Flow chart of samples collection and analysis.
In order to elucidate a potential interaction between mesenteric adipose tissue (mAT) microbiome and host responses in CD patients, a prospective cohort of 48 patients was recruited consisting of individuals with diagnosed CD. For non-CD controls, we recruited 16 patients diagnosed with colorectal cancer but CD free based on non-specific symptoms, endoscopic and histopathologic findings. All mAT from terminal ileum were resected under sterile conditions and further assessed using different omics datasets: 16S rRNA gene sequencing, human RNA sequencing (RNA-seq), metabolomes, isolation and culture (Fig. S1). After surgery, colonoscopy was performed to assess the endoscopic recurrence according to the Rutgeerts score.
Differential mAT microbiome associates with changes in host transcriptome and metabolome in CD patients.
To exclude the reagent and laboratory contamination, 4 negative sequencing controls (tissue-free blanks processed with the same DNA extraction, PCR amplification and sequenced on the same run) were introduced as quality controls. As expected, no bacterial DNA was detected in quality controls, which was significantly separated from the mAT from patients (Fig. S2a). Taxonomic analysis of mAT microbiome using principal component analysis (PCA) and principal coordinates analysis (PCoA) of 16S rRNA gene sequencing data showed that the mesenteric microbiome from CD and non-CD controls exhibited slight but significant separation (Fig. 1a and Fig. S2b). The α-diversity of microbiome was similar between mAT from CD and the non-CD controls (Fig. S2c and d). Then we applied linear discriminant analysis of effect size (LEfSe) to detect marked difference in the predominance of bacterial communities between CD and non-CD control. Multiple families such as Enterobacteriaceae, Micrococcaceae and Alcaligenaceae were enriched in mAT from CD patients, while the families such as Sphingomonadales, Sphingobacteriaceae and Rhodospirillales were enriched in those from non-CD controls (Fig. 1b). To further clarify the microbial differences, we compared the bacteria at the genus level. As shown in figure (Fig. S2e and f), twenty genera were significantly upregulated in mAT from CD, while five genera were significantly upregulated in non-CD controls. These results revealed the presence of a unique microbial signature in mAT from CD patients.
To further explore the role of the unique mAT microbiota in host alteration, mesenteric tissue (CD: n = 46, non-CD: n = 15; see methods) were incorporated into our analysis of the host transcriptome. The pattern of transcriptomic features significantly separated CD subjects from non-CD controls (Fig. 1c). We subsequently identified the significantly differentially expressed genes (DEGs) (FDR < 0.05), which was comprised of 512 upregulated and 229 downregulated DEGs. Several DEGs clustered in pathways of extracellular matrix (ECM)-receptor interaction and focal adhesion were significantly downregulated in mAT from CD patients (Fig. 1d), which was consistent with the previous study[15]. Strikingly, variance of transcriptome between CD and non-CD controls was tightly coupled with the changes of microbiome (Fig. S3a and b). Differential mesenteric microbiota between CD and non-CD controls also exerted significant effects on variance of the mesenteric transcriptome (Fig. S3c). These findings illustrated that variance of mesenteric microbiota in mAT of CD patients was associated with the alteration of host transcriptome.
Difference in mAT between patients with CD and those non-CD was also apparent at the metabolome level (CD: n = 48, non-CD: n = 16) (Fig. 1e). Out of all annotated mesenteric metabolites, 28 mesenteric metabolites were significantly elevated in CD relative to non-CD controls, while 19 were significantly depleted in CD patients (Fig. 1f and Fig. S4a). Detailed information about these differential metabolites were shown in Table S1. These differential metabolites were significantly correlated with the predominant bacterial communities in mAT (Fig. S4b). Similarly, the microbes with significant difference were also associated with the variance of metabolome in mAT (Fig. S4c). These results indicated that a specific pattern of metabolites in mAT from CD patients might be caused by the changes in microbiota. To further dissect interactions between host and microbiota that might underlie mesenteric features in CD, we constructed a large-scale network that incorporated microbiome, differential metabolites and immune-associated DEGs. Although strong connections were identified in both CD and non-CD groups, the correlation network in mAT from CD was clearly different from that of non-CD controls (Fig. S5a and Table S2).
