Diversity of the Human Intestinal
`Microbial Flora
`Paul B. Eckburg,1* Elisabeth M. Bik,2 Charles N. Bernstein,3
`Elizabeth Purdom,4 Les Dethlefsen,2 Michael Sargent,3
`Steven R. Gill,5 Karen E. Nelson,5 David A. Relman1,2,6*
`
`The human endogenous intestinal microflora is an essential ‘‘organ’’ in pro-
`viding nourishment, regulating epithelial development, and instructing innate
`immunity; yet, surprisingly, basic features remain poorly described. We exam-
`ined 13,355 prokaryotic ribosomal RNA gene sequences from multiple colonic
`mucosal sites and feces of healthy subjects to improve our understanding of
`gut microbial diversity. A majority of the bacterial sequences corresponded to
`uncultivated species and novel microorganisms. We discovered significant in-
`tersubject variability and differences between stool and mucosa community
`composition. Characterization of this immensely diverse ecosystem is the first
`step in elucidating its role in health and disease.
`
`The endogenous gastrointestinal microbial
`flora plays a fundamentally important role in
`health and disease, yet this ecosystem remains
`incompletely characterized and its diversity
`poorly defined (1). Critical functions of the
`commensal flora include protection against
`epithelial cell injury (2), regulation of host fat
`storage (3), and stimulation of intestinal an-
`giogenesis (4). Because of the insensitivity of
`cultivation, investigators have begun to ex-
`plore this ecosystem using molecular finger-
`printing methods (5) and sequence analysis of
`cloned microbial small-subunit ribosomal
`E
`^
`RNA genes
`16S ribosomal DNA (rDNA)
`(6–9). However, such studies have been
`limited by the relative paucity of sequenced
`gene fragments, the use of fecal biota as a
`surrogate for the entire gut microflora, and
`little attention given to potential differences
`between specific anatomical sites. In addition,
`variation associated with time, diet, and health
`status have not been adequately described, nor
`have the relative importance and contributions
`of each source (10).
`Surface-adherent and luminal microbial
`populations may be distinct and may fulfill
`different roles within the ecosystem. For ex-
`ample,
`the biofilm-like architecture of the
`
`1Division of
`Infectious Diseases and Geographic
`Medicine, Stanford University School of Medicine,
`Room S-169, 300 Pasteur Drive, Stanford CA 94305–
`5107, USA. 2Department of Microbiology and Immu-
`nology, 299 Campus Drive, Room D300, Fairchild
`Science Building, Stanford CA 94305–5124, USA.
`3Section of Gastroenterology, Department of Medi-
`cine, University of Manitoba, MS 779-820 Sherbrook
`Street, Winnipeg, Manitoba R3A 1R9, Canada. 4De-
`partment of Statistics, Sequoia Hall, 390 Serra Mall,
`Stanford University, Stanford CA 94305, USA. 5The
`Institute for Genomic Research, 9712 Medical Center
`Drive, Rockville, MD 20850, USA. 6Veterans Affairs
`Palo Alto Health Care System, 3801 Miranda Avenue,
`Palo Alto, CA 94304, USA.
`
`*To whom correspondence should be addressed.
`E-mail: eckburg1@stanford.edu (P.B.E.);
`relman@
`stanford.edu (D.A.R.)
`
`mucosal microbiota, in close contact with the
`underlying gut epithelium, facilitates benefi-
`cial functions including nutrient exchange and
`induction of host innate immunity (11). Fecal
`samples are often used to investigate the in-
`testinal microflora because they are easily
`collected. However, the degree to which com-
`position and function of the fecal microflora
`differ from mucosal microflora remains un-
`clear. We undertook a large-scale comparative
`analysis of 16S rDNA sequences to charac-
`terize better the adherent mucosal and fecal
`microbial communities and to examine how
`these microbial communities differed between
`subjects and between mucosal sites.
`Mucosal tissue and fecal samples were ob-
`tained from three healthy adult subjects (A, B,
`and C) who were part of a larger population-
`based case-control study (table S1) (12).
