Cross-trait mapping of shared susceptibility across inflammatory bowel disease and spondyloarthropathies

Abstract

Background

Inflammatory bowel diseases (IBDs) and spondyloarthropathies (SpAs) frequently co-occur, yet the subtype-specific genetic architecture and effector tissues that implement this gut–joint coupling remain poorly defined.

Methods

We integrated European-ancestry GWAS summary statistics for IBD, Crohn’s disease, ulcerative colitis, ankylosing spondylitis, psoriatic arthritis, reactive arthritis and enteropathic arthritis, covering 22.21 million variants across 1,263,767 individuals. We quantified genome-wide and local sharing (S-LDSC/LDSC, ρ-HESS), identified pleiotropic loci (MTAG, CPASSOC) in trait pairs with significant genome-wide genetic correlation, and prioritized shared loci by integrating local-correlation evidence with GWAS-PW and GCTA-COJO, mapped signals to tissues/cell types (LDSC-SEG, gsMap), prioritized effector genes (UTMOST, FUSION, SMR, immune-cell scPrediXcan), and assessed directionality using bidirectional Mendelian randomization.

Results

IBD showed 16.4% liability-scale SNP heritability (CD 21.5%; UC 15.2%), whereas SpA subtype heritability ranged from 2.4% (ReA) to 62.9% (AS). Genetic correlations were positive but heterogeneous, with the strongest sharing observed for EnA-related pairs. Local genetic analyses indicated structured but uneven regional sharing across subtype pairs, with the most robust local support concentrated in PsA- and EnA-related comparisons. Tissue and cell-type analyses converged mainly on immune- and barrier-related contexts, including blood, synovial fluid, spleen, lung, and lymphoid tissues, as well as T-lineage and myeloid immune compartments. Complementary gsMap projection suggested a developmentally informed gut–immune spatial scaffold rather than a uniform anatomical pattern. Pair-specific transcriptomic integration further prioritized recurrent candidate effectors across the IBD–SpA axis, while bidirectional Mendelian randomization supported a predominantly forward component from intestinal inflammatory liability to selected SpA phenotypes.

Conclusion

These results provide a multi-layer genetic and biological framework for gut–joint comorbidity, implicating shared immune and barrier niches and supporting a gut-first component in the IBD–SpA axis.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-026-08326-4.

Introduction

Inflammatory bowel diseases (IBDs), including Crohn’s disease (CD) and ulcerative colitis (UC), and spondyloarthropathies (SpAs), such as ankylosing spondylitis (AS), psoriatic arthritis (PsA), enteropathic arthritis (EnA) and reactive arthritis (ReA), are chronic immune-mediated disorders that predominantly affect the gut and the axial or peripheral skeleton. Clinically, these conditions rarely occur in true isolation. For decades, physicians have recognized a reproducible pattern of co-occurrence [1–4]. In a large cohort study, SpA was more prevalent at IBD diagnosis (OR 3.48) and incident SpA remained markedly elevated during follow-up (HR 7.15), with the strongest association observed for CD (HR 10.55) [5]. Epidemiological data also indicate that SpA occurs in up to 13% of IBD patients, including AS (~3%), sacroiliitis (~10%), and peripheral arthritis (~13%) [6]. Conversely, IBD is present in approximately 5% of patients with AS [7].

For affected individuals, concurrent IBD and SpA impose a compounded disease burden. The chronic abdominal pain, diarrhea, rectal bleeding and malabsorption characteristic of IBD intersect with the progressive spinal stiffness, peripheral joint destruction and loss of mobility that define SpAs [8]. This dual pathology creates complex therapeutic dilemmas: treatments effective for intestinal inflammation may incompletely control—or in some contexts exacerbate—musculoskeletal manifestations, while timing and intensity of immunomodulation must be balanced against infection risk and long-term toxicity. These challenges urge the need to understand why and how intestinal and skeletal inflammation are biologically coupled.

