In this study, we performed a summary-data-based Mendelian randomization (SMR) analysis using cis-expression quantitative trait loci (cis-eQTLs) data from three human tissues—blood, transverse colon, and sigmoid colon—alongside genome-wide association study (GWAS) data for CDI. Subsequently, Bayesian co-localization analysis (coloc) was conducted using blood cis-eQTL data from an independent database and CDI GWAS data. All eQTL and GWAS datasets were derived from previously published studies or publicly available summary statistics provided by consortia. These studies had obtained approval from their respective Institutional Review Boards (IRBs); thus, no additional ethical review was required. Animal experiments employed a GF mouse model colonized with LA. Successful colonization was verified by crypt-depth histology and 16S rRNA profiling. RT-qPCR assessed expression changes in key genes identified by SMR and co-localisation analyses, and relevant phenotypic read-outs were recorded. Animal experiments were approved by the Ethics Committee of the Experimental Animal Center at Fudan University (Approval No. 2025-HSYY-156). A schematic overview of the workflow is shown in Figure 1 .

For the SMR analysis, cis-eQTL summary statistics for blood were obtained from the CAGE study (Lloyd-Jones et al., 2017), which investigated transcript-level gene expression in peripheral blood from 2,765 individuals of predominantly European ancestry. Cis-eQTL data for the transverse colon and sigmoid colon were sourced from the Genotype-Tissue Expression (GTEx) project (Consortium et al., 2020). The GTEx study analysed 15,201 RNA-sequencing samples derived from 49 tissues of 838 postmortem donors (primarily European Americans), including 368 transverse colon and 318 sigmoid colon tissue samples.Bayesian co-localization analysis utilized cis-eQTL data from the eQTLGen Consortium (Vosa et al., 2021), which aggregates 37 datasets comprising 31,684 blood samples.

The CDI GWAS data were obtained from the FinnGen study (Kurki et al., 2023), a large-scale genomics initiative that integrates genetic variation and health records from over 500,000 Finnish biobank participants to investigate disease mechanisms and predispositions. FinnGen is a collaborative effort between Finnish research institutions, biobanks, and international industry partners. Cases were defined as patients with Clostridioides difficile-induced enterocolitis, totalling 3,384 cases and 406,048 controls.

SMR is a summary-level Mendelian randomisation framework that uses expression quantitative trait loci (eQTLs) as instrumental variables (IVs) to test the causal impact of tissue-specific gene expression on complex traits or disease risk (Zhu et al., 2016). The primary SMR test employs the most significant eQTL (topSNP)—located within ±2000 base pairs (bp) of the target gene and achieving genome-wide significance (P < 5 × 10⁻⁸)—to infer a causal relationship between expression and phenotype. Like conventional MR, SMR rests on three assumptions that minimise population stratification and confounding (Relton and Davey Smith, 2012): (i) the SNP is strongly associated with the exposure; (ii) the SNP is independent of confounders; and (iii) the SNP affects the outcome solely through the exposure. Adhering to these assumptions reduces vulnerability to reverse causation and residual confounding relative to observational studies.

We utilized single nucleotide polymorphisms (SNPs) from cis-eQTL data of three human tissues—blood, transverse colon, and sigmoid colon—as instrumental variables (IVs), with gene expression as the exposure and CDI as the outcome. SMR analysis was employed to integrate GWAS and eQTL summary statistics to test for pleiotropic associations between gene expression and CDI, driven by shared and potentially causal genetic variants at specific loci. The heterogeneity in dependent instruments (HEIDI) test was applied to exclude associations likely caused by high linkage disequilibrium (LD) between distinct genetic variants, requiring the P_HEIDI > 0.05. Default SMR settings were adopted, including: P_eQTL  < 5 × 10 ^–8, minor allele frequency (MAF) > 0.01, exclusion of SNPs in strong LD (r ²  > 0.9) with the top-associated eQTL, removal of SNPs in weak or no LD (r ²  < 0.05) with the top-associated eQTL, and cis-eQTLs located within ±2000 bp of each probe were selected for the SMR analysis) (Zhu et al., 2016). The P_SMR were adjusted for false discovery rate (FDR) using the Benjamini-Hochberg method, and genes with q-value < 0.2 were selected as key candidates.

Functional enrichment analysis was performed on key genes identified across all three tissues. Enrichment analysis included Gene Ontology (GO) Biological Processes (BP), Cellular Components (CC), Molecular Functions (MF), and KEGG pathways. False discovery rate (FDR) correction was applied using the Benjamini-Hochberg method, with a significance threshold set at q-value < 0.05. The top 10 most significant GO terms and the top 20 most significant KEGG pathways were visualized to highlight key biological mechanisms.

