We applied the 16S rRNA tag pyrosequencing strategy of Sogin et al. [39
] to characterize bacterial populations in the distal human gut, which constitute the predominant community of the human microbiota and one of the most densely populated microbial habitats on Earth. This approach revealed both the pervasive effects of an antibiotic that is considered to be relatively benign for the gut microbiota, and the resilience of the human gut microbiota following perturbation. As shown with marine, soil, and macaque gut samples [39
], massively parallel pyrosequencing permits the exploration of microbial diversity in complex communities to an unprecedented depth. In comparison to clone libraries and traditional dideoxy sequencing, tag pyrosequencing detects more taxa, and provides more-accurate estimates of the relative abundance of a large number of moderate- and low-abundance taxa. Such data will facilitate better ecological studies of the human microbiota, with immediate clinical application. For example, the composition of the microbiota at the start of antibiotic treatment may determine the likelihood of pathogen overgrowth and life-threatening antibiotic-associated diarrhea (AAD) [72
]; investigation of this hypothesis will depend on comprehensive characterizations of the community over time, including both rare and abundant taxa. Of the four bacterial pathogens associated with severe AAD, tags matching Klebsiella oxytoca
and Clostridium perfringens
were detected rarely in the current study (nine tags in two subjects and one tag, respectively); tags matching Clostridium difficile
or Staphylococcus aureus
were not found. The rarity of these pathogens and the absence of severe AAD in the current study is not surprising, since only a small number of healthy young adults were involved. It would be of tremendous interest and practical value to conduct a similar study of stool samples collected over time from elderly patients in hospitals or long-term care facilities, which have higher frequencies of both antibiotics use and serious complications from such use.
Cp is reported to have a lower rate of common gastrointestinal side effects than some other broad-spectrum antibiotics, and effects on gut microbial diversity (as detected by a low resolution “fingerprinting” technique) were less pronounced for Cp than for clindamycin or amoxicillin-clavulanic acid [25
]. Nonetheless, the relative abundance levels of about 30% of the taxa in the gut were affected by Cp treatment when the comparison was made across all individuals, and additional taxa were found to respond to Cp within a single participant. Some of these changes may be direct effects of varying sensitivity to the antibiotic among the taxa comprising the gut microbiota. The refOTUs that increased in relative abundance following Cp treatment may represent taxa with intrinsic resistance to the antibiotic, strains that are typically Cp-sensitive but that had already acquired resistance prior to this study, or strains that developed Cp-resistance due to the current exposure. However, many of the changes in the community are likely to be explained by indirect effects, mediated by ecological interactions among taxa such as resource competition, cross-feeding, or the cooperative lysis of polymeric substrates [74
Despite a pervasive disturbance of the gut microbiota, gut function remained normal as assessed subjectively by the participants, and the community composition in samples taken 4 wk after treatment were within the range of temporal variability of pretreatment samples. While the current study did not include samples showing the temporal features of the return of the community to its prior state, a pyrosequencing-based investigation of this rapid transition in a complex community is clearly of interest, and is continuing in our laboratory. The apparent continuity of gut function supports the hypothesis that the diversity of the microbial community provides functional redundancy [2
]. At the same time, continuity in the predominant metabolic activity of the community, i.e., hydrolysis and fermentation of polysaccharides, does not necessarily imply the continuity of more specialized activities [7
] such as bile transformation [77
], immune modulation [9
], or pathogen resistance due to specific inhibition [79
]. Yet, the rapid return of the gut community to its pretreatment state in each individual suggests that not all communities supporting similar functions are equivalent. A quantitative examination of the metabolic transformations and host interactions of the gut microbiota during perturbation will be necessary to assess functional stability. In addition, it will be important to examine the microbiota at different sites within the intestinal tract in order to appreciate possible local antibiotic effects that are not revealed in fecal specimens. Epidemiological studies associating antibiotic-induced changes in the gut microbiota with chronic diseases [15
] suggest that some consequences of community change may not be evident immediately.
An investigation into the factors responsible for community resilience is warranted. A mixture of selective forces intrinsic to the community (e.g., a competitive hierarchy based on relative growth rates and substrate affinity or interference mechanisms) and imposed by the environment (e.g., composition of the diet and of host-derived substrates) is likely to be involved, but nonselective forces such as re-colonization of the gut lumen from protected environments (perhaps the mucosa) must also be considered.