Genetic studies suggest that this coupling is not coincidental. Early work in the major histocompatibility complex (MHC) and HLA-B27 established overlapping susceptibility across IBD and SpA. More recent genome-wide association studies (GWAS) identified additional shared loci and reported polygenic correlations between gut and joint phenotypes, pointing to a partially shared genetic architecture [9, 10]. At the same time, clinical observations indicate substantial heterogeneity: not all IBD patients develop SpA; different SpA subtypes show different propensities for co-occurring intestinal disease; and UC and CD are not interchangeable in this context. These discrepancies raise unresolved questions: (i) how symmetrically do IBD subtypes and SpA subtypes share genetic risk at genome-wide and regional levels; (ii) through which tissues, cell types, genes and molecular pathways is this shared architecture realized; and (iii) is the gut–joint relationship dominated by shared predisposition or by directional causal effects?

Addressing these questions requires an approach that goes beyond single-disease GWAS or simple overlap of risk loci. Traditional analyses can detect shared variants but cannot decompose global versus local genetic correlations, resolve heterogeneity across subtypes, map polygenic risk onto anatomically and cellularly resolved substrates, or distinguish horizontal pleiotropy from directional causality. In this study, we integrated large-scale GWAS data for IBD and four SpA subtypes to construct a multi-layer map of the gut–joint axis. First, we quantified genome-wide and local genetic correlations and used cross-trait association analyses to identify comorbidity-specific loci. Second, we projected shared architecture onto tissues, cell types and genes by combining tissue/cell-type heritability enrichment, spatial transcriptomic mapping, and multi-tissue and single-tissue transcriptome-wide association studies, followed by gene-level causal inference and pathway enrichment. Third, we applied bidirectional Mendelian randomization to evaluate directional causal effects between IBDs and SpAs.

This study aims to define how polygenic risk is organized across the genome, body and immune system to generate distinct patterns of gut–joint comorbidity, and to nominate loci, tissues and molecular targets for future mechanistic and translational work. A schematic overview of the analytical framework is shown in Fig. 1.

Fig. 1.

Overall analytical workflow. Summary-level GWAS data for IBD (overall, Crohn’s disease, ulcerative colitis) and SpA phenotypes (ankylosing spondylitis, psoriatic arthritis, enteropathic arthritis, reactive arthritis) were first screened for SNP heritability and pairwise genetic correlation using S-LDSC/LDSC. Trait pairs with robust genetic sharing then underwent local genetic correlation and cross-trait meta-analysis to identify shared and pleiotropic risk SNPs, followed by GWAS-PW and GCTA-COJO to retain loci not driven by a single disease. Prioritised variants were mapped to tissues and cell types through single- and cross-tissue TWAS (FUSION, UTMOST, S-PrediXcan) and immune-cell prediction models, and further projected onto spatial transcriptomic data to define enrichment patterns. Finally, bi-directional MR and SMR were applied to these loci/genes to test causal directions between IBD and SpA

Methods

Study design and genomic data Integration

To investigate the shared genetic basis between inflammatory bowel diseases and spondyloarthritis, we conducted a multi-stage genomic analysis that integrated primary association signals from seven independent genome-wide association studies, collectively spanning 22.21 million high-quality genetic variants across 1,263,767 individuals of European ancestry. The analytical framework included IBD, CD, and UC, alongside the spectrum of spondyloarthritis manifestations including ankylosing spondylitis (AS), psoriatic arthritis (PsA), reactive arthritis (ReA), and enteropathic arthritis (EnA) [11, 12]. The IBD genomic landscape was constructed using summary statistics from the International IBD Genetics Consortium, providing association data for 9.73 million SNPs across 25,042 IBD cases and 34,915 controls. For individual IBD subtypes, CD associations were derived from 12,194 cases versus 28,072 controls (9.57 million SNPs), while UC included 12,366 cases and 33,609 controls (9.58 million SNPs). Spondyloarthritis genomic architecture was characterized through four complementary datasets: AS associations from 186 cases and 456,162 controls (11.83 million SNPs), PsA including 5,065 affected individuals and 21,286 controls (8.55 million SNPs), ReA associations from 3,779 cases versus 315,115 controls (21.32 million SNPs), and EnA including 861 cases and 315,115 controls (21.32 million SNPs).