To better assess the interplay between microbiome, host transcriptome and metabolome, we incorporated clinical variables prospectively assembled from our cohort. In addition to gender, age and diagnosis, contemporaneously collected information for each patient included disease type, disease activity, Limberg score, medication history (steroid, mesalamine and immunotherapy) and trajectories of disease progression with blood parameters (Table S3). Although patients with CD were significantly younger than those controls, age of patients did not significantly affect any omics datasets. Strikingly, the clinical variables associated with CD disease, such as group (CD or non-CD), disease type (locations of inflammation in CD group), Limberg score, as well as disease activity explained significant effects of the variance for those 3 omics datasets (Fig. S5b). Therefore, our data suggested that alteration of host transcriptome and host-derived metabolites were associated with the changes of microbiota in mAT, which was correlated with CD pathogenesis.
Presence of Proteobacteria in mAT is associated with the development of CD.
To evaluate the clinical significance of the mAT-associated microbiome in CD, we next applied random forest (RF) and LASSO logistic regression models to determine the role of microbiota in CD. The RF model that maximized the strength of the CD prediction identified 16 important families in mAT (Fig. 2a and Fig. S6a). The model based on these important families exhibited an accuracy of 0.91 for CD/non-CD stratification (Fig. S6b), which achieved an area under the receiver-operating characteristic (ROC) curve (AUC) of 0.986 to distinguish a patient with CD or non-CD (Fig. 2a). Interestingly, half of these important units belonged to the Proteobacteria phylum (Fig. S6c). Importantly, the microbiome selected from the RF model also exhibited an outstanding performance to predict post-operative endoscopic recurrence in CD patients (AUC = 0.852) (Fig. 2b). Similar results were obtained using the LASSO logistic regression classifier (shown in Fig. S6d-g), with an excellent ROC of 0.99 to detect CD patients and an ROC of 0.835 to predict those with endoscopic recurrence (Fig. S6h). These results implicated the important role of mAT-associated microbiota in CD development.
To isolate mAT resident bacteria, we anaerobically and aerobically cultured the mesenteric tissue using different culture mediums (Fig. 2c). Specifically, 174 colonies were picked from the various culture mediums and identified by 16S rRNA gene sequencing, leading to the identification of 32 species belonging to 13 different families presented in mAT from CD (Fig. 2c and Table S5). Consistent with our results described above, eight of these 13 families belonged to Proteobacteria phylum. We next carried out correlation analysis of these candidate bacteria with metabolites and immune-associated DEGs. Four families exclusive to CD patients and coexisting in the same mAT, including Alcaligenaceae, Brucellaceae, Pseudomonadaceae and Hyphomicrobiaceae, exhibited a significantly co-varying relationship with a variety of host transcripts and metabolites (Fig. 2d and Fig. S7). Thus we identified 5 candidate strains (5-mix), including Achromobacter pulmonis (A. pulmonis from Alcaligenaceae), Ochrobactrum anthropi (O.anthropi from Brucellaceae), Pseudomonas alcaliphila (P.alcaliphila from Pseudomonadaceae), Achromobacter deleyi (A.deleyi from Alcaligenaceae), and Devosia riboflavina (D.riboflavina from Hyphomicrobiaceae) (Table S5) to represent the major pathogenic members of mesenteric microbiota. In addition, higher abundance of these 4 candidate families of bacteria in mAT was significantly associated with higher risk of CD patients with endoscopic recurrence (Fig. 2e), indicating the clinical significance of these bacteria in colitis development.
Colonization of a defined mesenteric resident bacteria consortium exacerbated colitis in mice.
