Link between a harmful E. coli bacteria and severe gut imbalance in Hong Kong Chinese people with Crohn's disease

To investigate whether the presence of AIEC impacted the human mucosa microbiota, 16s ribosomal RNA sequencing was performed on 63 biopsy samples, including 21 biopsies from 18 AIEC positive CD patients (3 pairs of inflamed and noninflamed tissues), 17 biopsies from AIEC negative patients and 25 biopsies from healthy controls. The Chao1 richness in CD tissues was significantly lower than control tissues (p < .001, Wilcoxon rank-sum test), whereas no significant difference in Shannon diversity index was observed between CD and control tissues (p = .19, Wilcoxon rank-sum test, Supplementary Figure 1e). The overall microbiome composition in CD tissues was significantly different compared to control tissues (p = .019, PERMANOVA, Figure 1a). The Shannon diversity index and Chao1 richness in AIEC-positive tissues was significantly lower than AIEC-negative and control tissues (p = .033 and p = .042, respectively, Wilcoxon rank-sum test, Figure 1b). Significant difference in microbiome composition was observed among AIEC-positive CD tissues, AIEC-negative CD tissues and control tissues (p = .023, PERMANOVA, Figure 1a). Subgroup analysis showed that the microbiome composition in AIEC-positive CD tissues was significantly different compared to control tissues (p = .048, PERMANOVA, Figure 1a). No significant difference was observed between AIEC-negative tissues and control tissues (p = .052), as well as between AIEC-positive and AIEC-negative CD tissues (p = .249, PERMANOVA, Figure 1a). The intragroup variation of AIEC-positive microbiota was significantly higher than AIEC-negative tissues (p < .001, Wilcoxon rank sum test, Figure 1c). Linear discriminant analysis effect size (LEfSe) analysis revealed 76 taxa that were differentially abundant between AIEC-positive and AIEC-negative microbiota (LDA>2, p < .05, Figure 1d). AIEC-positive tissues were enriched with Peptoniphilus, Finegoldia and Mycetocola and depleted with Bacteroides, Coprococcus, Sutterella, Roseburia, Megamonas, Delftia, Moraxella, Bacillus, Lactococcus, Anaerostipes, compared with AIEC-negative tissues (LDA>2, p < .05, Figure 1d). At the phylum level, AIEC-positive tissues were depleted with Firmicutes compared with AIEC-negative tissues (p = .045. Wilcoxon rank-sum test, Figure 1e).

Biopsies were obtained from 116 Hong Kong Chinese subjects (60 CD and 56 healthy controls) undergoing colonoscopy at the Prince of Wales Hospital, the Chinese University of Hong Kong. Patients were included if they were 18 years or older with a diagnosis of ileal CD defined by endoscopy, radiology, and histology (ileal or ileocolonic); had active ileal disease at the time of endoscopy and on stable CD-related medication. Subjects with a use of antibiotics, probiotics or prebiotics or a history of enteric infection in the past three months were excluded. Subjects with no prior gastrointestinal disease and a normal colonoscopy as part of screening checkup were recruited as healthy controls. Informed consent was signed by all participants. During colonoscopy, mucosal biopsies were taken from terminal ileum. Tissues were immersed in the cell culture medium MEM supplemented with 15% sterile glycerol, snapped frozen and stored at −80°C until use. This study was approved by the Ethics Committee of the Chinese University of Hong Kong (CRE 2014.026).

Isolation and characterization of AIEC have been previously described.¹⁴ Briefly, the number of E. coli adhering to the intestinal mucosa were counted on Drigalski agar. Bacterial load was normalized by colony-forming units per milligram (CFU/mg) tissue samples. Colonies were cultured in 96-well microplates in Luria-Bertani medium supplemented with 15% glycerol and stored at −80°C. Adhesive and invasive properties were initially evaluated on intestinal epithelial cell line Intestine-407 (ATCC® CCL-6™) by antibiotic protection assay. Regarding the invasion, the results were expressed as percentage of the number of invasive bacteria compared to the number of bacteria present in the initial inoculum, defined as 100%. The survival and replication capabilities of invasive bacteria was measured within THP-1 macrophage (ATCC® TIB-202™). Survival rate was expressed as the mean number of bacteria recovered, compared with LF82, defined as 100%. E. coli strains were considered AIEC if the invasion rate was equal to or greater than 10% that of AIEC reference strain LF82.

For optimal isolation of bacterial DNA, mucosal biopsies were disrupted by bead-beating upon digestion in lysozyme (Sigma Aldrich, Northbrook, IL) before extraction and purification by Maxwell Promega Tissue DNA Purification kit according to manufacturer’s instructions. 16S ribosomal RNA gene amplicon sequencing was performed on DNA extracted from 63 biopsy samples (25 normal and 38 CD), including 21 biopsies from 18 AIEC positive patients (3 pairs of inflamed and noninflamed tissues) and 17 biopsies from AIEC negative patients. Mice fecal DNA was extracted using the QIAamp DNA Mini Kit according to manufacturer’s instructions. Mice colonic tissue DNA was extracted following same protocol as human tissues. The V3-V4 hypervariable region of 16S ribosomal RNA gene was amplified by forward primer 341 F-CCTAYGGGRBGCASCAG and reverse primer 806RGGACTACNNGGGTATCTAAT. Illumina HiSeq 2500 platform (Illumina, San Diego, CA) was used to generate PE 250 bp raw reads by Novogene (Beijing, China). PhiX was included for each sequencing run at a concentration of 15 nMol/ul (50% of the library).