To minimize confounding effects from population substructure, all genomic association datasets were restricted to European ancestry. Quality control procedures followed standard GWAS protocols including variant filtering for minor allele frequency ≥ 0.01, imputation quality ≥ 0.8, Hardy-Weinberg equilibrium p > 1 × 10^{− 6}, and removal of ambiguous or strand-inconsistent SNPs. Detailed characteristics of these GWAS data were provided in STable 1.

Single-trait heritability estimation using stratified LDSC

To quantify the genetic architecture of inflammatory disorders, we estimated single-trait SNP heritability using stratified linkage disequilibrium score regression (S-LDSC) with the baseline-LD model [13, 14]. S-LDSC represents an advancement over standard univariate LDSC by partitioning heritability across diverse functional genomic annotations while accounting for linkage disequilibrium structure. This stratified approach prevents heritability underestimation that can occur with univariate methods due to negative selection pressure, where functional variants in low-LD regions contribute disproportionately to trait heritability. SNP heritability estimates were converted from the observed scale to the liability scale to account for binary trait ascertainment bias, using established epidemiological prevalence parameters. Population prevalence rates were set at 0.54%, 0.29%, and 0.25% for IBD, UC, and CD respectively [15], and 0.13%, 0.1%, 0.1%, and 0.1% for AS, PsA, EnA, and ReA correspondingly, based on current epidemiological estimates [16–19].

Cross-trait genetic correlation analysis using LDSC

To quantify shared genetic architecture between IBDs and SpA subtypes, we employed linkage disequilibrium score regression (LDSC) for cross-trait genetic correlation estimation [20]. LDSC exploits the polygenic nature of complex traits by regressing chi-squared association statistics against LD scores, effectively separating true genetic covariance from confounding factors such as population stratification and cryptic relatedness through the underlying linkage disequilibrium structure. Genetic correlations were computed using HapMap3-filtered SNPs to ensure high-quality LD reference panels and correlation estimates. LDSC generates genetic correlation coefficients ranging from −1 to +1, where extreme values indicate complete genetic opposition or concordance respectively, while intermediate values reflect partial genetic overlap. The method incorporates an intercept term to adjust for sample overlap between studies, ensuring unbiased correlation estimates. Statistical significance was assessed using Bonferroni correction for multiple comparisons (p < 0.0041; 0.05/12 tests).

Local genetic correlation mapping using ρ-HESS

To investigate genome-wide genetic correlations into regional components and identify specific genomic loci contributing to shared genetic architecture, we employed Heritability Estimation from Summary Statistics (ρ-HESS) for local genetic covariance analysis [21, 22]. ρ-HESS partitions the genome into approximately linkage disequilibrium-independent segments, enabling estimation of both local SNP heritability and cross-trait genetic correlations within discrete genomic regions. The analysis included 1,699 predefined genomic regions (mean size ≈1.5 Mb) selected for their approximate LD independence based on the 1000 Genomes Project reference panel. For each trait pair, we first obtained region-level estimates across the genome and used p < 0.05 to identify suggestive local-sharing regions for exploratory characterization of regional heterogeneity. We additionally reported FDR-corrected ρ-HESS.

Multi-trait GWAS meta-analysis using MTAG

To enhance statistical power for detecting shared genetic variants by leveraging cross-trait correlations, we performed multi-trait analysis of GWAS (MTAG) for joint phenotypes comprising IBDs and each SpA subtype [23]. The method assumes uniform SNP heritability across traits and accounts for sample overlap between studies through its variance-covariance structure. Prior to analysis, we validated these assumptions using the maximum false discovery rate (maxFDR) threshold to ensure homogeneous effect patterns across examined traits. Only trait pairs demonstrating significant genetic correlations through LDSC were selected for MTAG analysis to maximize the statistical power gains from multi-trait integration.

