We enrolled a total of 310 adult subjects, of whom 112 were male and 198 were female. In this cohort, mean age and BMI was higher, and manifestations of the metabolic syndrome 
, were more common in women than in men (). We performed 16S rRNA multiplex pyrosequencing of V1–V3 amplicons using bar-coded primers on the 454 Titanium platform 
to characterize the fecal microbiota. We obtained 10,357±3764 pyrosequencing reads per sample, with an average read length of 303 bp (Table S1
). Reads were binned into individual samples based on the barcode sequence, and complementary phylogenetic and taxon-based analysis methods were used to compare 16S rRNA sequences across the fecal microbial communities (see Methods). In total, 203 genera were identified in the gut microbiota in the Amish; the 10 most abundant genera accounted for 67% of the reads (Figure S1
). Seven species were each represented by more than 1% of the total sequence reads, including three species in the Bacteroidetes (Prevotella copri, Bacteroides vulgatus
and Bacteroides plebius
) and four species in the Firmicutes (Faecalibacterium prauznitzii, Eubacterium rectale, Eubacterium biforme,
and Roseburia faecis)
). The percent of reads for which a species-level taxonomic assignment could not be made averaged 47% across all the samples (range of 15% to 84%).
Characteristics of subjects enrolled in this study.
All but one of the twenty-five most abundant genera were present in at least 75% of all subjects, and these include members of the Firmicutes, Bacteroidetes, Tenericutes, Actinobacteria, and Proteobacteria (Figure S2
). The prevalence of genera in the gut microbiota follows a bi-modal distribution, with a large peak above zero associated with many genera that are present in a small number of subjects and a second peak of 17 genera that are present in 95% or more of the 310 subjects (Figure S3
). These 17 genera were deemed to comprise the core microbiota 
in the Amish () and represent members of the Firmicutes, Bacteroidetes and Tenericutes.
Core gut microbiota at the genus level (present in ≥95% of subjects).
Despite the substantial overlap in genera present in the gut microbiota of the 310 subjects, a significant amount of inter-individual variation was observed with respect to the relative abundance of both the predominant and rare genera (Figure S4
). For example, the relative abundance of Bacteriodetes across all the samples ranges between 3 and 81%. Phyla-level phylogenetic binning of 16S rRNA data revealed a nominally significant correlation between the Bacteroidetes:Firmicutes ratio and age- and sex- adjusted BMI (r
0.04), but no significant correlation with metabolic syndrome traits [fasting glucose (r
0.19), systolic or diastolic blood pressure (r
0.78 and (r
0.45), respectively), fasting triglycerides (r
0.17), HDL-cholesterol (r
0.36); all adjusted for age and sex) Results from unweighted UniFrac analysis 
also did not distinguish among the subjects based on BMI (Figure S5
Using a random matrix theory-based framework 
we identified three networks of interacting bacteria in the human gut, that correlated with the three gut enterotypes recently reported by Arumugam et al. 
(). In our dataset, these groups appear to represent a strong gradient effect between a few dominant taxa, rather than distinct clusters () because the silhouette width statistic 
does not prefer three (mean silhouette width of 0.3) clusters over two (mean silhouette width of 0.34). Each subject was assigned membership to one of the bacterial groups based on the predominant genus present in his/her microbiota. The most prevalent group (I) in this cohort (47% of subjects: 55% male and 45% female) is dominated by Prevotella
, the most abundant genus in the OOA gut microbiota. The least common group (II) in this cohort (14% of subjects: 23% male and 77% female), is Bacteroides
-dominated. A Firmicutes-dominated group (III; 39% of subjects: 31% male and 69% female) is characterized by diverse Firmicutes genera, Oscillospira
being the most abundant. Group III displays a statistically significant higher Shannon diversity index of both OTUs and named taxa than either network I or II (; p
by Wilcoxon rank-sum test).
Bacterial networks in the human gut microbiota.
UniFrac Principle Co-Ordinates Analysis plot, showing each study sample positioned according to its first two principle coordinates.