`Mucosal samples were obtained during colo-
`noscopy from healthy-appearing sites within
`the six major subdivisions of the human colon:
`cecum, ascending colon, transverse colon, de-
`scending colon, sigmoid colon, and rectum.
`Fecal samples were collected from each sub-
`ject 1 month following colonoscopy (12). We
`focused on 16S rDNA given its universal dis-
`tribution among all prokaryotes, the presence
`of diverse species-specific domains, and its
`reliability for inferring phylogenetic relation-
`ships (13). The 16S rDNA was amplified
`from samples with polymerase chain reaction
`(PCR) and broad-range bacterial and archaeal
`primers (12). The 7 samples from subject B
`and the fecal sample from subject C yielded
`archaeal products; all 21 samples yielded
`bacterial products. PCR products were cloned
`and sequenced bidirectionally, and numerical
`ecology approaches were applied.
`Initially, a phylotype census was performed
`on each sample (table S2). A total of 11,831
`bacterial and 1524 archaeal near-full-length,
`nonchimeric 16S rDNA sequences were sub-
`jected to phylogenetic analysis. Using 99%
`
`R E P O R T S
`
`minimum similarity as the threshold for any
`pair of sequences in a phylotype (or opera-
`tional
`taxonomic unit) as calculated by
`dissimilarity matrices and the DOTUR pro-
`gram (12), we identified a total of 395 bacte-
`rial phylotypes (Fig. 1). In contrast, all 1524
`archaeal sequences belonged to a single
`phylotype (Methanobrevibacter smithii); these
`archaeal sequences were excluded from fur-
`ther analyses. This remarkable apparent dif-
`ference in diversity of the two prokaryotic
`domains in the gut was reminiscent of results
`from soil and ocean (14).
`Of the 395 bacterial phylotypes, 244 (62%)
`were novel (table S3), and 80% represented
`sequences from species that have not been
`cultivated (12). Most of the inferred organisms
`were members of the Firmicutes and Bacte-
`roidetes phyla (Fig. 1 and fig. S1), which is
`concordant with other molecular analyses of
`the gut flora (6, 7, 9). The Firmicutes phylum
`consisted of 301 phylotypes, 191 of which were
`novel; most (95%) of the Firmicutes sequences
`were members of the Clostridia class. We de-
`tected a substantial number of Firmicutes re-
`lated to known butyrate-producing bacteria
`(2454 sequences, 42 phylotypes) (15, 16), all
`of which are members of clostridial clusters
`IV, XIVa, and XVI. We expected prominent
`representation of this functional group among
`our healthy control subjects, given its role in
`the maintenance and protection of the normal
`colonic epithelium (16). Large variations
`among the 65 Bacteroidetes phylotypes were
`noted between subjects (Fig. 1), as described
`previously (6, 7). B. thetaiotaomicron was de-
`tected in each subject and is known to be
`involved in beneficial functions, including nu-
`trient absorption and epithelial cell maturation
`and maintenance (17). Relatively few sequences
`were associated with the Proteobacteria, Actino-
`bacteria, Fusobacteria, and Verrucomicro-
`bia phyla (fig. S1). The low abundance of
`Proteobacteria sequences (including Esche-
`richia coli) was not surprising, given that
`facultative species may represent È0.1% of
`the bacteria in the strict anaerobic environ-
`ment of the colon;
`this is consistent with
`previous findings (6, 8, 9). Three sequences
`from two subjects (represented by AY916143)
`clustered with unclassified sequences previ-
`ously identified from mammalian gut samples.
`These sequences appear to represent a novel
`lineage, deeply branching from the Cyano-
`bacteria phylum and chloroplast sequences.
`No complex microbial community in na-
`ture has been sampled to completion. In ad-
`dition to its biases and inability to distinguish
`live from dead organisms, the limited sensi-
`tivity of broad-range PCR may hinder detec-
`tion of rare phylotypes. We used several
`nonparametric methods to explore the diversi-
`ty and coverage of our clone libraries. Phylo-
`type richness estimations suggested that at
`least 500 phylotypes would be detected with
`
`www.sciencemag.org
`
`SCIENCE VOL 308 10 JUNE 2005
`
`1635
`
`Genome Ex. 1037
`Page 1 of 4
`
`

`

`R E P O R T S
`
`continued sequencing from our samples
`(Q130, Q300, and Q200 phylotypes in subjects
`A, B, and C) (Fig. 2 and figs. S2 and S3).
