Statistical and computational analyses
Microbial pathways were analysed with reference to the UniPathway database (23). The R languages’ vegan package was used for calculating Shannon’s index within sample diversity, α-diversity (24), and Bray-Curtis dissimilarity between microbiome and microbial pathway sample pairs. Similarity between pairs of profiles were calculated as 1-d, where d is the associated Bray-Curtis dissimilarity. These dissimilarity matrices were used to evaluate sample-to sample diversity, β-diversity, between cohorts and timepoints using the permutational multivariate analysis of variance (PERMANOVA) implemented in vegan’s ADONIS function, as well as to visualize the relationship between sample groups using the principal coordinate analysis implemented in R’s phyloseq package (25). When PERMANOVA analysis gave a significant result, the R’s DESeq2 package (26) with test and fitType parameters set to “Wald” and “parametric”, and with pseudocount of 1 addition to phylum-/genus-/species-level MetaPhlAn2 RPKM data, were used to identify differentially-abundant microbiome taxa and microbial pathways. The p-values following Wald test were corrected for multiple testing using the Benjamini-Hochberg method.
Kruskal-Wallis test was employed to assess differences between groups at the taxonomic pathway level. When a significant difference was obtained, the Mann-Whitney U test with a Holm-Bonferroni correction was used for post-hoc comparisons. Wilcoxon Signed-Rank test was used to test differences in alpha-diversity over time using paired baseline and longitudinal follow-up. Spearman correlations were used to determine the relationship between microbiome abundance and clinical traits and inflammatory biomarkers. All statistical tests performed were two-tailed.