194 Intra-individual microbial dissimilarity was lowest in all our comparative analyses of paired tissue samples (Figure 3D). Hierarchical clustering analysis performed on paired samples demonstrated a clear tendency of these samples to cluster together, a finding that we could also replicate in the HMP2 cohort data (Extended Data Figure S5A).13 Overall, our data demonstrate that the composition of the mucosal microbiota is highly personalized and that inter-individual variability dominates over the effects of tissue location or inflammatory status. We then aimed to identify phenotypic factors that shape the composition of the mucosal microbiota using Hierarchical All-against-All association (HAllA) analysis. This allowed us to study the relative associations between microbial taxa and phenotypic factors and disease characteristics (Figure 3E, Supplementary Table S6). Analysis at bacterial genus level revealed that the main factors correlating with mucosal microbiota composition are stricturing disease in CD (fibrostenotic CD, Montreal B2), usage of TNF-α-antagonists, age at time of sampling, age of onset and the comparisons of patients with CD vs. controls, UC vs. controls and CD vs. UC. In contrast, inflammatory status and tissue location did not show a significant effect, and this was also the case within the HMP2 cohort data (Extended Data Figure S5B). These findings are in line with several previous observations from which age at diagnosis, age at sampling and TNF-αantagonist use emerged as critical determinants of mucosal microbiota composition.22 Distinct host–microbe interaction modules are identified in relation to IBD To capture the main microbial taxa associated with inflammation-associated gene expression, we combined the data and performed sparse canonical correlation analysis (sparse-CCA) on 1,441 inflammation-associated genes and 131 microbial taxa (Figure 4). This approach enabled us to identify gene pathways and groups of microbiota and their potential correlations. In total, we found six distinct pairings of groups of genes with bacterial taxa to be significantly correlated with each other (FDR<0.05, Supplementary Tables S7-S18). To prioritize the individual genes and bacteria involved in the sparse-CCA analysis, we performed individual pairwise gene–bacteria associations, which revealed 312 significant gene–bacteria pairs, with most pairs (94.17%) overlapping with the sparse-CCA results. We then replicated these associations in the HMP2 cohort (Spearman correlation ρ=0.16, P=0.005, Supplementary Table S19, Extended Data Figure S6, Methods). Further details on the most intriguing individual pairwise gene–bacteria associations are discussed in Box 1. Chapter 6
RkJQdWJsaXNoZXIy MjY0ODMw