`These estimates must be considered as lower
`bounds, because both the observed and the
`estimated richness have increased in parallel
`with additional sampling effort (Fig. 2 and fig.
`S3). Coverage was 99.0% over all bacterial
`clone libraries combined, meaning that one
`new unique phylotype would be expected for
`every 100 additional sequenced clones (18).
`The microbial community appeared more
`diverse in subject B than in A or C, based on
`inspection of the richness and evenness of the
`clone distribution across the phylogenetic tree
`
`(Fig. 1). The Rao diversity coefficient (19),
`which accounts for both phylotype abundance
`and dissimilarity, was indeed higher for B
`than for the other subjects (fig. S7). This
`pattern was not found with traditional, that is,
`Shannon and Simpson, diversity indices,
`which assess only relative phylotype abun-
`dance (20). Within each subject, the mucosal
`samples demonstrated similar diversity pro-
`files, regardless of the index used (fig. S7).
`Previous investigations have not rigorously
`addressed possible differences in the intestinal
`microflora between subjects, between anatom-
`ical sites, or between stool and mucosal com-
`munities. We applied techniques that are
`
`based on the relative abundance of sequences
`within communities and the extent of genetic
`divergence between sequences. We first com-
`pared inter- and intrasubject variability using
`double principal coordinate analysis (DPCoA)
`(19). The greatest amount of variability was
`explained by intersubject differences; stool-
`mucosa differences explained most of the
`variability remaining in the data (Fig. 3).
`The relative lack of variation among mucosal
`sites was further examined. The FST statistic
`of population genetics (21 ) was used to com-
`pare genetic diversity within each subject; this
`revealed that the mucosal populations of sub-
`jects A and B were significantly distinct com-
`
`Fig. 1. Number of sequences per phy-
`lotype for each sample. The y axis is
`a neighbor-joining phylogenetic tree
`containing one representative of
`each of the 395 phylotypes from this
`study; each row is a different phylo-
`type. The phyla (Bacteroidetes, non-
`Alphaproteobacteria, unclassified near
`Cyanobacteria, Actinobacteria, Firmi-
`cutes, Fusobacteria, and Alphaproteino-
`bacteria, ordered top to bottom) are
`color coded as in Fig. 3 and fig. S1.
`Each column is labeled by subject
`(A, B, C) and anatomical site. For
`each phylotype, the clone abundance
`is indicated by a grayscale value.
`
`1636
`
`10 JUNE 2005 VOL 308 SCIENCE www.sciencemag.org
`
`Genome Ex. 1037
`Page 2 of 4
`
`

`

`pared with the overall mucosal diversity (table
`S5). However, in both of these subjects, a
`single mucosal library had a deviant genetic
`diversity index; exclusion of this library from
`the analysis led to an insignificant FST statistic
`in each case (12). Taken as a whole, these re-
`sults confirmed little genetic variation among
`subject-specific mucosal libraries.
`We then asked whether nonrandom dis-
`tributions of phylogenetic lineages accounted
`for any variation among all samples. Using a
`modification of the phylogenetic (P) test
`(12, 21), we found that stool and pooled
`mucosal libraries harbored distinct lineages
`G
`(P
`0.001) (table S5); however, distinct lin-
`eages were not found among the individual
`
`mucosal libraries. We sought further anatomic
`precision in explaining library distinctions
`using the X-LIBSHUFF program (22). We
`found that mucosal clone libraries were similar
`to the other mucosal libraries from the same
`subject, with two exceptions (fig. S6). The
`library from the ascending colon of subject
`A was a subset of every other mucosal pop-
`G
`ulation from that subject (P values
`0.0017),
`and the descending colon library from sub-
`ject B was a subset of the ascending colon
`0
`library in that subject (P
`0.0005). Such in-
`consistencies among mucosal subpopulations
`suggested a pattern of patchiness in the dis-
`tribution of mucosal bacteria rather than a
`homogenous gradient along the longitudi-
`nal axis of the colon. X-LIBSHUFF also re-
`vealed that nearly all mucosal libraries from
`subjects B and C were significantly distinct
`from the corresponding stool library, whereas
`each mucosal library from subject A was a
`
`R E P O R T S
`
`subset of the stool library. We postulate that
`the fecal microbiota represents a combi-
`nation of shed mucosal bacteria and a sepa-
`rate nonadherent luminal population; however,
`these data must be interpreted with caution,
`given the delay between stool and mucosa
`sampling.