Bayesian co-localization analysis was conducted using the coloc method (Wallace, 2021). For each key gene, blood cis-eQTL data from the eQTLGen Consortium were co-localized with CDI GWAS data. The coloc.abf() function was applied to all variants within a ±1 megabase (Mb) region around the top-associated SNP (topSNP) in the eQTL data and the corresponding region in the disease GWAS data. The posterior probability (PP.H4) of shared causal variants between gene expression and CDI was calculated, with PP.H4 > 50% considered robust evidence for co-localization (Huang et al., 2023).

GF and specific pathogen-free (SPF) male C57BL/6 mice (6 weeks old) were obtained from Slac Laboratory (Shanghai, China). GF mice were housed in sterile isolators, with weekly verification of their GF status via microscopy and aerobic/anaerobic culturing of fresh faecal samples. All GF mice received autoclaved water and a gamma-irradiated (50 kGy) standard diet. Mice were maintained at 20–24°C under a 12-hour light/dark cycle (lights on at 07:30) with 55–65% humidity. Two independent experiments were conducted with LA strains from different suppliers and distinct mouse batches. Experiment 1 used 13 mice: 5 GF mice colonised with LA (experimental), 4 GF mice gavaged with phosphate-buffered saline (PBS, negative control), and 4 SPF mice gavaged with PBS (positive control). Experiment 2 followed the identical design with 12 mice (4 per group).

Bacteria were cultured for 24 h at 37°C in MRS broth (Hopebio, Qingdao, China) under anaerobic conditions (5% H₂, 10% CO₂, 85% N₂). Bacterial cells were harvested by centrifugation at 3,000 × g for 15 minutes, washed, and resuspended in sterile PBS containing 0.1% peptone. GF mice received a daily oral gavage of 1 × 10⁹ colony-forming units (CFU) in 200 μL PBS for 14 consecutive days. On day 15, all mice were euthanized via cervical dislocation under isoflurane anesthesia. Colon tissues and luminal contents were immediately collected for downstream analyses.

Approximately 30 mg of freshly excised colon tissue was rinsed twice in PBS, freeze-dried, and homogenised in 600 µL of lysis buffer from the Cell/Tissue Total RNA Kit (NCM Biotech) using an electric homogeniser. After centrifugation, the supernatant was processed according to the manufacturer’s protocol to isolate total RNA, which was reverse-transcribed with a cDNA synthesis kit (Monad Biotech). Quantitative PCR was then carried out with the RT-PCR kit (Selleck) as per the kit instructions: each reaction contained 1 µL of forward primer and 1 µL of reverse primer (10 µmol/L each), 120 ng of template cDNA, and nuclease-free water to a final volume of 20 µL. The Thoc5 primers (Guria et al., 2011) were as follows: forward: 5′- TCTGCCTTTTCACCTGGAAG -3′, reverse: 5′- CTCGGTACTTTTCTGCCAGC -3′. The β-actin reference gene primers (Kong et al., 2023) were: forward: 5′- AGAAGATCTGGCACCACACC -3′, reverse: 5′- TACGACCAGAGGCATACAGG -3′. Thermal cycling conditions included an initial denaturation at 95°C for 30 sec, followed by 40 cycles of denaturation at 95°C for 15 sec, annealing/extension at 60°C for 30 sec, and a final extension at 72°C for 30 sec. Relative Thoc5 expression levels were normalized to β-actin and calculated using the 2-ΔΔCT method.

A 2 cm segment of colon was excised from each mouse and gently rinsed with PBS at low speed and low pressure. Tissues were then fixed in 4% (wt/vol) paraformaldehyde, and embedded in paraffin. Longitudinal sections (4 μm thickness) were stained with hematoxylin (Vector Laboratories, Burlingame, CA, USA) and eosin (Sigma-Aldrich, Zwijndrecht, Netherlands). Images were captured using a digital microscope (100×), and crypt depth was measured in 10 well-preserved crypts per mouse. For the immunofluorescence (IF) study, colon sections from the first experiment were analysed (n = 3 mice per group). Sections were incubated with a primary anti-F4/80 antibody (Cell Signaling Technology, 1:500) followed by an anti-rabbit IgG secondary antibody (Jackson, 1:500). Fluorescent labelling was developed with the TYR-520 fluorophore, TSA+ signal amplifier, DAPI nuclear counterstain, and an antifade mounting medium supplied in the IF kit (Huilanbio). Slides were scanned on a fluorescence microscope, and images were captured at ×400 total magnification. Three random fields per animal were photographed, and the mean fluorescence intensity of F4/80 was quantified in Fiji. Crypt-depth measurements and IF quantification were performed independently by two pathologists who were blinded to the group assignments.

