Exploring the role of gut microbiota in inflammatory bowel disease patients comorbid with non-alcoholic fatty liver disease

The present study consisted of two parts. In the first part, information of patients with IBD, NAFLD, and COMO from Peking University Third Hospital and Peking University First Hospital was retrospectively collected, including general information (gender, age, height, weight, smoking and drinking history), medical history (hypertension, diabetes, and hyperlipidemia), and laboratory results (glucose, total cholesterol, total triglycerides and glycosylated hemoglobin) (Fig. S1a). In the second part of this study, another cohort of patients with IBD, NAFLD and COMO, as well as healthy controls from Peking University Third Hospital, were prospectively recruited to explore the gut microbiota features. The information mentioned above was also retrieved from these subjects. In addition, fecal samples were collected and analyzed via bacterial 16S rRNA gene sequencing and bioinformatic analysis (Fig. S1b). The overall study protocol was approved by the Ethics Committee of Peking University Third Hospital (Approval No. M2020447). Informed consent was obtained from all individuals involved in this study.

Individuals (> 18 years old) who fulfilled the diagnostic criteria for IBD, NAFLD or COMO, and healthy volunteers were recruited. IBD was diagnosed according to the 2018 consensus on the diagnosis and treatment of inflammatory bowel disease by the Chinese Society of Gastroenterology [8]. Activity of CD and UC was evaluated using the Crohn’s disease activity index [9] and modified Mayo score [10], respectively. NAFLD was diagnosed by abdominal computed tomography scans, magnetic resonance imaging, or ultrasound data collected up to 3 months before recruitment. Patients with alcoholic liver disease, hepatitis B and C virus infection, autoimmune hepatitis, hepatolenticular degeneration, drug-induced fatty liver diseases, total parenteral nutrition, celiac disease, hypothyroidism, Cushing’s syndrome, β-lipoprotein deficiency, lipoatrophy, diabetes, or other fatty liver diseases were excluded from the study. Individuals without gastrointestinal symptoms and with normal blood, urine, and fecal routine tests, as well as normal blood biochemistry and colonoscopy or abdominal imaging, were recruited as healthy controls. The study exclusion criteria were as follows: diagnosis of psychological disorders; current respiratory, digestive, urinary, or other system infectious diseases; severe heart, liver, lung, kidney, blood, endocrine, nervous system, autoimmune, or peripheral vascular diseases; organic intestinal disease other than IBD; liver disease other than NAFLD; and gestation or lactation. For subjects whose fecal samples were collected for gut microbiota analysis, those who were taking antibiotics, probiotics, or prebiotics within one month before recruitment were also excluded. Blood sample collection was performed according to hospital standard procedures in either inpatient wards or outpatient testing centers. Fecal sample was collected using a stool sample collection kit, including sample tubes, several ice packs, and an insulated bag.

Fecal samples were collected and stored at − 80 °C until analysis. Microbial DNA was extracted, and the V3–V4 region of 16 S rRNA was amplified and subsequently sequenced using the HiSeq 2500 platform (Illumina, San Diego, CA, USA). The Parallel-Meta Suite (PMS) was used to processing the amplicon sequencing data and to determine taxonomy [11]. PMS performs amplicon sequence variants (ASVs) denoising [12] and de-chimera [13] for marker genes to avoid sequencing inaccuracy. Then sequences are aligned against reference databases by the built-in vsearch [14] for profiling and taxonomy annotating from kingdom level to species level. The relative abundance of community members on each taxonomy level is also corrected using marker gene copy number normalization.

Gut microbiota analysis was performed using R software (v4.1.0; R Foundation for Statistical Computing, Vienna, Austria) after the respective operational taxonomic unit table was obtained. Alpha diversity was analyzed via the ‘vegan’ R package (v2.5-7). Principal coordinate analysis was conducted with the ‘vegan’ (v2.5-7) and ‘ape’ (v5.5) R packages. Functional prediction of 16 S rRNA was conducted with the Parallel-Meta Suite [11]. Chord diagrams and heatmaps were drawn with the ‘circlize’ (0.4.13) and ‘pheatmap’ (v1.0.12) R packages. Other diagrams were drawn with ‘ggplot2’ (v3.3.5). Cooccurrence network analysis was performed with ‘igraph’ (v1.2.11) and visualized via ‘gephi’ (v0.9.2). Machine learning models used were implemented by ‘randomForest’ (v4.7-1.2), ‘xgboost’ (v1.7.9.1) and ‘e1071’ (v1.7-16).