`Bacterial diversity within the human co-
`lon and feces is greater than previously de-
`scribed, and most of it is novel. Differences
`between individuals were significantly greater
`than intrasubject differences, with the ex-
`ception of variation between stool and ad-
`herent mucosal communities. Complicating
`this picture is our evidence for patchiness and
`heterogeneity. This patchiness did not display
`an obvious pattern along the course of the
`colon but may reflect microanatomic niches.
`Given that each mucosal sample contained
`a similar distribution of organisms within
`higher order taxa (Fig. 1), the variation we
`
`Fig. 2. Collector’s curves of observed and
`estimated phylotype richness of pooled muco-
`sal samples per subject. Each curve reflects the
`series of observed or estimated richness values
`obtained as clones are added to the data set in
`an arbitrary order. The curves rise less steeply
`as an increasing proportion of phylotypes have
`been encountered, but novel phylotypes con-
`tinue to be identified to the end of sampling.
`The relatively constant estimates of the num-
`ber of unobserved phylotypes in each subject
`as observed richness increases (the gap be-
`tween observed and estimated richness) indi-
`cate that estimated richness is likely to
`increase further with additional sampling. The
`Chao1 estimator and the abundance-based
`coverage estimator (ACE) are similar, but the
`ACE is less volatile because it uses more
`information from the abundance distribution
`of observed phylotypes. Individual-based rare-
`faction curves are depicted in figs. S4 to S6.
`
`Fig. 3. DPCoA for (A) colonic mucosa (solid lines) and stool (dashed lines), (C) colonic mucosal
`sites alone, and (D) mucosal sites excluding Bacteroidetes phylotypes. Phylotypes are represented
`as open circles, colored according to phylum as in Fig. 1. Phylotype points are positioned in
`multidimensional space according to the square root of the distances between them. Ellipses
`indicate the distribution of phylotypes per sample site, except in (A), where all mucosal sites are
`represented by one ellipse. Percentages shown along the axes represent the proportion of total
`Rao dissimilarity captured by that axis. (A) is the best possible two-dimensional representation of
`the Rao dissimilarities between all samples (12). (B) is an enlarged view of (A), depicting the
`centroids of each site-specific ellipse. Subject ellipse distributions remain distinct after stool
`phylotypes (C) and Bacteroidetes phylotypes (D) are excluded from the analysis.
`
`www.sciencemag.org
`
`SCIENCE VOL 308 10 JUNE 2005
`
`1637
`
`Genome Ex. 1037
`Page 3 of 4
`
`

`

`R E P O R T S
`
`observed at the genus or species level may
`be the result of colonization resistance by the
`more abundant members within similar func-
`tional groups (23). Whether the gut micro-
`biota undergoes such nonrandom assembly
`remains unclear.
`Ecological statistical approaches reveal
`previously unrecognized irregularities in the
`architecture of complex microbial commu-
`nities. High-resolution spatial, temporal, and
`functional analyses of the adherent human in-
`testinal microbiota are still needed. In ad-
`dition,
`the effects of host genetics and of
`perturbations such as immunosuppression,
`antimicrobials, and change in diet have yet to
`be carefully defined. We anticipate that micro-
`arrays, single-cell analysis, and metagenomics
`E
`B
`[
`e.g., a
`Second Human Genome Project
`^
`(24)
`will complement the approach we have
`illustrated and hasten our understanding of
`human-associated microbial ecosystems.
`
`References and Notes
`1. L. V. Hooper, J. I. Gordon, Science 292, 1115 (2001).
`2. S. Rakoff-Nahoum, J. Paglino, F. Eslami-Varzaneh,
`S. Edberg, R. Medzhitov, Cell 118, 229 (2004).