To functionally link these mesenteric resident bacteria to intestinal inflammation, we colonized antibiotics-treated specific pathogen-free (SPF) mice with our 5 candidate strains (5-mix) and examined effects on DSS-induced colitis (Fig. 3a). Mice colonized with the commensal bacteria, Escherichia fergusonii, isolated from SPF wide type C57BL/6J mice was used as controls, which has been demonstrated to be not virulent in a mouse model[19, 20]. These 5 bacteria were capable of colonizing the intestinal lumen as measured by 16S rRNA PCR amplification (Fig. S8a-f). The body weight of mice in those 3 groups were comparable before DSS administration (Fig. S8a). DSS-mediated body weight loss was significantly exacerbated in mice colonized with 5-mix bacteria compared to mice colonized with E. fergusonii or culture media alone (BHI) (Fig. 3b). Additionally, DSS-induced colitis as measured by disease activity index (DAI) and colon length was significantly augmented in mice colonized with 5-mix bacteria compared to mice colonized with E. fergusonii or BHI (Fig. 3c-e). These clinical parameters correlated with exacerbated histological assessment of colonic inflammation characterized by increased mucosal erosion, crypt destruction and inflammatory cell infiltration in the mice colonized with 5-Mix bacteria compared with mice colonized with E. fergusonii or BHI group (Fig. 3f and g). Consistent with the histological alteration, colonic mRNA expression of TNF-α, IL-6 and IL-1β were significantly elevated in mice colonized with 5-mix bacteria (Fig. 3h). Results were validated in a spontaneous colitis Il10−/− mouse model (Fig. 3i). Mesenterium-resident bacteria significantly induced shortening (Fig. 3j and k) and destruction of colon tissue (Fig. 3l and m). Collectively, our results demonstrated that the consortium of 5 mesenteric resident bacteria exacerbated DSS-induced colitis.
As our previous transcriptomic analysis indicated the down-regulation of ECM-receptor interaction and focal adhesion in mAT, we subsequently evaluated these transcripts in mice treated with 5-mix bacteria. Several transcripts associated with focal adhesion, such as Egfr, Itga9, Lama2, Pdgfra and Thbs2, were significantly down-regulated in mAT from mice treated with 5-mix bacteria (Fig. S8g). In addition, mRNA expression of genes associated with ECM-receptor interaction including Col1a1, Col 6a1, Col 6a2, Col 6a3, FN1 and Tnxb were decreased in mAT from mice treated with 5-mix bacteria (Fig. S8h). Interestingly, these transcriptional changes were similar to those observed in CD patients. These data implicated the impact of these 5 bacteria on the remodeling of mesenteric structure.
To examine the in vitro inflammatory potential of the 5-mix as well as contribution of individual bacteria, we co-cultured these bacteria with murine RAW264.7 macrophages. Consistent with the results observed in mice, the 5-mix bacteria induced significant secretion of inflammatory cytokines (IL-1β, IL-6 and TNF-α) from macrophages. Although the inflammatory capacity varied among the individual bacterium, we observed that the Achromobater pulmonis (A. pulmonis) exhibited a significant effect in pro-inflammatory response (Fig. 3n). The secretory pattern of inflammatory cytokines from A. pulmonis- treated macrophages was even similar to those treated with 5-mix (Fig. 3n). In addition, A. pulmonis also demonstrated strong inducibility of cytotoxicity activity in pre-adipocytes 3T3-L1 or epithelial cell IEC6 by detection of lactate dehydrogenase (LDH) (Fig. S9a and b). We therefore hypothesized that the A. pulmonis strain isolated from mAT of CD patients might be the major contributor of exacerbated colitis in vivo.
Mesenteric resident Achromobacter pulmonis exacerbates colitis
To verify whether colonization of A. pulmonis alone was sufficient to exacerbate colitis, we compared the efficacy of A. pulmonis versus the remaining 4 strains (4-mix) from the 5-mix and 5-mix bacteria using the DSS-induced colitis model (Fig. 4a). A. pulmonis-colonized mice exhibited comparable weight loss and shortening colons to the mice treated with 5-mix, while the weight loss (Fig. 4b) and shortening colons were moderate in 4-mix-treated mice (Fig. 4d and e). DAI scores in A. pulmonis group and 5-mix group were comparable following DSS exposure, which were both significantly worse than those of the 4-mix group (Fig. 4c). Further, histological analysis revealed that both A. pulmonis- and 5-mix-colonized mice displayed severe loss of crypt architecture and extensive inflammation (Fig. 4f and g). In contrast, 4-mix-colonized mice had reduced intestinal tissue damage and inflammation. Consistently, fecal presence of A. pulmonis in A. pulmonis- or 5-mix-colonized mice were significantly higher than other groups (Fig. S9c). To further verify the pro-inflammatory role of A. pulmonis, we colonized SPF Il10−/−mice with A. pulmonis (109 CFU/mouse) daily after treatment of a broad spectrum antibiotics cocktail (Fig. 4h). Consistent with DSS-induced colitis, Il10−/− mice colonized with A. pulmonis exhibited a significant shortened colon (Fig. 4i and j) and developed more severe colitis after 3-week colonization compared to mice colonized with E. fergusonii or BHI (Fig. 4k and l). In line with the histopathologic changes, A. pulmonis induced higher colonic gene expression of pro-inflammatory cytokines, such as IL-1β, TNF-α and IL-6 in Il10−/− mice (Fig. S9d). In the Il10−/− spontaneous colitis model, A. pulmonis colonization significantly downregulated gene expression of several intestinal barrier molecules including Muc2, ZO-1 and Occludin (Fig. S9e). These data suggested that A. pulmonis was sufficient to directly exacerbate colitis in both mouse models.