Genomic DNA was extracted from mouse intestinal luminal content samples and assessed for quality via 1% agarose gel electrophoresis. Target regions were amplified using barcoded primers (TransStart FastPfu DNA Polymerase, ABI GeneAmp^® 9700 PCR System) under low-cycle PCR conditions, with triplicate reactions pooled and purified for each sample. Amplification products were quantified using the Quant-iT™ PicoGreen fluorescence assay (Promega, USA), and libraries were constructed with the TruSeq™ DNA Sample Prep Kit (Illumina, USA). Paired-end sequencing was performed on the NextSeq platform (Illumina, USA).

Bioinformatics analysis followed the QIIME2 pipeline: raw data underwent quality filtering, paired-end read merging, and denoising via DADA2 to generate amplicon sequence variants (ASVs). Taxonomic annotation was performed using the SILVA database. Alpha diversity indices (Shannon, Simpson, Chao1) and beta diversity metrics (Bray-Curtis, UniFrac distances) were calculated. Group differences were assessed using ANOSIM/PERMANOVA tests. Community structure was visualized via bar plots, heatmaps, and Venn diagrams. Differential taxa analysis was conducted using LEfSe and MaAsLin2 (adjusted for confounders). Functional profiling of microbial communities was inferred using PICRUSt2 and Tax4Fun, while microbial phenotypes were evaluated via BugBase.

Statistical analyses were performed using R (version 4.3.1) and GraphPad Prism (version 9.0). Functional enrichment analysis was conducted with the clusterProfiler R package (version 4.10.0) (Wu et al., 2021). Bayesian co-localization analysis was implemented using the coloc R package (version 5.2.3) (Wallace, 2021). SMR was performed with the SMR software tool (https://yanglab.westlake.edu.cn/software/smr/#Overview) (Zhu et al., 2016). RT-qPCR data, crypt-depth measurements, and IF intensities were analysed in GraphPad Prism. Variance homogeneity was checked by one-way ANOVA, followed by Welch’s t-test where appropriate. Processing of 16S rRNA sequencing data utilized the vegan (version 2.6-4) and phyloseq (version 1.46.0) R packages. Visualization of results was performed using the ggplot2 R package (version 3.4.4). The code used in the study can be found at https://github.com/syxdavid/Bayesian-co-localization-analysis.

We performed SMR analysis using gene expression quantitative trait loci (eQTL) data from three tissues—blood, transverse colon, and sigmoid colon—integrated with CDI GWAS data. Genes meeting the significance threshold (P_SMR < 0.05) and passing the heterogeneity (HEIDI) test (P_HEIDI > 0.05) numbered 365, 223, and 203 in blood, transverse colon, and sigmoid colon tissues, respectively ( Supplementary Table 1 ). We detected 14 key genes across the three tissues analysed ( Figure 2A ).

Sigmoid colon: XXbac-BPG181B23.7 (OR = 0.83, 95% CI 0.79–0.88; P_SMR = 5.36 × 10⁻¹²; q = 2.61 × 10⁻⁸), AGER (0.55, 0.41–0.73; 5.43 × 10⁻⁵; 3.31 × 10⁻²), PSMB9 (0.68, 0.56–0.83; 1.45 × 10⁻⁴; 6.42 × 10⁻²), TNNI3K (0.87, 0.81–0.94; 4.58 × 10⁻⁴; 1.49 × 10⁻¹), C4A (1.24, 1.10–1.40; 4.96 × 10⁻⁴; 1.51 × 10⁻¹), HLA-DQA1 (1.17, 1.07–1.29; 5.78 × 10⁻⁴; 1.54 × 10⁻¹), HLA-DQB1 (1.09, 1.04–1.15; 5.52 × 10⁻⁴; 1.54 × 10⁻¹), THOC5 (0.91, 0.86–0.96; 7.48 × 10⁻⁴; 1.82 × 10⁻¹).

Transverse colon: Y RNA (0.74, 0.66–0.83; 2.15 × 10⁻⁷; 2.92 × 10⁻⁴), RNF5 (1.62, 1.31–2.01; 1.10 × 10⁻⁵; 8.51 × 10⁻³), HLA-DRB9 (0.79, 0.71–0.88; 1.33 × 10⁻⁵; 8.61 × 10⁻³), AGER (0.55, 0.41–0.72; 2.29 × 10⁻⁵; 1.25 × 10⁻²), PSMB9 (0.64, 0.52–0.80; 5.84 × 10⁻⁵; 2.64 × 10⁻²), C4A (1.21, 1.09–1.34; 3.82 × 10⁻⁴; 1.27 × 10⁻¹), TNNI3K (0.81, 0.72–0.91; 3.97 × 10⁻⁴; 1.27 × 10⁻¹), HLA-DQB2 (0.91, 0.87–0.96; 4.58 × 10⁻⁴; 1.38 × 10⁻¹), VARS2 (1.15, 1.06–1.25; 5.54 × 10⁻⁴; 1.51 × 10⁻¹), HLA-DQB1 (1.12, 1.05–1.19; 5.55 × 10⁻⁴; 1.51 × 10⁻¹), THOC5 (0.85, 0.77–0.93; 7.91 × 10⁻⁴; 1.94 × 10⁻¹).

