. In either case, the community canshift to an alternative state. In reality, continuous feedback between the community and its environment means thatthey change together. See (57) for applications to therapy.

www.sciencemag.org SCIENCE VOL 336 8 JUNE 2012 1259

SPECIALSECTION

on

July

1, 2

012

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

The Gut Microbiotalikelihood of microbial disease. Host-associatedmicrobial communities play an important role inpreventing disease, and although outside thescope of this discussion, they may also pro-mote pathology, as in the case of inflammatorybowel disease (59). It may be useful to viewsome pathogens as invasive species, and to viewthe consequences of invasion as a special caseof community assembly. As with invasive spe-cies in more traditionally studied settings, theability of a species to invade a particular com-munity depends largely on niche opportunities—that is, the filters imposed by the abiotic envi-ronment and the resistance of the community tocolonization by an exotic species (60). A suc-cessful invasion involves dispersal of an in-vader to a new community, initial colonization,and proliferation; these steps are influenced bythe same processes as in community assemblymore generally.

The environment created by the host deter-mines the number of potential niche opportunities.The nature of this environment is influenced by anumber of conditions, including “abiotic” factors(such as oxygen levels, pH, and temperature) aswell as the abundance and types of available re-sources (such as the composition of the host’sdiet) (61) and carbon sources provided directlyby the host (such as mucosal poly- and oligo-saccharides) (62). In addition, the host immunesystem acts as an important environmental filterto limit the spatial extent of the microbiota’savailable niches. The main functions of the mu-cosal immune system are to create an inhos-pitable buffer zone between the microbiota andthe host epithelium, and to minimize the incidenceof systemic inflammation that would normally beinduced in the face of so many bacterial products(13, 63, 64). The immune system performs thesefunctions through three general mechanisms:(i) physical barriers such as the inner mucuslayer of the colon and stomach, which is gen-erally impenetrable to bacterial cells (65); (ii)antimicrobial peptides and mucosal antibodiesin the mucus layer that further hinder bacterialcolonization of the epithelium (63); and (iii)innate and adaptive immune responses withinthe regional lymphatic tissues (64). These threemechanisms, in most healthy hosts, select for bac-terial species that do well at or near mucosalsurfaces or strong barriers such as the skin. How-ever, host filtering is not the only factor influ-encing the ability of pathogens to invade thehost-microbiota community.

One of the most important roles of the micro-biota in mediating host-pathogen interactions isprotection of the host from pathogen invasion, or“colonization resistance” (64, 66). Protection isachieved through induction of the innate andadaptive branches of the immune system, creatingan environment that is unfavorable to pathogens[as illustrated by the observation that axenic mice(67) and zebrafish (68) have diminished immune

responses and impaired barriers to infection], andthrough direct competition (or community filter-ing, e.g., by lowering vaginal pH). In the lattercase, pathogens are kept at bay by competitionwith the microbiota for space and resources. Thisprotective effect is demonstrated by the increasedsusceptibility to infection of hosts that have hadtheir microbiota altered by antibiotics, a phenom-enon well documented by Miller and Bohnhoff inthe early 1960s with Salmonella invasion of micepretreated with antibiotics (69). The ability of cer-tain anaerobes to limit the invasion and growth ofClostridium perfringens in a diet-dependent man-ner is an example of competition for resources(70). Bifidobacterium breve produces an exo-polysaccharide that protects it from the immuneresponse; this allows it to compete for space andcolonize the mouse gut at high loads both in thelumen and at the epithelial surface without in-ducing inflammation (71). Even if invaders dogain a foothold, the indigenous microbiota canblock lethality: In mice, some B. longum strainscan protect against enterohemorrhagic E. coli–mediated death by inhibiting translocation ofShiga toxin from lumen to blood (72).