The current study can be compared to that of Young and Schmidt, who investigated changing bacterial populations in stool samples from a single patient with self-resolving AAD subsequent to amoxicillin-clavulanic acid treatment [82
]. These investigators noted the complete absence of sequences from Clostridium
cluster XIVa on the fourth day of antibiotic treatment (down from 20% on day 0) and a reduction from 33% to 15% of sequences affiliated with the genus Faecalibacterium
; these two groups include the majority of butyrate-producing bacteria in the human gut [83
]. (Butyrate is the preferred energy source of colonocytes.) In contrast, none of the participants in the current study experienced AAD. While Faecalibacterium
refOTUs declined in all participants in response to Cp, each individual had other refOTUs with dominant tags exactly matching known butyrate-producing organisms that maintained or increased their abundance during Cp treatment (e.g., V3_Roseburia
_refOTU_1 in individual B matches Butyrivibrio fibrisolvens
and Roseburia intestinalis
_refOTU_1, in individuals A and C, matches the unnamed butyrate-producing strains SS2/1 and SSC/2 [85
Although the community as a whole showed a substantial return to the pre-Cp composition within 4 wk of the end of Cp treatment, there were examples of taxa that were affected by Cp and did not recover. V3_Clostridiales
_refOTU_23 was the 41st and 52nd most abundant taxon before Cp treatment in individuals A and C, respectively, and present in all pre-Cp samples (means of 206 and 158 tags per sample); it was completely absent from all samples after the start of Cp treatment. This refOTU includes perfect matches to clones obtained from the human gut and from swine manure lagoons, but none to a cultivated bacterium. The most similar cultivated species are Clostridium piliforme
and C. colinum
, both zoonotic pathogens. V3_Bilophila
_refOTU_1 was the 82nd and 79th most abundant pre-Cp refOTU in individuals A and C (means of 60 and 91 tags/sample); it was reduced to an average of less than two tags per sample after Cp treatment began. V3_Bilophila
_refOTU_1 responded differently in individual B, in whom it was absent from the Cp-associated sample but subsequently rebounded to pre-Cp levels. The dominant tag of this refOTU exactly matches Bilophila wadsworthia,
a common gut microbe that is frequently isolated in cases of appendicitis and from intestinal and extraintestinal abscesses [86
]. The clinical significance of these persistent Cp-induced changes is unknown, but some gut bacteria are known to have important health effects despite being present at moderate or low abundance. For example, Oxalobacter formigenes
is the most important known oxalate degrader in the human gut, an activity associated with a reduced risk for the development of calcium oxalate kidney stones [22
]. O. formigenes
is found at low abundance in most children, but is absent from many adults, which is generally attributed to its susceptibility to many antibiotics [22
]. Tags matching O. formigenes
were not detected in the current study.
Certain abundant taxa responded differently to Cp in different individuals, as shown in . This result is not surprising when we consider that antibiotic resistance, even high-level Cp resistance resulting from chromosomal mutations [87
], can be acquired far more rapidly than the rate of evolutionary change in the 16S rRNA gene. Although the participants had not taken any antibiotics in the year prior to the study, their earlier exposure to Cp or other antibiotics is unknown. Once established, antibiotic resistant strains can persist in the human gut for years in the absence of further antibiotic use [28
]. Another potential explanation for divergent Cp responses of the same refOTU between individuals is the existence of indirect effects; the response may be mediated by ecological linkages that differ between individuals.
As shown in the companion paper of Huse et al. [48
], the inferred bacterial community composition in these samples at taxonomic levels from phylum to genus was very similar, whether based on tags from the V6 or V3 variable regions or on full-length 16S rRNA sequences. All three techniques involve PCR amplification, and hence could fail to detect certain phylogenetic groups or lead to skewed levels of tag or sequence relative abundance with respect to the starting material [88
]. However, the similarity of the results obtained from three distinct sets of primers suggests that the PCR bias of these primers (and the cloning bias for full-length sequences) for the bacterial taxa present in this habitat is minimal. The interpretation of these results as a measure of the relative abundance of cells in the gut habitat must still be made with caveats regarding differential cell lysis and the unequal numbers of rRNA genes per genome, as for any 16S rRNA-based technique [88
Clearly, full-length sequences will provide the highest possible phylogenetic resolution for any genetic locus. However, pyrosequencing tags from either of two carefully chosen, short variable regions of the 16S rRNA gene have sufficient resolution to reveal taxonomic richness that exceeds any previously observed for samples of host-associated microbial communities, even though most diversity in these habitats is concentrated at the species and strain level [2
]. This success occurred despite our use of reference databases to define OTUs, which necessarily constrained the resolution of tag pyrosequencing to the diversity already represented in public 16S rRNA sequence databases. A more accurate assessment of microbial diversity that accounts for the novel 16S rRNA variants discovered by pyrosequencing will depend on defining OTUs with reference to the tags themselves (e.g., using tag sequence divergence), with appropriate treatment of potential pyrosequencing errors. A high proportion of tags in the current study have an exact match in the database, due to the substantial sequencing effort that has already been directed to the human gut microbiota [7
]. Hence, the constraint of defining OTUs with respect to a database is not as severe for this habitat as it may be for others [39
]. However, the limitations of full-length 16S rRNA sequence surveys for discovering rare members of the gut biosphere are demonstrated by the contrast between the abundant tags in this study, almost all of which have an exact match in the database, and rare tags, which occur at a range of distances from their nearest database match ().
This study has confirmed the existence of more than 5,600 bacterial taxa in the human gut, exceeding earlier predictions made on the basis of nonparametric richness estimators and full-length 16S rRNA sequencing studies [29
], but the presence of singleton OTUs in our data leads to a still larger predicted richness for this habitat. As has long been appreciated, the highly uneven abundance distribution among taxa in this community (, Dataset S3
) hampers the identification of rare taxa via traditional sequencing approaches. However, coverage values exceeding 99% for our V6 and V3 data correspond to estimates that fewer than 1% of all detectable 16S rRNA tags in these samples belong to taxa that have not yet been detected. Individually, any of the rare, unobserved taxa in these samples are unlikely to exceed 0.001% of all 16S rRNA genes, the relative abundance at which we have roughly 99% confidence of detection. Because so many distinct reference tags are matched exactly by the tags obtained in this study, we conclude that a large number of bacterial taxa previously identified as gut inhabitants in a range of previous 16S rRNA sequencing studies can be present simultaneously or sequentially in a single individual, albeit at widely varying abundance. As suggested in a recent revision to the maxim of Baas Becking, “Everything may be everywhere, but not in equal amounts” [91
]. This result challenges microbial ecologists to account for persistent differences in the composition of the microbiota between individuals, either by ongoing selective forces that vary among individuals such as diet and host genotype, or by factors such as founder effects and specific mutualistic interactions that stabilize community composition. However, any explanation will also have to account for the temporal variability shown by some taxa, and the resilience of the community to perturbations involving a significant fraction of its constituent populations.