A retrospective cohort comprising 647 patients was analyzed, including 490 IBD (352 UC and 138 CD), 89 NAFLD, and 68 COMO patients (Table 1). Comparative demographic analysis revealed that COMO and IBD patients were younger and had lower body mass index compared to NAFLD patients. Metabolic profiling demonstrated that IBD patients had markedly reduced metabolic risk factors relative to NAFLD patients, including a lower prevalence of hypertension and diabetes, as well as decreased levels of blood glucose, glycosylated hemoglobin, total cholesterol (TC) and triglycerides (TG). Notably, COMO patients also displayed a reduced metabolic burden, with a lower hypertension incidence and lower levels of glucose, TC and TG compared to NAFLD subjects.

Next, the second cohort of 117 patients (30 IBD, 26 COMO, 32 NAFLD and 29 controls) was enrolled prospectively for further gut microbiome analysis (Table 2). Similar to the first cohort, IBD comorbid with NAFLD patients were also younger than NAFLD alone patients (COMO vs. NAFLD: 45.92 vs. 53.47 years, ^* p < 0.05). Although there was no difference in known history of diabetes, the fasting glucose of NAFLD patients at enrollment was higher than that of the COMO group and within the range of impaired glucose regulation (NAFLD vs. COMO: 6.32 vs. 5.30 mmol/L, ^* p < 0.05). With respect to inflammatory factors, IBD and COMO patients had increased level of fecal calprotectin (IBD vs. NAFLD: ^*** p < 0.001; COMO vs. NAFLD: ^* p < 0.05). Moreover, hemoglobin and albumin, although within the normal range, were lower in the IBD patients than in the NAFLD group. Concerning the IBD patients with or without NAFLD, no major difference was found in symptoms or disease activity (Table 3). In the IBD-only group, 50% (n = 15) of the patients were treated with biologics, which was higher than that in the COMO group (11.5%, n = 3, ^** p = 0.005).

To explore the role of gut microbiota in the development of IBD-NAFLD comorbidity, fecal samples were collected and analyzed via 16 S rRNA gene sequencing. The alpha diversity of the gut microbiota, as measured by the Shannon (Fig. 1a) and Simpson indices (Fig. 1b), was lower in COMO patients than in IBD or NAFLD patients. Principal coordinate analysis (PCoA) further revealed that beta diversity was significantly different among the four study groups (Fig. 1c, PERMANOVA: p = 0.001). The overall gut microbiota structure of COMO patients was similar to that of IBD patients (PERMANOVA: p = 0.600) but markedly different from that of NAFLD patients (PERMANOVA: p = 0.006). Network analysis revealed that the co-occurrence of the gut microbiota in COMO patients was not only much simpler but also more dispersed than that in any of the other groups (Fig. 1d, Table S1). Additionally, IBD, NAFLD, and COMO patients had fewer negatively correlated gut microbiota, especially COMO patients, which again suggested a weakened antagonistic relationship in this group (Table S1).

At the phylum level, comparative analysis revealed significant alterations in microbial composition across disease groups (Fig. 2a, b). Compared with the controls, all patient groups (IBD, COMO and NAFLD) exhibited decreased relative abundance of Firmicutes and Actinobacteria, while presenting increased proportions of Bacteroidetes, Proteobacteria, and Fusobacteria.

Genus-level analysis was performed via Kruskal-Wallis test which identified significant differences among all groups, followed by pairwise comparisons using Nemenyi test, DESeq2 and limma methods (Table S2). The number of differentially abundant taxa identified by each method was presented in Venn diagrams in Fig. 2c, with the name of the taxa commonly detected by all three methods displayed in Fig. 2d. Notably, the IBD and COMO groups demonstrated minimal microbial divergence, with no differentially abundant genera identified by all three methods (Fig. 2c), reinforcing the resemblance of the gut microbiota between COMO and IBD patients but not the NAFLD group.

