213 score was treated as a continuous value. To determine whether these interactions were observed by chance, we also performed permutation tests that randomly shuffled the dysbiosis score 100 times across all samples, and then repeated the interaction models. On average, only three FDRadjusted significant results were obtained for each round of permutation testing, suggesting that the rate of total false positives was approximately ~ 0.014 (3/204). Gene ~ intercept + taxa + dysbiosis + taxa * dysbiosis + inflammation + location + age + sex + BMI + medication + batch Fourth, enrichment of specific intestinal cell types was inferred from the RNA-seq data using the Xcell package in R. The effects of tissue location, inflammatory status and type of IBD diagnosis on expression levels of mucosal cell types were assessed using linear models, adjusting for age, sex, BMI, batch and medication usage. Subsequently, we used the glmnet R package to investigate the variation of cell type–enrichment that could be explained by the mucosal microbiota using lasso regression while employing a nested 10-fold cross-validation using six models: Cell enrichment ~ age + gender + BMI + batch Cell enrichment ~ medication (aminosalicylates, thiopurines, steroids, biologicals) Cell enrichment ~ inflammation Cell enrichment ~ tissue location Cell enrichment ~ bacteria abundance Cell enrichment ~ full factors mentioned above The percentage of explained variance (R2) was calculated to estimate the variation in cell type– enrichment explained by the mucosal microbiota. All analyses were corrected for multiple testing using a FDR significance threshold of 0.05. All gene pathway enrichment analyses were conducted using the Reactome database from MsigDB.105,106 Replication in the HMP2 dataset RNA-seq and 16S raw data were obtained from https://ibdmdb.org and reprocessed using the same pipeline in this study. After harmonizing with the phenotype file, we included 152 intestinal biopsies from the 85 patients with CD, 46 patients with UC and 45 non-IBD controls. First, gene expression and mucosal microbiota patterns were compared separately between this study and HMP2. Second, given the limited overlap in clinical phenotypes between the two cohorts, we restricted the replication analysis to inflammation-related host–microbiota interactions. Individual gene–microbiota associations were calculated using the same linear models used in this study while adjusting for age, gender, tissue location and inflammation status. Spearman correlation coefficients were used to assess the concordance between the Z-scores of gene–microbiota associations from the two studies. 2. Mucosal host-microbe interactions in IBD 1. 2. 3. 4. 5. 6.
RkJQdWJsaXNoZXIy MjY0ODMw