Dysbiosis has been demonstrated in Crohn's disease (CD) patients using feces, colonic biopsies, and ileal biopsies. The mucosa-associated microbiome has been reported to be more predictive of disease than the fecal microbiome. However, biopsies as a sampling method for microbiome studies are limited by the difficulty of characterizing bacterial proteins and metabolites, which are masked by the preponderance of human material. We have previously demonstrated that lavage sampling of the mucosal-luminal interface (MLI) - representing the surface loose colonic mucus and adherent microbes after bowel preparation - yields suitable material for bacterial proteomics and metabolomics. We undertook a study to characterize the MLI microbiome of CD patients relative to healthy controls and its relationship to disease phenotype, location, and genetics.Methods:
Eighty-eight CD patients and 110 healthy controls undergoing colonoscopy were recruited at the Cedars-Sinai Medical Center. All CD patients had mucosal healing by endoscopic appearance. The mucosal surface of the cecum and sigmoid colon were rinsed with sterile saline during colonoscopy. DNA was extracted from pelleted microbes by bead beating and the V4 region of 16S ribosomal RNA was sequenced by HiSeq to a mean depth of 575,708 sequences/sample. Alpha and beta diversity analysis were performed in QIIME. Microbial abundances were fitted to negative binomial models for the cecum and sigmoid colon data. Diagnosis (CD versus healthy), disease behavior, disease location, sex, obesity, and CD genetic risk score (an aggregate measure calculated from IBD-associated polymorphisms genotyped by Immunochip) were covariates in these models. Random forests classifiers were used to predict CD status.Results:
CD patients had lower microbial diversity and altered composition compared to controls in both the cecal and sigmoid MLI microbiome. High-depth sequencing revealed 759 differential operational taxonomic units (OTUs) in the sigmoid and 687 in the cecum between CD and controls, predominantly representing enrichment of members of the Proteobacteria phylum and depletion of Firmicutes. Random forests classifiers based on microbial composition had high accuracy for CD diagnosis, with area under the curve of 0.91 in the sigmoid colon and 0.93 in the cecum. B3 (penetrating) and B2 (stricturing) phenotypes were associated with 329 and 321 differential OTUs, respectively, in the cecum relative to B1. Location had a comparatively small effect, with 31 differential OTUs in the cecum between L1 (ileal) and L3 (ileocolonic) and 55 between L2 (colonic) and L3. Healthy controls in the highest quartile of CD genetic risk were found to have 229 differential features in a pooled analysis of cecum and sigmoid, with 88 showing concordant changes in CD versus healthy including depletion of Faecalibacterium prausnitzii.Conclusions:
CD had a distinctive MLI microbiome that differentiated patients by disease behavior and location. We further demonstrated that genetic risk factors for CD influenced microbial composition in healthy controls, supporting recent findings from studies of IBD-associated single nucleotide polymorphisms in genes such as FUT2 and NOD2 that have indicated that genetic variation can garden the microbiome. These results support MLI sampling as a strategy for studies investigating the mucosal microbiome of IBD patients.