Cross-phenotype association testing using CPASSOC

As a complementary approach to identify pleiotropic variants affecting multiple traits, we employed the cross-phenotype association test (CPASSOC) with the SHet algorithm implementation [24]. CPASSOC synthesizes association signals from GWAS summary statistics when genetic variants demonstrate correlation with at least one phenotype, accommodating potential effect heterogeneity across traits through its flexible statistical framework. Genetic variants underwent pairwise linkage disequilibrium pruning (r² threshold = 0.2) using PLINK software to ensure independence. We prioritized variants meeting dual significance criteria: (1) genome-wide significance (p < 5 × 10^{− 8}) in both MTAG and CPASSOC analyses, and (2) physical localization within genomic regions identified as statistically significant by ρ-HESS local genetic correlation analysis.

Joint association analysis using GWAS-PW

To ensure that IBD-associated variants identified through cross-trait meta-analysis represented genuine pleiotropic effects rather than single-trait-driven associations, we implemented pairwise genome-wide association analysis (GWAS-PW) to detect genomic regions with shared genetic architecture between IBDs and SpAs [25]. GWAS-PW employs Bayesian statistical modeling to calculate posterior probabilities for distinct colocalization scenarios, specifically: (1) the posterior probability of a locus being causally shared by both traits (PPA3), and (2) the posterior probability of a locus being associated with both traits through distinct causal variants (PPA4). We established a stringent significance threshold of PPA3/PPA4 > 0.8 to prioritize variants with high confidence for genuine cross-trait associations [26].

Conditional independence testing and functional annotation

To identify conditionally independent association signals and resolve complex linkage disequilibrium structures, we employed GCTA-COJO (v.1.94.1) [27] with stepwise model selection procedures using default parameters (p < 5 × 10^{− 8}, 10 Mb window size, collinearity threshold < 0.9). This approach enabled detection of multiple independent causal variants within associated genomic regions while controlling for LD confounding. Independent SNPs underwent functional annotation using ANNOVAR [28], with variants classified as “previously reported” if catalogued in GWAS databases via FUMA mapping [29, 30], or designated as “novel” discoveries. Due to complex LD patterns characteristic of the major histocompatibility complex, MHC region variants were excluded from final annotation results to ensure reliable functional interpretation.

Colocalization analysis

We used the R package “coloc” [31] to assess locus-level colocalization between IBD and SpA association signals. For each prioritized cross-trait risk SNP, we extracted variants within ±500 kb of the index SNP and calculated the posterior probability that both traits were driven by a single shared causal variant. Genomic loci with a posterior probability H4 (PPH4) exceeding 0.7 were considered to show evidence of colocalization.

LDSC-SEG heritability enrichment analysis

To identify disease-relevant tissues and cell, we performed linkage disequilibrium score regression applied to specifically expressed genes (LDSC-SEG) [32]. This method employs stratified LD score regression to test whether genomic regions proximal to genes with tissue-specific expression patterns demonstrate significant enrichment for disease heritability. The analysis utilized precomputed LD score annotations including 53 key tissue types (with an average of 161 samples per tissue) from the GTEx project [32, 33] and 152 tissue and cell types derived from 37,427 human microarray samples in the Franke laboratory dataset [32, 34], providing comprehensive coverage of human tissue diversity. Due to complex linkage disequilibrium patterns that could confound enrichment estimates, the major histocompatibility complex (HLA) region was excluded from all LDSC-SEG analyses. Tissue-specific heritability enrichment was assessed using the baseline model framework to control for genomic features and ensure robust statistical inference of disease-relevant biological contexts.

Cross-tissue transcriptome-wide association analysis

Recognizing that single-tissue analyses may overlook systemically acting regulatory factors influencing IBD-SpA comorbidity patterns, we implemented cross-tissue transcriptome-wide association analysis using UTMOST [35] to evaluate gene-trait associations at the organismal level. This analytical framework addresses the limitation of tissue-specific approaches by capturing pleiotropic gene effects that operate across multiple biological contexts. UTMOST employs penalized multivariate regression to train cross-tissue expression imputation models, effectively integrating heterogeneous eQTL effect sizes and directions across diverse tissue types [36]. The method leverages tissue-specific heritability patterns to enhance both the detection of context-specific gene associations and the accuracy of genetically predicted expression levels. This approach captures shared regulatory architecture while preserving tissue-specific expression patterns, enabling comprehensive assessment of gene-trait relationships across the human transcriptome. To synthesize association evidence across multiple tissues and account for inter-tissue correlation structures, we implemented a generalized Berk-Jones test incorporating covariance matrices derived from single-tissue summary statistics [35]. This statistical framework appropriately weights tissue-specific evidence while controlling for multiple testing across both genes and tissues.

