212 performed using general linear models (see below). Per sample, the mucosal dysbiosis score was defined as the median Aitchison distance from that sample to a reference sample set of non-IBD controls. Dysbiotic status was defined as being at the 90th percentile of this score.13 Associations between microbial taxa and biopsy inflammation/location: Taxa ~ intercept + inflammation + location + age + sex + BMI + medication + batch + surgical resection Associations between microbial taxa and clinical phenotypes: Taxa ~ intercept + Montreal/anti-TNF therapy + inflammation + location + age + sex + BMI + medication + batch + surgical resection Gene–microbiota interaction analysis We first focused on host inflammation-related genes (n=1,441) to investigate their potential associations with mucosal microbiota. Group-level correlations between gene expression and mucosal microbiota were performed using sparse-CCA using the residuals of genes and microbiota after correcting for age, gender, BMI, inflammation, tissue location and surgical resection separately. Sparse-CCA identifies the PCs from two related datasets that maximize the correlation between the two components. A set of enriched host pathways for all significant components was combined while adjusting for multiple comparisons using the FDR approach. Individual pairwise gene–microbiota associations were assessed by fitting a general linear model while adjusting for age, sex, BMI, inflammation status, tissue location, sequencing batch and medication use (including the use of aminosalicylates, thiopurines and steroids, see below). A gene–microbiota network analysis was visualized using the R package ggview. Individual gene–bacteria associations were determined using the following model: Gene ~ intercept + taxa + inflammation + location + age + sex + BMI + medication + batch Second, we focused on host–microbiota interactions associated with fibrostenotic CD and usage of TNF-α-antagonists. Genes and taxa that were differentially abundant between clinical phenotypes were selected and then served as input for CentrLCC-network analysis using the NetCoMi R package. Hub nodes were defined as those with an eigenvector centrality value above the empirical 95% quantile of all eigenvector centralities in the network. This analysis was done in different groups separately (e.g. users and non-users of TNF-α-antagonists). To assess whether the taxa-associated gene networks were altered between groups, the associated genes for each taxa node were ranked within the total gene set background based on Z-scores. TheWilcoxon test was used to compare the two gene rank lists for each taxa. Third, we assessed whether gene–microbiota associations depend on intestinal dysbiosis by modeling these associations using an additional interaction term in linear models. The dysbiosis Chapter 6 1. 2. 1.
RkJQdWJsaXNoZXIy MjY0ODMw