Blood: HLA-C (0.78, 0.70–0.87; 1.10 × 10⁻⁵; 3.12 × 10⁻²), and PSMB9 (0.90, 0.85–0.94; 3.23 × 10⁻⁵; 6.87 × 10⁻²).

Of these, AGER, TNNI3K, C4A, HLA-DQB1, and THOC5 were significant in two tissues, while PSMB9 reached significance in all three ( Figure 2B ). Notably, most of these genes clustered within the major histocompatibility complex (MHC) region on chromosome 6 (29–33 Mb) (D’Antonio et al., 2019; Douillard et al., 2021), with their top-associated SNPs localized to adjacent genomic positions.

After excluding duplicate and unannotated genes, 12 key genes were subjected to functional enrichment analysis. Gene Ontology (GO) analysis revealed 30 significantly enriched terms (q-value < 0.05), highlighting antigen binding, processing, and presentation, MHC class II molecule activity, and immune regulation ( Supplementary Figure 11A ). KEGG pathway analysis identified 20 enriched pathways (q-value < 0.05), including T-cell differentiation, inflammatory/autoimmune diseases, infectious diseases, and pathogen clearance ( Supplementary Figure 11B ).

Bayesian co-localization analysis of key gene eQTLs (from the eQTLGen Consortium) and CDI GWAS data revealed robust evidence for shared causal variants only at the THOC5 locus (posterior probability, PP.H4 = 82.6%; Figure 3B ). Results for other genes are provided in Supplementary Figures 1–10 . THOC5 exhibited significant SMR associations (P_SMR < 0.05) across all three tissues, with FDR-corrected q-values < 0.2 in both colon tissues ( Figure 3A ).

To assess whether THOC5 expression is functionally involved in the LA–mediated defence against CDI, we carried out two independent experiments. GF mice were gavaged with live LA strains obtained from two different sources to achieve intestinal colonisation. Throughout housing and handling, no animals exhibited signs of distress or abnormal behaviour: there was no kyphosis, lethargy, reluctance to move, or motor impairment. Fur remained smooth and well groomed, respiration was normal, and no ocular, nasal, or oral discharge was observed. In Experiment 1, GF mice were colonised with LA strain ATCC 4356. RT-qPCR analysis of colon tissues revealed a significant upregulation of Thoc5 expression in LA-colonized GF mice compared to controls (fold change [FC] = 5.19, P = 0.034; Figure 4A ). Similarly, Thoc5 expression differed markedly between GF and SPF mice (FC = 5.30, P = 0.047; Figure 4B ). Experiment 2 produced concordant results: colonic Thoc5 mRNA remained significantly up-regulated after LA colonisation (fold change = 6.79, P = 0.025; Figure 4C ), and levels in the LA-treated GF mice were still markedly higher than in SPF controls (fold change = 5.00, P = 0.031; Figure 4D ). To verify successful engraftment, all colonic samples were subjected to H&E staining and crypt-depth measurement. GF intestines typically display transient crypt hyperplasia within the first 16 days after exposure to an external microbiota, reflecting increased epithelial proliferation (El Aidy et al., 2012). In both experiments, crypts were significantly deeper following LA colonisation, confirming effective microbial establishment ( Figures 5C–L ). 16S rRNA sequencing of luminal contents from LA-colonized mice confirmed exclusive dominance of LA, ruling out contamination by other bacterial species ( Figures 5A, B ).

Most work on THOC5 has centred on its role in monocyte-to-macrophage differentiation and maturation (Carney et al., 2009; Tran et al., 2013; Tran et al., 2014). We therefore hypothesised that the LA–induced increase in colonic Thoc5 expression might drive macrophage differentiation in the gut. F4/80 is a mouse-specific marker of colonic macrophages; its abundance reflects both differentiation status and the extent of macrophage infiltration (Deng et al., 2024; Liu et al., 2025). Accordingly, we stained colonic sections for F4/80 by immunofluorescence and quantified the signal intensity. LA colonisation markedly increased F4/80 expression in the colon of GF mice (fold change = 1.17, P = 0.039; Figure 6 ), indicating enhanced macrophage differentiation and infiltration in response to the probiotic treatment.