By viewing pathogens as invasive species, wesee that the contexts in which they are able tocause disease are the same as those required forany other species that invades and proliferates ina community. Niche opportunities can result fromexploiting novel or excess resources (from thehost’s food), outcompeting a commensal speciesfor the same resource, or, perhaps most important,exploiting niches left open after a disturbance.The importance of exploiting disturbance is wellillustrated by the increasing number of cases ofdisease caused by C. difficile (73), which is a“weedy” exotic (or native) species that can rap-idly fill niches once they are vacant but, in mostcases, is eventually removed or kept at low abun-dance in the absence of disturbance. Salmonellaenterica serovar Typhimurium is an example of anexotic invasive that exploits disturbance, but inthis case, it also causes the disturbance it exploits.S.Typhimurium expresses many virulence factorsthat induce inflammation in the mammalian in-testine. A mutant S. Typhimurium strain lackingthese virulence factors is unable to invade the gutcommunity and cause disease; however, if in-flammation is provided by some other mecha-nism, avirulent strains are able to invade the hostcommunities (74). Inflammation likely reducesthe abundances of other bacteria that would com-pete with pro-inflammatory pathogens for space.As one possible mechanism, inflammation causesthe intestine to produce tetrathionate, whichS. Typhimurium uses as an electron acceptor forrespiring ethanolamine, a carbon source that can-not be exploited by other bacteria, thus avoidingcompetition for nutrients (75). By causing acuteinflammation, the pathogen is able to alter thenative microbiota and effectively colonize andproliferate.

In contrast to the above examples, a reductionin disturbance frequency can also promote invasionby pathogens. Patients with cystic fibrosis producethickened mucus, which inhibits the ability of thecilia to remove foreignmaterial from normally ster-ile lung airways. This lack of constant removal(i.e., impaired innate immune host-filteringmech-anism), among other factors, allows for the estab-lishment of bacterial communities that wouldnormally not be able to persist at that site (76).

In summary, predicting the success and out-come of infection by pathogens can be aided byframing the issue as an ecological problem ofcommunity assembly. Invasion ecology highlightsthe importance of niche opportunities as deter-minants of success of invasion, and the manip-ulation of which might help in pathogen controland disease prevention. Experimental modelsusing gnotobiotic organisms such as mice andzebrafish will be helpful in understanding therole of community diversity, as well as the roleof particular community members in conferringcolonization resistance through indirect inhibi-tion and resource competition. In addition, thefrequency and magnitude of disturbance plays acrucial role in facilitating colonization by exoticinvasives as well as the expansion of native spe-cies. Finding ways, through prebiotics, probiotics,or pharmabiotics, to alter pathogen or other bac-terial species abundances (or to inhibit their detri-mental effects on the host) in a specific mannerwithout causing additional disturbance to thecommunity will be very important for preventingand treating disease caused by invasive species.

Translating Ecological Understandinginto Clinical PracticeAn improved understanding, informed by eco-logical theory, of how microbiomes assemblecould alter clinical practice by changing theperspective clinicians bring to the treatment ofinfectious disease and chronic inflammatorydisorders. The traditional perspective has beento think of the human body as a battleground onwhich physicians attack pathogens with in-creasing force, occasionally having to resort to ascorched-earth approach to rid a body of disease.Although this perspective has been very success-ful for several diseases, it has come at a greatcost. Even for those diseases for which it hasworked, the collateral damage can be severe. Aswe have discussed, antibiotics often kill morethan the target organisms (52) and can increasethe chance of invasion by unwanted organisms[such as C. difficile (73)].

The body-as-battleground approach ignoresthe community context of infectious disease andchronic inflammation, and does not take intoaccount our increasing knowledge regarding theassembly of the human microbiome. We suggestthat it is time for clinicians to abandon the warmetaphor (77). Given the ecological parallels be-tween assembly of the human microbiome and

8 JUNE 2012 VOL 336 SCIENCE www.sciencemag.org1260

on

July

1, 2

012

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

assembly of other ecological communities, wesuggest that human medicine has more in com-mon with park management than it does withbattlefield strategy. To effectively manage a plantor animal community requires a multiprongedapproach of habitat restoration, promotion ofnative species, and targeted removal of invasives.We describe below some examples of how such ahuman-as-habitat approach might alter clinicalpractice.

An ecological approach to managing in-vasions. An understanding of the dominantmechanisms of community assembly could di-rectly alter how clinicians treat disease. Con-sider, for example, the rise of drug-resistantpathogens during the course of drug treatment.We can consider this a “special case” of com-munity assembly, much as we did invasion bya pathogen (see above). We can ask: What isthe relative importance of dispersal, diversifica-tion, environmental selection, and ecologicaldrift in the successful invasion by this drug-resistant strain? If the source of these strainsis primarily through random sampling of theexternal environment, then the most effectivepreventive strategy may be quarantine andenhanced hygiene. In contrast, if such strainsarise primarily through diversification of resi-dent taxa, then multidrug treatment may bemore effective (to make successful evolutionmore difficult). If the drug-resistant strains arealready present at the outset of treatment andincrease in abundance via environmental se-lection, then drug cycling may be the mosteffective treatment to reduce the overall com-petitive advantage of the resistant strains. Ifdrug-resistant strains establish primarily throughecological drift, then disturbance may be cru-cial to their establishment (freeing up ecolog-ical “space” for their invasion). In this case,reducing disturbance of the resident micro-biota may be most effective. In this way, adetailed understanding of the relative impor-tance of different community assembly pro-cesses can be used to tailor the treatment ofdisease.