FUSION-based tissue-level transcriptome-wide association studies

Transcriptome-wide association study (TWAS) provides a robust analytical framework for post-GWAS investigation by modeling transcriptional regulation mechanisms to detect statistically significant gene-trait relationships [37]. This approach integrates expression quantitative trait loci (eQTL) information to identify functional genes modulated by disease-associated genetic variants. We systematically evaluated associations between individual genes and IBD/SpA traits across tissue-specific contexts using FUSION software [38] integrated with eQTL reference data from 49 tissues in 838 people (85.3% European ancestry) [36, 39]. FUSION employs ensemble machine learning approaches, implementing multiple predictive models (including elastic net, LASSO, and Bayesian sparse linear mixed models) to assess aggregate SNP effects on gene expression. The optimal prediction model for each gene-tissue pair was selected based on cross-validation performance metrics to assign appropriate gene expression weights for downstream association testing.

To distinguish genuine associations from linkage disequilibrium-driven signals, we implemented conditional and joint (COJO) analysis within each genomic locus. This approach identifies conditionally independent gene-trait associations by modeling the joint effects of multiple genes while accounting for local LD structure. Genes demonstrating independent associations through COJO analysis were classified as jointly significant and retained for biological interpretation, while those exhibiting only marginal significance in the presence of conditionally independent signals were excluded from subsequent functional analyses.

Single-cell spatial transcriptomic analysis using gsMap

To provide a complementary spatial view of disease-prioritized signals, we integrated single-cell spatial transcriptomic data with GWAS summary statistics using the genetically informed spatial mapping of cells for complex traits (gsMap) algorithm [40]. The analysis utilized a high-resolution embryonic mouse atlas (E16.5) and was intended to explore whether prioritized genes showed coherent spatial organization in a developmentally informed framework. Because this atlas does not directly represent adult human inflammatory tissues, gsMap was used as a supportive, hypothesis-generating spatial contextualization layer rather than as a stand-alone source of adult disease validation. For each of the seven analyzed traits (IBD subtypes and SpA conditions), gsMap evaluates the spatial enrichment of trait-associated gene expression signatures across anatomically defined regions. The method calculates enrichment scores by comparing observed versus expected expression levels of GWAS-prioritized genes within specific spatial domains, accounting for background expression patterns and tissue-specific transcriptional programs.

Multi-method Mendelian randomization (MR) for causal inference

To investigate causal relationships between IBDs and SpAs, five complementary MR methods with distinct underlying assumptions were utilized: Generalized Summary-data-based Mendelian Randomization (GSMR) [41], MR-Egger [42], inverse variance weighting (IVW) [43], weighted median [44], and weighted mode [45]. These methods operate under distinct assumptions regarding the extent and nature of horizontal pleiotropy. Briefly, GSMR assumes the absence of correlated pleiotropy but identifies and excludes SNPs demonstrating significant evidence of uncorrelated pleiotropy through the HEIDI-outlier detection method [41]. The MR-Egger approach further accommodates directional (i.e., non-zero mean) uncorrelated pleiotropy by incorporating an intercept term in the IVW regression to mitigate confounding bias arising from such pleiotropic effects [42]. The IVW method presupposes that any existing uncorrelated pleiotropy has a mean effect of zero, implying that such pleiotropy would only introduce multiplicative random-effect noise into the combined meta-analytic regression model of SNP effects [43]. The weighted median method yields reliable causal effect estimates even under conditions where nearly half of the genetic instruments are invalid [44]. The weighted mode method classifies SNPs into groups based on their estimated causal effects and evaluates causal evidence using only the largest cluster of SNPs. This approach essentially relaxes the assumptions of MR (Mendelian Randomization) and can identify the true causal effect even when the majority of instrumental variables are invalid SNPs [45]. Therefore, the application of weighted median and weighted mode methods can partially mitigate correlated pleiotropy.