In order to explore pro-inflammatory response of A. pulmonis, we co-cultured A. pulmonis with murine macrophages cell line RAW264.7, rat intestinal epithelial cell line IEC6 and mouse pre-adipocytes cell line 3T3-L1, and the expression of pro-inflammatory cytokines and chemokines were assessed by RT-PCR. In addition, the accumulation of IL-6, TNF-α, IL-1β and MCP-1 mRNA were also significantly increased in cells exposed to A. pulmonis compared to controls (Fig. S9f-h). Taken together, these results indicated that A. pulmonis could directly induce inflammatory response to mediate colitis development.
Bacterial translocation into mAT is associated with colitis.
CD is characterized by increased gut permeability, which could lead to bacterial or bacterial components translocation into the extra-intestinal compartment, such as mAT. The mucous layer is part of the intestinal barrier controlling bacterial translocation. To explore the impact of mAT-associated bacteria on intestinal barrier, we evaluated various molecules associated with intestinal barrier. Although expression of different tight junctional molecules was relatively stable, the mRNA level of Muc2 was significantly downregulated in mice colonized with A. pulmonis without DSS intervention (Fig. S10a-d), suggesting association between A. pulmonis and mucous impairment. Consistently, the mucous layer thickness from A. pulmonis-colonized mice was significantly decreased compared to those from control groups in DSS-induced colitis (Fig. 5a and b). Furthermore, increased presence of A. pulmonis was observed in extra-intestinal mAT in DSS colitis model (Fig. 5c). These results suggest that bacterial translocation was associated with impairment of intestinal mucous layer by mesenteric bacteria from CD patients.
To further confirm the bacterial translocation into mAT in CD patients, we performed fluorescence in situ hybridization (FISH) using A. pulmonis-specific probes and general eubacterial probes under sterile conditions visualize and quantify A. pulmonis in human mAT. This analysis showed increased presence of A. pulmonis in the mAT from CD patients (Fig. 5d and Fig. S11), but decreased in those from non-CD control (Fig. S11). Furthermore, we quantified A. pulmonis 16S rRNA gene in human mAT using quantitative RT-PCR and found that A. pulmonis were significantly higher in mAT from CD patients compared with those in non-CD controls (Fig. 5e), whereas the total bacterial DNA was also higher in mAT from CD patients than those from non-CD controls (p = 0.0591) (Fig. S12c). As for the patients with endoscopic recurrence, the level of A. pulmonis in mAT from patients with endoscopic recurrence was significantly higher than those without endoscopic recurrence (Fig. 5f).
Although the colitogenic role of A. pulmonis from mAT has been evaluated, their role in mesenteric translocation is limited. To assess the potential source of mAT associated microbiota, we retrieved fecal 16S rRNA gene sequencing data from RISK cohort[21]. Similar to the result in our cohort, the abundance of Alcaligenaceae, Brucellaceae, Pseudomonadaceae and Hyphomicrobiaceae were significantly up-regulated in feces from CD than those from non-CD controls (Fig. S12a). More specifically, we found that the abundance of Achromobacter spp. from Alcaligenaceae was also significant higher in feces from CD than those from non-CD controls (Fig. S12b). To further evaluate the mesenteric translocation of A. pulmonis, a new clinical cohort including mATs and corresponding mucosal tissue from 12 CD patients, as well as the mesenteric and mucous tissue from 6 non-CD controls, were recruited for validation analysis. Colonization of A. pulmonis in mAT by qPCR detection was significantly higher than those from non-CD controls (Fig. 5g). Consistently, the signal of total bacterial DNA and A. pulmonis were both significantly higher in corresponding mucosal layer from CD patients than those from non-CD controls (Fig. 5h and Fig. S12d). These data suggested that high colonization of A. pulmonis in mAT was associated with the bacterial translocation from gut microenvironment.