`
`3. F. Backhed et al., Proc. Natl. Acad. Sci. U.S.A. 101,
`15718 (2004).
`4. T. S. Stappenbeck, L. V. Hooper, J. I. Gordon, Proc.
`Natl. Acad. Sci. U.S.A. 99, 15451 (2002).
`5. E. G. Zoetendal et al., Appl. Environ. Microbiol. 68,
`3401 (2002).
`6. H. Hayashi, M. Sakamoto, Y. Benno, Microbiol.
`Immunol. 46, 535 (2002).
`7. G. L. Hold, S. E. Pryde, V. J. Russell, E. Furrie, H. J.
`Flint, FEMS Microbiol. Ecol. 39, 33 (2002).
`8. A. Suau et al., Appl. Environ. Microbiol. 65, 4799 (1999).
`9. X. Wang, S. P. Heazlewood, D. O. Krause, T. H. Florin,
`J. Appl. Microbiol. 95, 508 (2003).
`10. M. C. Horner-Devine, K. M. Carney, B. J. M. Bohannan,
`Proc. R. Soc. London 271, 113 (2003).
`11. J. L. Sonnenburg, L. T. Angenent, J. I. Gordon, Nat.
`Immunol. 5, 569 (2004).
`12. Materials and methods are available as supporting
`material on Science Online.
`13. N. R. Pace, Science 276, 734 (1997).
`14. T. P. Curtis, W. T. Sloan, Curr. Opin. Microbiol. 7, 221
`(2004).
`15. A. Barcenilla et al., Appl. Environ. Microbiol. 66, 1654
`(2000).
`16. S. E. Pryde, S. H. Duncan, G. L. Hold, C. S. Stewart,
`H. J. Flint, FEMS Microbiol. Lett. 217, 133 (2002).
`17. L. V. Hooper et al., Science 291, 881 (2001).
`18. Good’s coverage estimates were 99.3%, 97.9%, and
`98.3% for subjects A, B, and C, respectively.
`19. S. Pavoine, A. B. Dufour, D. Chessel, J. Theor. Biol. 228,
`523 (2004).
`20. R. K. Colwell, EstimateS, version 7 (2004) (http://
`viceroy.eeb.uconn.edu/estimates).
`
`Fungal Pathogen Reduces
`Potential for Malaria Transmission
`Simon Blanford,1 Brian H. K. Chan,1 Nina Jenkins,2 Derek Sim,1
`Ruth J. Turner,1 Andrew F. Read,1 Matt B. Thomas3*
`
`Using a rodent malaria model, we found that exposure to surfaces treated
`with fungal entomopathogens following an infectious blood meal reduced
`the number of mosquitoes able to transmit malaria by a factor of about 80.
`Fungal infection, achieved through contact with both solid surfaces and netting
`for durations well within the typical post-feed resting periods, was sufficient
`9
`to cause
`90% mortality. Daily mortality rates escalated dramatically around
`the time of sporozoite maturation, and infected mosquitoes showed reduced
`propensity to blood feed. Residual sprays of fungal biopesticides might replace
`or supplement chemical insecticides for malaria control, particularly in areas
`of high insecticide resistance.
`
`The use of pyrethroid insecticides on bednets
`or on walls and ceilings is the mainstay of
`malaria vector control. However, some forms
`of resistance are a threat to the long-term
`effectiveness of such measures (1, 2). With
`practical implementation of novel molecular
`interventions still years off (3–6), there is a
`pressing need for practical alternatives for
`malaria control (7).
`
`1Institutes of Evolution, Immunology, and Infection
`Research, School of Biological Sciences, Ashworth
`Laboratories, University of Edinburgh, Edinburgh EH9
`3JT Scotland, UK. 2CABI Bioscience at Department of
`Agricultural Sciences, Imperial College London, Wye
`Campus, Wye, Kent, TN25 5AH, UK. 3Division of
`Biology and NERC Centre for Population Biology,
`Imperial College London, Wye Campus, Wye, Kent,
`TN25 5AH, UK.
`
`*To whom correspondence should be addressed.