Health as a product of ecosystem services.Humans benefit from a variety of processessupplied by natural ecosystems. Collectivelythese benefits are known as ecosystem services(78). There is growing evidence that humanhealth is a collective property of the humanbody and its associated microbiome, and thuscould be considered a net effect of ecosystemservices. We envision clinical medicine focusedon managing the human body and its associ-ated microbiome to preserve these ecosystemservices. How might this be accomplished? Ingeneral ecology, the management of an eco-system service requires four basic steps (79):(i) identification of ecosystem service providers(ESPs; taxa that provide specific ecosystemservices) and characterization of their func-

tional roles; (ii) determination of how commu-nity context influences the function of theseproviders; (iii) assessment of key environmen-tal factors influencing the provision of ser-vices; and (iv) measurement of the spatial andtemporal scales at which these providers andtheir functions operate. This general frameworkwould work equally well for human health–associated ecosystem services. If studies of thehuman microbiome were structured around thesefour priorities, the development of an ecologicalapproach to medicine could be accelerated. Pro-gress has been made in identifying ESPs [“bio-markers”; see (57)]; for example, declines inFaecalibacterium prausnitzii are associated withinflammatory bowel disease, and this organismmay be an ESP for health in the human gut (80).

Adaptive management of the human body.Transitioning clinical practice from the body-as-battleground to the human-as-habitat perspec-tive will require rethinking how one managesthe human body. In the management of plantand animal communities, a system-level ap-proach known as “adaptive management” hasbecome popular. This approach is a structured,iterative process of decision-making, one thatuses system monitoring to continually updatemanagement decisions (79). It has been success-fully used to manage biodiversity in a variety ofhabitats, including communities in highly dis-turbed environments affected by overfishing andby climate change (79). For the human body, weenvision that this approach would involve mon-itoring of the microbiome during health, to es-tablish a healthy baseline, with more intensivemonitoring during disease and treatment. Thiswill require the development of new diagnostictools that are both accurate and sufficientlyrapid to inform decisions regarding therapeu-tics [see (57)]. Such diagnostics are not yet fea-sible, but given recent advances in our abilityto survey the human microbiome, this possi-bility is not far in the future, especially if weare able to identify particular components ofthe human microbiome that contribute dispro-portionately to the maintenance of human health.An adaptive management approach to clinicalmedicine provides a wonderful example of per-sonalized medicine, with treatments tailored toindividuals on the basis of diagnostic changesin an individual’s microbiome, and continuallyadjusted through regular monitoring. Such aninformation-intensive approach, guided by eco-logical theory, has the potential to revolutionizethe treatment of disease.

References and Notes1. J. R. Mihaljevic, Trends Ecol. Evol. 10.1016/j.tree.2012.01.011

(2012).2. J. I. Prosser et al., Nat. Rev. Microbiol. 5, 384 (2007).3. C. Palmer, E. M. Bik, D. B. DiGiulio, D. A. Relman,

P. O. Brown, PLoS Biol. 5, e177 (2007).4. J. Ravel et al., Proc. Natl. Acad. Sci. U.S.A. 108

(suppl. 1), 4680 (2011).

5. G. D. Wu et al., Science 334, 105 (2011).6. M. Vellend, Q. Rev. Biol. 85, 183 (2010).7. L. Dethlefsen, M. McFall-Ngai, D. A. Relman, Nature 449,

811 (2007).8. M. A. O’Malley, Nat. Rev. Microbiol. 5, 647 (2007).9. J. Koopman, Annu. Rev. Public Health 25, 303 (2004).

10. C. S. Smillie et al., Nature 480, 241 (2011).11. M. C. Urban et al., Trends Ecol. Evol. 23, 311 (2008).12. T. Fukami, H. J. Beaumont, X. X. Zhang, P. B. Rainey,

Nature 446, 436 (2007).13. L. V. Hooper, D. R. Littman, A. J. Macpherson,

Science 336, 1268 (2012).14. B. Kerr, M. A. Riley, M. W. Feldman, B. J. Bohannan,

Nature 418, 171 (2002).15. S. P. Hubbell, The Unified Neutral Theory of Biodiversity

and Biogeography (Princeton Univ. Press, Princeton, NJ,2001).

