Supplementary Methods:
Microbial DNA Extraction and Analysis
Sequences were parsed based on barcodes using a customized Perl script. Data was then quality trimmed and aligned to a SILVA reference database (version SSUref_102) using mothur (version 1.12.3)[21]. This resulted in removal of sequences which were less than 250 base pairs long, and/or which contained ambiguous bases or homopolymers extending eight or more nucleotides. The ChimeraSlayer algorithm[22] was used to remove potentially chimeric sequences from the dataset. Additionally, a pre-cluster step was performed, as has been previously recommended by Huse et al., allowing detection and removal of rare spurious sequences[23]. Individual sequences were then assigned to taxonomic outcome groups, at the genus level where possible, using the Silva taxonomy with mothur[24]. A Bayesian classification approach with 100 iterations, and an applied bootstrap cutoff of 60 was used to assign sequences to a taxonomic outcome groups.
Statistical analysis
Taxonomic groups which were detected in less than 5% of the samples were excluded from further analysis. Analyses were performed separately on pouch and afferent limb samples in order to avoid falsely inflating significance. Results were first dichotomized based on whether or not a given genera was detected within a sample. Fisher’s exact test was applied to determine whether there were statistically significant differences between groups. Results at the genus level were confirmed using exact logistic regression with smoking, country of birth and gender included as covariets to further validate results. Continuous abundance was calculated by normalizing the sequences for each taxon against the total number of sequences detected in a sample[25]. Genera abundance was assessed for significance using the non-parametric Kruskal-Wallis and Wilcoxon signed rank test. FDR corrected p-values below .05 were considered significant in both the dichotomized and continuous abundance analyses. All statistical analyses were conducted using STATA version 11.1 (StataCorp Texas, USA) and R version 2.13.1 (R Foundation for Statistical Computing, Vienna, Austria). Additional analyses were carried out using the Linear Discriminant Analysis (LDA) Effect Size (LEfSe)[26] tool (version 1.1.0) to get a more robust understanding of differences in community structure across phenotypic outcome groups (Significance threshold P<.05).
Quantitative PCR (qPCR)
Organisms previously implicated in inflammation including F. prausnitzii, Clostridial cluster IV, AIEC, and Roseburia were evaluated using qPCR with previously described primer pairs (Supplementary Table 2). Relative abundance was estimated as has been previously described[27], with Eubacterial primers used to construct a standard curve. Reactions were carried out using the Power SYBR green PCR master mix (Applied Biosystems, Foster City, CA) with the AB 7300 system and sequence detection software (version 1.3; Applied Biosystems, Foster City, CA). Results from individual samples were log2 transformed and analyzed using Kruskal-Wallis test. Additionally, primers specific for the Bacteroides genus were used to confirm pyrosequencing results by evaluating the correlation between dichotomized (Matthew’s) and frequency (Spearman) pyrosequencing and qPCR data. Dichotomous results were generated based on a pyrosequencing detection limit of 10-4 sequences per sample, estimated from the theoretical number of sequences possible per sample in the multiplexed pyrosequencing reaction