`E-mail: m.thomas@imperial.ac.uk
`
`Several studies have investigated the use
`of microbial agents for biological control
`of mosquitoes (see 8–10 for reviews). The
`most successful approach has been the use
`of microbial biopesticides, such as Bacillus
`thuringiensis, for control of the larval stages
`(8, 9). Here, we report the potential of fungal
`entomopathogens for indoor use against adult
`mosquitoes to reduce malaria transmission.
`Oil-based formulations of fungal entomopath-
`ogens are a relatively recent innovation that
`enables economically viable spore applica-
`tion in ultralow-volume sprays under a wide
`range of environmental conditions (11, 12).
`These formulations create an opportunity to
`apply fungal pathogens for use on indoor sur-
`faces of houses or curtains where some ma-
`laria vector species rest after a blood meal or
`on bednets to which mosquitoes are attracted
`
`21. A. P. Martin, Appl. Environ. Microbiol. 68, 3673 (2002).
`22. P. D. Schloss, B. R. Larget, J. Handelsman, Appl. Environ.
`Microbiol. 70, 5485 (2004).
`23. J. Fargione, C. S. Brown, D. Tilman, Proc. Natl. Acad.
`Sci. U.S.A. 100, 8916 (2003).
`24. D. A. Relman, S. Falkow, Trends Microbiol. 9, 206
`(2001).
`25. We thank B. Bohannan, M. B. Omary, and S. Holmes
`(Stanford University) for helpful comments on the
`manuscript. This research was supported by grants
`from the NIH (no. AI51259) and Ellison Medical
`Foundation (D.A.R.), Canadian Institutes of Health
`Research and Crohn’s and Colitis Foundation of
`Canada (C.N.B., M.S.), National Science Foundation
`(E.P.), and Defense Advanced Research Projects Agen-
`cy (S.R.G., K.E.N.). Representatives of novel phylo-
`types (AY916135 to AY916390) and all other
`sequences (AY974810 to AY986384) were deposited
`in GenBank.
`
`Supporting Online Material
`www.sciencemag.org/cgi/content/full/1110591/DC1
`Materials and Methods
`SOM Text
`Figs. S1 to S8
`Tables S1 to S6
`References
`
`2 February 2005; accepted 5 April 2005
`Published online 14 April 2005;
`10.1126/science.1110591
`Include this information when citing this paper.
`
`by the odor of the occupant. Fungal entomo-
`pathogens infect through external contact, with
`spores germinating and penetrating through
`the cuticle before proliferating in the hemo-
`coel. Natural mosquito-mosquito transmission
`is unlikely, because this would require contact
`between an uninfected adult and a sporulat-
`ing cadaver. In other pest-control contexts,
`contact with fungal spores in a spray residue
`has proved to be a highly efficient means of
`infecting insects (13, 14) because it does not
`rely on direct hit of the target and enables
`accumulation of high doses of pathogen over
`time through continued or repeat exposure.
`Our experimental system comprised Anoph-
`eles stephensi and the rodent malaria Plas-
`modium chabaudi. Our first experiment was a
`basic mortality screen of eight Hyphomycetes
`fungal isolates from two common species, Beau-
`veria bassiana and Metarhizium anisopliae.
`The specific isolates were selected based on
`their known biological activity, that is, either
`known generalists or those originally iso-
`lated from dipteran hosts (15). The basic
`assay technique exposed blood-fed adult fe-
`male mosquitoes to oil-based spray residues
`inside replicated cardboard pots (16). Mosqui-
`toes were introduced 24 hours after the pots
`were sprayed and, for this initial screen, re-
`mained in the sprayed pots for 14 days. The
`eight fungal isolates varied in virulence to A.
`stephensi (Fig. 1 and table S2), six of which
`9
`produced
`80% mortality by day 14, with
`9
`70% of the cadavers bearing sporulated
`fungi. High mortality by day 14 is encour-
`aging because that is about the time taken
`for Plasmodium to develop from ingested
`gametocytes to infective sporozoites.
`
`1638
`
`10 JUNE 2005 VOL 308 SCIENCE www.sciencemag.org
`
`Genome Ex. 1037
`Page 4 of 4
`
`