For the selection of instrumental variables, a genome-wide significance threshold of p = 5 × 10⁻⁸ was applied across all five methods. However, due to the limited number of eligible SNPs available as instruments, a more lenient threshold of p = 1 × 10⁻⁶ was adopted when inferring causal relationships of AS/EnA on IBD/CD. Independent single nucleotide polymorphisms (SNPs) were identified through linkage disequilibrium clumping using PLINK v1.9 [46], with reference to the European population sample from the 1000 Genomes Project (clumping parameters: r² < 0.05, window size = 1000 kb). Potential confounders were further screened via PhenoScanner to exclude SNPs associated with outcome variables or their known risk factors.

The inverse variance weighted (IVW) method served as the primary analytical approach, while the other four methods were employed as sensitivity analyses. Causal relationships demonstrating consistent evidence across all MR methods were deemed more reliable and noteworthy. Instrument strength was assessed using per-variant F-statistics, and all retained instruments exceeded the conventional threshold of F > 10. In addition to Cochran’s Q and the MR-Egger intercept, we performed MR-PRESSO global tests, leave-one-out analyses, and Steiger directionality testing to evaluate potential pleiotropy, outlier influence, and the plausibility of causal direction. GSMR analysis was implemented through the GCTA software, which offers enhanced computational efficiency and greater flexibility compared to the R package GSMR [41]. The other four MR analyses were implemented using the R package TwoSampleMR [47].

Summary-data-based Mendelian randomization (SMR) analysis

To validate potential causal relationships between gene expression and IBD/SpA traits while mitigating confounding from linkage disequilibrium, SMR analysis was implemented [48]. This method integrates GWAS summary statistics with eQTL data to assess whether trait-associated genetic variants influence disease risk through modulation of gene expression levels. SMR analysis was coupled with the heterogeneity in dependent instruments (HEIDI) test to distinguish genuine causal relationships from spurious associations arising from linkage disequilibrium between trait-associated variants and eQTL signals. The HEIDI test evaluates whether multiple genetic instruments within a locus demonstrate consistent causal effects, with significant heterogeneity indicating potential linkage confounding rather than true causality.

We conducted SMR analysis across 49 tissues from a cohort of 838 individuals, with the majority being of European descent to capture tissue-specific causal relationships between gene expression and disease traits. Significant causal associations were identified using stringent statistical criteria: FDR-adjusted SMR P_FDR < 0.05 combined with HEIDI test p-value > 0.01, ensuring both statistical significance and absence of heterogeneity indicative of linkage confounding. We complemented transcriptomic prioritization with drug–gene interaction analysis using the Drug–Gene Interaction Database 5.0 DGIdb 5.0 to assess translational relevance.

Pathway enrichment analysis

To elucidate the biological mechanisms underlying identified gene-trait associations, functional enrichment analysis was performed using Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway database. Analysis was implemented using the clusterProfiler R package [49], with false discovery rate (FDR) correction applied to control multiple testing burden across pathway categories (P_FDR < 0.05). To further assess the robustness of the pathway-level findings, we repeated the enrichment analysis using the Reactome database.

Cell-level transcriptome-wide association studies

Although classical TWAS approaches have established disease-gene associations, their capacity to elucidate pathogenic mechanisms at cellular resolution remains constrained. To identify specific immune cell populations where disease genes exert functional effects, we employed scPrediXcan—a computational method that integrates single-cell transcriptomic data with GWAS summary statistics to detect cell type-specific gene-trait associations [50]. The analysis utilized single-cell RNA-sequencing data including 29 distinct immune cell types derived from 982 individuals of European ancestry. The dataset underwent quality control including Demuxlet-based [51] sample demultiplexing and Scrublet-mediated doublet removal to ensure high-quality single-cell profiles [52]. Cell type annotations were validated through canonical marker gene expression patterns and transcriptomic clustering approaches. scPrediXcan employs the Enformer deep learning model to predict epigenomic features surrounding each gene’s transcription start site (TSS), capturing regulatory elements that influence cell type-specific expression patterns. These epigenomic features serve as inputs to ctPred, a four-layer fully connected neural network that estimates gene expression percentiles within each immune cell type. The ctPred model was subsequently linearized into ℓ-ctPred to enable integration with GWAS summary statistics [53].