16. M. A. Leibold et al., Ecol. Lett. 7, 601 (2004).17. A. M. Ellis, L. P. Lounibos, M. Holyoak, Ecology 87, 2582

(2006).18. J. B. Logue, N. Mouquet, H. Peter, H. Hillebrand;

Metacommunity Working Group, Trends Ecol. Evol. 26,482 (2011).

19. I. D. Ofiteru et al., Proc. Natl. Acad. Sci. U.S.A. 107,15345 (2010).

20. K. Van der Gucht et al., Proc. Natl. Acad. Sci. U.S.A. 104,20404 (2007).

21. S. R. Hovatter, C. Dejelo, A. L. Case, C. B. Blackwood,Ecology 92, 57 (2011).

22. W. T. Sloan et al., Environ. Microbiol. 8, 732 (2006).23. R. I. Mackie, A. Sghir, H. R. Gaskins, Am. J. Clin. Nutr. 69,

1035S (1999).24. R. Murgas Torrazza, J. Neu, J. Perinatol. 31 (suppl. 1),

S29 (2011).25. Th. Escherich, Rev. Infect. Dis. 10, 1220 (1988).26. D. C. Savage, Annu. Rev. Microbiol. 31, 107 (1977).27. M. G. Dominguez-Bello et al., Proc. Natl. Acad. Sci. U.S.A.

107, 11971 (2010).28. R. Mändar, M. Mikelsaar, Biol. Neonate 69, 30 (1996).29. R. M. Brotman, J. Ravel, R. A. Cone, J. M. Zenilman,

Sex. Transm. Infect. 86, 297 (2010).30. P. Gajer et al., Sci. Transl. Med. 4, 132ra52 (2012).31. M. Arumugam et al.; MetaHIT Consortium, Nature 473,

174 (2011).32. R. Bennet, C. E. Nord, Infection 15, 332 (1987).33. J. Penders et al., Pediatrics 118, 511 (2006).34. M. Kuitunen et al., J. Allergy Clin. Immunol. 123, 335

(2009).35. F. A. van Nimwegen et al., J. Allergy Clin. Immunol. 128,

948, e1 (2011).36. D. A. Sela et al., Proc. Natl. Acad. Sci. U.S.A. 105, 18964

(2008).37. A. Marcobal et al., J. Agric. Food Chem. 58, 5334

(2010).38. C. F. Favier, E. E. Vaughan, W. M. De Vos, A. D. Akkermans,

Appl. Environ. Microbiol. 68, 219 (2002).39. J. E. Koenig et al., Proc. Natl. Acad. Sci. U.S.A. 108

(suppl. 1), 4578 (2011).40. P. Trosvik, N. C. Stenseth, K. Rudi, ISME J. 4, 151 (2010).41. F. Savino et al., Acta Paediatr. 100, 75 (2011).42. T. Olszak et al., Science 336, 489 (2012).43. P. J. Turnbaugh et al., Nature 457, 480 (2009).44. J. G. Caporaso et al., Genome Biol. 12, R50 (2011).45. B. Walker, C. S. Holling, S. R. Carpenter, A. Kinzig,

Ecol. Soc. 9, 5 (2004).46. H. Goossens, M. Ferech, R. Vander Stichele,

M. Elseviers; ESAC Project Group, Lancet 365, 579(2005).

47. A. P. Roberts, P. Mullany, Expert Rev. Anti Infect. Ther. 8,1441 (2010).

48. A. A. Salyers, A. Gupta, Y. Wang, Trends Microbiol. 12,412 (2004).

49. H. H. Lee, M. N. Molla, C. R. Cantor, J. J. Collins,Nature 467, 82 (2010).

50. N. M. Vega, K. R. Allison, A. S. Khalil, J. J. Collins,Nat. Chem. Biol. 8, 431 (2012).

51. L. Dethlefsen, S. Huse, M. L. Sogin, D. A. Relman,PLoS Biol. 6, e280 (2008).

www.sciencemag.org SCIENCE VOL 336 8 JUNE 2012 1261

SPECIALSECTION

on

July

1, 2

012

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om

The Gut Microbiota52. L. Dethlefsen, D. A. Relman, Proc. Natl. Acad. Sci. U.S.A.