The three networks appear to reflect the main large-scale trends in gut microbial populations and suggest that gut bacterial community structures are driven by high abundance and high variability in populations of Prevotella, Bacteroides
, or Firmicutes (with a degree of mutual exclusion between them). Prevotella
appear to co-exist at lower levels if the community is Firmicutes-dominated, but are nearly mutually exclusive when either is abundant in Bacteroidetes-dominated communities. Using community structure as a proxy, we examined the stability of the dominant taxa in 19 second samples from the subjects in this study 2–5 months following initial samples. The dominating taxa in 15 of the 19 subjects (79%) did not change (Figure S6
). The communities became Prevotella-
dominated in 4 subjects where a change was observed; in 3 cases first samples were collected in winter and the second samples were collected 3–5 months later in the spring. These changes may represent seasonal changes in diet that will need to be investigated further.
Our cohort of 310 subjects included 113 nuclear families having two or more members phenotyped. These nuclear families contributed a total of 54 spouse pairs, 76 sibling pairs, and 42 parent-offspring pairs. Community structure concordance rates tended to be lower for the spouse pairs (38.9%) than for the sibling (46.1%) and parent-offspring (52.4%) pairs (p
0.32 for comparison of spouse pairs with the combined set of sibling and parent-offspring pairs). Larger sample sizes will be required to get better estimates of the relative contributions of relatedness and household effects to community structure.
We tested the association of groups (coded as a class variable with 3 levels) with each available clinical factor, while adjusting for age and sex (a 2 df test). Neither BMI nor any metabolic syndrome trait was associated with a particular group (). These data suggest that individual traits associated with obesity and its metabolic complications do not correlate with a major shift in the relative abundance of these two predominant phyla in the Amish, an observation reported in other cohorts 
Regression analyses between bacterial networks and metabolic phenotypes.
To determine if gut community type was associated with occupation, we classified study subjects into occupational classes (farmers, tradesmen, farmer’s wives, teachers/shopkeepers, and unknown/retired) and tested the association of phenotype with each occupational class. In men, the occupation of farming was over-represented among those with the Prevotella-
dominated network (71.4%) compared to those with either the Bacteroidetes
-dominated (21.4%) or Firmicutes
-dominated (36.5%) networks (p
0.002; ). The distribution of networks did not differ significantly within any other occupational class in men or with any occupational class in women ().
Distribution of subjects in each microbiota network, according to occupational class.
The availability of extensive clinical data from the OOA cohort allowed us to evaluate the potential role of the microbiota (both species and OTUs) in obesity and its metabolic derangements using network analysis, independent of enterotype. Twenty-two species of bacteria from the phyla Bacteroidetes, Firmicutes, and Actinobacteria, and four OTUs from the order Clostridiales, displayed both positive and inverse correlations with BMI, serum triglycerides, HDL cholesterol, total cholesterol, fasting glucose levels and C-reactive protein () and with each other (). These bacteria differ in both abundance and prevalence. Collectively, this group of species and OTUs represent between 0.03 and 31% of the total 16S rRNA sequence reads in the 310 subjects. Phylogenetic analysis revealed that the majority of these OTUs, along with the majority of known species associated with clinical phenotype are members of the order Clostridiales (). While a majority of the correlations observed were between one metabolic trait and one taxon, Lachnobacterium bovis
and Anaerotruncus colihominis
were inversely correlated with both high BMI and elevated serum triglycerides. Although these two species of gut bacteria are known to produce short chain fatty acids as end-products of metabolism, a more thorough understanding of the potential impact of these species on host metabolism awaits further studies that will be facilitated by completion of ongoing reference genome sequencing projects for gut-associated bacterial species 
Bacterial species and OTUs correlated with metabolic syndrome phenotype.
In a subset of 32 older (≥59 years of age) obese subjects with more severe manifestations of the metabolic syndrome there was an increase in the relative abundance of three gram-negative bacterial genera, Fusobacterium
, J2–29, and Tenacibaculum
, as compared to younger subjects (Table S2
). It is not possible to determine whether these represent age- or disease-related changes in the gut microbiota 
; however, the increase in the abundance of Fusobacterium
is of interest in light of previous findings that periodontal Fusobacterium
species have been associated with inflammation, atherosclerosis 
and colon cancer