Gene-trait association testing was performed using the Summary-PrediXcan (S-PrediXcan) framework to compute standardized effect sizes (Z-scores) and evaluate statistical significance across cell types [54]. Results from all 29 immune cell types were integrated using the Aggregated Cauchy Association Test (ACAT) p-value combination method to identify genes demonstrating significant associations with IBD and SpA traits across the immune system landscape [55].

Results

Genome-wide and local shared genetic architecture between IBDs and SpAs

The extent and symmetry of shared genetic architecture between IBDs and SpAs were first quantified on the genome-wide and regional levels. Liability-scale SNP-based heritability estimated by S-LDSC with the baseline-LD model confirmed substantial polygenic components for all traits. IBD showed 16.4% heritability, with CD and UC exhibiting 21.5% and 15.2%, respectively. Among SpA subtypes, AS demonstrated the highest heritability at 62.9%, followed by EnA (18.6%), PsA (9.9%), and ReA (2.4%) (STable 2). Bivariate LDSC revealed broadly positive genetic correlations between IBDs and SpAs but with marked heterogeneity across subtype pairs (Fig. 2a, STable 3). EnA showed the strongest genome-wide sharing with IBD, particularly with CD (IBD–EnA: rg = 0.6081, p = 8.12 × 10^{− 9}; CD–EnA: rg = 0.6566, p = 1.11 × 10^{− 8}; UC–EnA: rg = 0.4107, p = 1.97 × 10^{− 5}). Correlations involving ReA were generally weaker, whereas those between UC and PsA were moderate. Across trait pairs, genetic correlations between IBD and individual SpA subtypes consistently fell between the corresponding CD–SpA and UC–SpA estimates. Quality-control assessment of LDSC intercepts supported the integrity of these correlations: sample overlap between PsA and IBDs was modest (intercepts 0.12–0.16), whereas intercepts for AS, EnA, and ReA with IBDs remained below 0.015. Trait pairs meeting corrected significance (p < 0.0041 for 12 comparisons) were carried forward to downstream cross-trait analyses.

Fig. 2.

Genome-wide and local genetic correlations between IBD traits and SpA phenotypes. (a) Heatmap of bivariate LDSC genetic correlations (r₍g₎) between IBD, Crohn’s disease (CD), ulcerative colitis (UC) and four spondyloarthropathy entities—ankylosing spondylitis (AS), psoriatic arthritis (PsA), enteropathic arthritis (EnA) and reactive arthritis (ReA). * p < 0.05. Strongest genome-wide correlations were observed for IBD/CD with EnA and AS. (b–j) boxplots summarizing significant local genetic-correlation patterns estimated by ρ-HESS across disease-specific locus categories (disease-specific, subtype-specific, intersection, and neither) for the LDSC-significant IBD–SpA trait pairs. FDR-corrected ρ-HESS results are provided in supplementary table 4 and distinguish more robust local-sharing signals from suggestive regional patterns. (k–m) additional regional correlation plots for remaining IBD/IBD-subtype–SpA combinations, displayed in the same format, highlighting that the global IBD–SpA overlap is assembled from a limited number of LD-independent hotspots rather than uniformly shared across the genome

To move beyond genome-wide averages, local genetic correlations were quantified using ρ-HESS to map region-specific sharing across the genome. Using a nominal threshold (p < 0.05), we identified 162, 31, and 37 unique suggestive regions across IBD–SpA, CD–SpA, and UC–SpA comparisons, respectively (STable 4). After FDR correction, however, these numbers were reduced to 10, 2, and 4 unique regions. FDR-supported local sharing was concentrated in IBD–PsA, IBD–EnA, IBD–AS, CD–EnA, UC–PsA, and UC–EnA comparisons, whereas no local regions for IBD–ReA, CD–AS, or UC–ReA remained significant after multiple-testing correction. Stratification by disease-specific risk loci still suggested subtype-dependent regional heterogeneity at the nominal level (Fig. 2b–j, SFigure 1-2), with EnA- and PsA-related comparisons showing the most consistent local sharing patterns. However, because most locus-category contrasts did not survive multiple-testing correction, these regional configurations should be regarded as exploratory trends.