For each disease state, the analysis revealed distinct microbial signatures, with IBD showing enrichment of Lachnospiraceae UCG 004, COMO demonstrating increased abundance of Bacteroides, Oscillibacter, and Phascolarctobacterium, and NAFLD exhibiting elevation of Candidatus Saccharimonas, Lachnospiraceae ND3007 group, Psychrobacter, and Rikenellaceae RC9 gut group (Fig. 2d, Fig. S2a-h). Depletion patterns were observed for Subdoligranulum in COMO and for Faecalibacterium and Solobacterium in NAFLD patients (Fig. 2d, Fig S2i-k). Shared microbial alterations across disease groups included decreased Agathobacter, Dorea, Eubacterium coprostanoligenes group, and Ruminococcaceae UCG 013 in IBD and COMO patients (Fig S2l-o), whereas reduced Ruminiclostridium 5 in NAFLD and COMO patients (Fig S2p). Notably, the increase in Lactococcus (Fig. 2e) and decrease in Coprococcus 3 and Ruminococcus 2 were consistent across all disease groups (Fig. 2d-g), suggesting their potential role as common contributors to the development of both IBD and NAFLD. LEfSe analysis further supported the importance of Ruminococcus 2 and Coprococcus 3 in discriminating controls from disease groups (LDA > 3.5, Fig. 2h). Furthermore, clinical correlation analysis revealed negative associations between Coprococcus 3 as well as Ruminococcus 2 and the inflammatory marker fecal calprotectin, whereas Lactococcus was positively associated with blood glucose level (Fig. 2i), indicating potential links between gut dysbiosis and disease manifestations.

To further investigate the functional changes of gut microbiota in IBD and NAFLD patients, functional prediction analysis was performed using PICRUSt2 algorithm with KEGG pathway annotation. Among the 277 KEGG categories recognized, 145 pathways exhibited significant differences in COMO patients versus controls, with 106 pathways overlapping with IBD group, and within these pathways, 88 were concurrently altered in NAFLD patients (Fig. 3a). Correlation analysis revealed 60 pathways associated with the differential genera Ruminococcus 2, Coprococcus 3 and Lactococcus, half of which were related to metabolism (Fig. 3b). Three KEGG level 2 pathways demonstrated significant differences, namely metabolism of cofactors and vitamins (Fig. 3c), biosynthesis of other secondary metabolites (Fig. 3d) and metabolism of other amino acids (Fig. 3e). Therein, 11 subordinate level 3 pathways also demonstrated notable differences (Fig. S3). Further correlation analysis between these 11 metabolic pathways and the clinical features shown in Table 2 revealed that BMI was correlated with four vitamin pathways, including thiamine metabolism (KO00730), riboflavin metabolism (KO00740), folate biosynthesis (KO00790) and biotin metabolism (KO00780), whereas blood glucose was associated with folate biosynthesis (KO00790), phosphonate and phosphinate metabolism (KO00440) and ubiquinone and other terpenoid quinone biosynthesis (KO00130) (Fig. 3f).

To evaluate whether these identified metabolic pathways and microbial signatures could serve as biomarkers for distinguishing COMO patients from IBD patients, we further employed a machine learning framework. Feature selection was performed using random forest importance scoring, with seven key predictors (mean decrease accuracy > 1.0) identified for subsequent model construction (Fig. 4a). The optimized random forest classifier achieved an area under the curve (AUC) of 0.864 through 10-fold cross-validation (Fig. 4b). Additionally, three alternative algorithms, including logistic regression, XGBoost and support vector machine, were applied and consistently demonstrated robust discriminative capacity and classification precision (AUC > 0.8, F1 score > 0.75) (Fig. 4b-c), validating the biomarker potential of the seven selected features (Ruminococcus 2, Lactococcus, and the metabolic pathway of isoflavonoid, thiamine, vitamin B6, biotin and caffeine).