108 (suppl. 1), 4554 (2011).53. H. E. Jakobsson et al., PLoS ONE 5, e9836 (2010).54. C. Jernberg, S. Löfmark, C. Edlund, J. K. Jansson,

ISME J. 1, 56 (2007).55. V. B. Young, T. M. Schmidt, J. Clin. Microbiol.42, 1203 (2004).56. R. C. Owens Jr., C. J. Donskey, R. P. Gaynes, V. G. Loo,

C. A. Muto, Clin. Infect. Dis. 46 (suppl. 1), S19 (2008).57. K. P. Lemon, G. C. Armitage, M. A. Fischbach, Sci. Transl.

Med. 4, 137rv5 (2012).58. R. T. Paine, M. J. Tegner, E. A. Johnson, Ecosystems 1,

535 (1998).59. E. Elinav et al., Cell 145, 745 (2011).60. K. Shea, P. Chesson, Trends Ecol. Evol. 17, 170 (2002).61. P. J. Turnbaugh et al., Nature 444, 1027 (2006).62. J. L. Sonnenburg, L. T. Angenent, J. I. Gordon,

Nat. Immunol. 5, 569 (2004).63. B. A. Duerkop, S. Vaishnava, L. V. Hooper, Immunity 31,

368 (2009).64. A. J. Macpherson, M. B. Geuking, E. Slack, S. Hapfelmeier,

K. D. McCoy, Immunol. Rev. 245, 132 (2012).

65. M. E. Johansson et al., Proc. Natl. Acad. Sci. U.S.A. 105,15064 (2008).

66. I. Sekirov, B. B. Finlay, J. Physiol. 587, 4159 (2009).67. A. J. Macpherson, N. L. Harris, Nat. Rev. Immunol. 4, 478

(2004).68. M. Kanther, J. F. Rawls, Curr. Opin. Immunol. 22, 10 (2010).69. M. Bohnhoff, C. P. Miller, J. Infect. Dis. 111, 117 (1962).70. N. Yurdusev, M. Ladire, R. Ducluzeau, P. Raibaud,

Infect. Immun. 57, 724 (1989).71. S. Fanning et al., Proc. Natl. Acad. Sci. U.S.A.109, 2108 (2012).72. S. Fukuda et al., Nature 469, 543 (2011).73. C. P. Kelly, J. T. LaMont, N. Engl. J. Med. 359, 1932 (2008).74. B. Stecher et al., PLoS Biol. 5, e244 (2007).75. P. Thiennimitr et al., Proc. Natl. Acad. Sci. U.S.A. 108,

17480 (2011).76. V. Klepac-Ceraj et al., Environ. Microbiol. 12, 1293 (2010).77. J. Lederberg, Science 288, 287 (2000).78. G. C. Daily et al., Issues Ecol. 2, 1 (1997).79. C. Allan, G. H. Stankey, Eds., Adaptive Environmental

Management: A Practitioner’s Guide (Springer,Heidelberg, 2009).

80. H. Sokol et al., Proc. Natl. Acad. Sci. U.S.A. 105, 16731(2008).

81. J. M. Chase, Oecologia 136, 489 (2003).82. T. Fukami, in Community Ecology: Processes,

Models, and Applications, H. A. Verhoef, P. J. Morin,Eds. (Oxford Univ. Press, Oxford, 2010),pp. 45–54.

Acknowledgments: We thank members of the Relman andBohannan laboratories, and in particular E. Bik, N. Qvit-Raz,G. Schmid, and K. Shelef (Stanford), and K. Guillemin,J. Green, Z. Stephens, and A. Burns (Oregon) for valuableideas and critique. E.K.C. is a Walter V. and Idun BerryPostdoctoral Fellow. B.J.M.B. is supported by NIH grantR01GM095385. D.A.R. is supported by a Distinguished ClinicalScientist Award from the Doris Duke Charitable Trust, NIHPioneer Award DP1OD000964, and the Thomas C. andJoan M. Merigan Endowment at Stanford University. Thereare no conflicts of interest.

10.1126/science.1224203

REVIEW

Host-Gut MicrobiotaMetabolic InteractionsJeremy K. Nicholson,1* Elaine Holmes,1 James Kinross,1 Remy Burcelin,2

Glenn Gibson,3 Wei Jia,4 Sven Pettersson5*

The composition and activity of the gut microbiota codevelop with the host from birth and issubject to a complex interplay that depends on the host genome, nutrition, and life-style. Thegut microbiota is involved in the regulation of multiple host metabolic pathways, giving rise tointeractive host-microbiota metabolic, signaling, and immune-inflammatory axes thatphysiologically connect the gut, liver, muscle, and brain. A deeper understanding of theseaxes is a prerequisite for optimizing therapeutic strategies to manipulate the gut microbiotato combat disease and improve health.