Raw 16S rRNA amplicon reads were processed with FastX-Toolkit to remove reads with quality score below 20. Paired reads were then merged and filtered using VSEARCH²⁹ remove reads with over 1% error rate, yielding an average of 65835 (11054–106171) (min-max) clean reads per sample (66813 (11054–106171) for human samples and 63910 (45479–80168) for mice samples). Remaining reads were converted to FASTA format and subjected to chimera filtering using VSEARCH de novo mode and the rdp_gold.fa database. Remaining reads were classified using the UTAX algorithm with RDP database and reads classified as either chloroplast or not classified at domain level were removed. Operational taxonomic units (OTU) tables were generated using the UPARSE algorithm³⁰ implemented in VSEARCH software and all reads across samples. The remaining reads were then clustered at 97% similarity and annotated by the rdp_16s_v18.fa database. All reads per sample were then mapped to representative sequences using VSEARCH to generate OTU abundance tables. Human samples were rarefied to 12,126 reads One sample from AIEC-positive group with less than the said numbers were removed. Mice samples were rarefied according to the sample with lowest number of reads (45,479 reads per sample). The resulting OTU table and representative sequences were then used to predict the metabolic functional profiles with PICRUSt2 (Phylogenetic Investigation of Communities by Reconstruction of Unobserved States).³¹ Metabolic pathways were annotated using MetaCyc.³²

Random forest (RF) was used to build AIEC-positive versus AIEC-negative prediction model using fecal microbes because of its superior performance for classification with binary features.³³ AIEC-positive and AIEC-negative samples were divided into 80% of training set and 20% test set by random stratified sampling by caret R package. The importance value of each OTU to the classification model was evaluated by recursive feature elimination using the training set. Sequential forward selection was employed to identify most important OTUs. Model performances were compared in terms of binary classifiers with AUC of the test set. This process is repeated 100 times to avoid over fitting. OTUs with highest average importance scores were included in the model. The number of features was chosen when best accuracy for the test set was achieved. These analysis was done using R packages randomForest v4.6–14³⁴ and pROC v1.15.3.³⁵

The dextran sodium sulfate (DSS) colitis model was chosen due to its rapidity, simplicity, reproducibility, controllability, and compatibility for AIEC infection studies.^36,37 Male C57B/L6 mice (7–8 weeks of age) were treated with 2% DSS for 7 days to induce colitis (Figure 4a). In a preliminary experiment, AIEC 62d was able to colonize the mice gut when mice were treated with a 7-day course of 2% DSS and persisted in the gut for at least 21 days after initial inoculation, with or without antibiotic treatment prior to gavage (supplementary Figure 4). Therefore, we gavage the mice with the E. coli strains without antibiotics pretreatment to minimize the disturbance of gut microbiota.

Mice were inoculated with 10⁹CFU of AIEC62d or nonpathogenic control strain K12 C600 for three consecutive days. Another group of mice were included as blank control and stool donors for FMT. At day 8, six mice from AIEC and K12 group were sacrificed as “before FMT” group. One mouse from AIEC group was moribund and euthanatized at day 5 and removed from subsequent analysis. Mice in FMT groups were treated with 200 μl of fecal solution (stool collected from control mice diluted 1:10 with sterile PBS) for three consecutive days. Mice in “no FMT” groups were treated with sterile PBS (n = 8 per group). Bodyweight of the mice was measured daily. Stool samples were collected every two days. DNA was extracted from stool and colonic tissues for 16S rRNA sequencing. For stool samples, pre-FMT and post-FMT samples were sequenced. For colonic tissue samples, post-FMT samples were sequenced (supplementary table 3). One stool sample from pooled FMT solution was sequenced as donor microbiome profile.

After weighing and serial dilutions, stool suspensions were plated on LB agar plates supplemented with 100 mg/L ampicillin and cultured at 37°C overnight to determine the AIEC load in mice gut (CFU/mg). Fecal lipocalin-2 (Lcn-2) were measured by ELISA according to manufacturer’s instruction (R&D). At sacrifice, colon length and histology score in distal colonic segments were measured to assess the severity of colitis. All slides were stained with hematoxylin and eosin (H&E). Blinded histologic scoring was performed. Each slide was assigned four scores by assessment of epithelial damage and inflammatory infiltration into the submucosa, mucosa, and muscularis/serosa, resulting in a total score of 12. Additional colon tissues from AIEC-infected mice were fixed in Carnoy’s solution for Fluorescence in situ hybridization (FISH) to visualize the location of E. coli within the mouse gut mucosa. Tissue sections were stained with 25 μg/mL FITC-labeled E. coli specific probe and nuclei was labeled with DAPI for confocal microscopy imaging (Leica confocal SP2 1P/FCS). High magnification images were obtained under the × 63 objective.

All measurements were shown as mean ± standard deviation. Wilcoxon’s rank-sum test was used to test for differences in numerical variables, and Fisher’s exact test was used to evaluate the proportional difference in categorical variables between groups. Linear discriminant analysis Effect Size (LEfSe) analyses were performed on the Huttenhower lab Galaxy server (http://huttenhower.sph.harvard.edu/galaxy/↗) by importing the relative abundance values and associated sample metadata, with FDR-adjusted q value < 0.05 considered significant and effect size calculated. We used observed community richness and Chao1 community richness indexes for comparison of bacterial OTU richness and Shannon diversity indexes for comparison of bacterial OTU diversity. Origin of microbiota analysis was performed using SourceTracker v. 0.9.8 with default parameters.³⁸ Donor and pre-FMT samples from same cage as source, and individual post-FMT samples as sink. Multiple group comparisons were made by the Kruskal Wallis test. P < .05 was considered statistically significant. All tests were performed with the R Project v3.6.0 for statistical computing or GraphPad Prism 7.0 (GraphPad, La Jolla, CA).