Coelomate animals possess an internal bodycavity surrounding the gut and other or-gans and have coevolved with a diverse

range of symbiotic gut bacteria and other mi-croorganisms collectively known as the gut mi-crobiota [see Perspective by Gordon (1)]. Thismutually beneficial relationship between the hostand its resident microbiota results in productionof metabolites by microbes that contribute to theevolutionary fitness of the host (2). The diversity

and composition of the gut microbiota withinand between individuals of a host species isinfluenced by topographical and temporal var-iation in the microbial communities, with partic-ular bacterial species occupying specific nichesin the body habitat or being associated withparticular growth ormaturation phases of the host(1, 3). In humans, the primary individual micro-biota may reflect the maternal hand-over of “seedecology” species at birth (4, 5). Subsequentshaping of the microbial landscape is then drivenby a series of complex and dynamic interactionsthroughout life, including diet, life-style, dis-ease, and antibiotic use. This developmentaltrajectory of the microbiome, incorporating themicrobes and their collective genomes, modu-lates the metabolic phenotype of the host andgreatly influences host biochemistry and sus-ceptibility to disease (6).

Interactions between the gut microbiota andthe host immune system begin at birth. The mi-crobiota shapes the development of the immunesystem, and the immune system in turn shapesthe composition of the microbiota. This cross-talk between the microbes and the host immune

system is transmitted through a vast array of sig-naling pathways that involve many differentclasses of molecules and extend beyond the im-mune system. These immune-mediated signalingprocesses, together with direct chemical inter-actions between the microbe and host, act uponmultiple organs such as the gut, liver, muscle, andbrain. Together these complex interactions com-prise a series of host-microbe metabolic axes. Wedefine a host-microbe metabolic axis as a multi-directional interactive chemical communicationhighway between specific host cellular pathwaysand a series of microbial species, subecologies,and activities. Within these metabolic axes,multiple bacterial genomes can sequentiallymodulate metabolic reactions, resulting in com-binatorial metabolism of substrates by the mi-crobiome and host genome, exemplified byproduction of bile acids, choline, and short-chainfatty acids (SCFAs) that are essential for hosthealth (6). In addition, the production of thesemetabolites by microbes contributes to the hostmetabolic phenotype and hence to disease risk.The profound influence of the gut microbiota onthe host immune system is strongly associatedwith long-term health prospects [see Review byHooper et al. in this issue (7) and Review byBlumberg and Powrie (8)]. The composition ofthe core gut microbiota is considered to be es-sentially stable throughout adulthood. However,there are components that are dynamic and bio-logically and metabolically flexible, respondingto perturbations such as environmental stresses orchanges in diet by alteration in species composi-tion that may influence health or disease risk (9).

The increased incidence of gut dysbiosis (animbalance in the intestinal bacteria leading todisease) in western populations over the past60 years is associated with a variety of factors,ranging from the now-textbook story of gastriculcers caused by infection with the bacteriumHelicobacter pylori to life-style-related diseasessuch as diabetes and obesity. Hence, there ismuch interest in developing new therapeutic tools

1Biomolecular Medicine, Department of Surgery and Cancer,Faculty of Medicine, Imperial College London, Exhibition Road,South Kensington, London SW7 2AZ, UK. 2Institut National dela Santé et de la Recherche Médicale, U1048, and Institut desMaladies Métaboliques et Cardiovasculaire I2MC, Rangueil Hos-pital, BP84225, 31432 Toulouse, France. 3Department of Foodand Nutritional Sciences, The University of Reading, White-knights, Reading, UK. 4Department of Nutrition, University ofNorth Carolina at Greensboro, North Carolina Research Campus,Kannapolis, NC 28081, USA. 5Department of Microbiology,Tumor, and Cell Biology (MTC), Karolinska Institutet, Stockholm117 77, Sweden, and School of Biological Sciences and NationalCancer Centre, 11 Hospital Drive, Singapore 169610.

*To whom correspondence should be addressed. E-mail:j.nicholson@imperial.ac.uk (J.K.N.); sven.pettersson@ki.se (S.P.)

8 JUNE 2012 VOL 336 SCIENCE www.sciencemag.org1262

on

July

1, 2

012

ww

w.s

cien

cem

ag.o

rgD

ownl

oade

d fr

om