Cross-trait meta-analysis identifies shared and comorbidity-specific risk loci

To identify specific loci contributing to the shared architecture, cross-trait association analysis was performed. Across all examined pairs, MTAG and CPASSOC produced highly concordant results (STable 5–7). Candidate variants were restricted to genomic regions exhibiting significant local genetic correlations in ρ-HESS (p < 0.05). These variants were then evaluated using GWAS-PW to distinguish associations shared between traits from those driven by either IBD or SpA alone. GCTA-COJO conditional analysis was subsequently applied to account for linkage disequilibrium and to derive independent SNP sets for IBD and its subtypes within each SpA context (STable 8). This multi-step framework revealed additional potential risk SNPs for IBDs that were not apparent in single-trait analyses (STable 9–11).

Functional annotation of independent risk SNPs identified five variants not previously catalogued in the NHGRI-EBI GWAS Catalog. Three variants (rs2891409, rs34426521, and rs62324212) were specific to IBD in the presence of PsA co-occurrence. One variant (rs3761158) emerged for CD under AS influence, and rs10408351 was identified for UC in association with EnA (STable 12). At the locus level, we further performed cross-trait colocalization for these prioritized risk variants. This analysis provided robust evidence of shared local causal-variant structure for 21 variants, including the novel loci rs34426521 in IBD–PsA and rs10408351 in UC–EnA (STable 12). These findings indicate that part of the gut–joint genetic architecture reflects genuinely shared cross-trait loci.

Tissue and cell-type landscape of the gut–joint genetic axis

The anatomical contexts in which IBD risk alleles act under different SpA influences were next examined. LDSC-SEG was used to evaluate tissue- and cell-type–specific enrichments, with statistical significance defined as adjust p < 0.05. Across all IBD–SpA trait pairs, significant heritability enrichment was observed in blood, synovial fluid, spleen, lung and lymphoid tissues (Fig. 3; SFigure 4; STable 13). At the cellular level, T-lymphocyte-related annotations showed the strongest and most consistent enrichment, while monocyte/macrophage-related compartments were also recurrently enriched across multiple IBD–SpA trait pairs. Blood and synovial fluid emerged as the most strongly enriched contexts across analyses. Synovial fluid demonstrated robust enrichment for all IBD subtypes, with UC consistently displaying higher enrichment rankings than IBD or CD under any SpA influence. Blood retained top enrichment rankings across all IBD subtypes and SpA contexts, underscoring the central role of circulating immune compartments in mediating gut–joint communication.

Fig. 3.

Tissue and immune-cell contexts enriched for shared IBD–SpA heritability. LDSC-SEG heritability enrichment for IBD-based SpA pairs across GTEx tissues and the Franke immune/tissue panel, showing consistent signals in blood, spleen, lung, lymphoid tissues and synovial fluid. Enrichment profile for CD-based SpA pairs, emphasising T lymphocytes, monocytes/macrophages, regulatory T cells and NK cells as repeatedly prioritised immune populations. Enrichment profile for UC-based SpA pairs, again highlighting blood and synovial fluid, with UC showing relatively higher ranks under several SpA influences. All enrichments are FDR-adjusted; HLA was excluded from LDSC-SEG to avoid LD-driven artefacts. Together, these panels indicate that shared IBD–SpA genetics is preferentially realised in immune- and barrier-associated environments

To provide complementary spatial contextualization for the tissue and cell-type signals identified above, we applied gsMap to the original single-trait GWAS summary statistics for the nine LDSC-significant trait pairs. gsMap projection onto an embryonic day 16 mouse atlas allowed prioritized signals to be visualized in a developmentally informed spatial framework (STable 14–15). Across IBD subtypes, these projections were concentrated in gastrointestinal and immune-related elements, which were consistent with the immune- and barrier-associated contexts highlighted by LDSC-SEG (Fig. 4b–d). Among SpA subtypes, the projected patterns were more heterogeneous, with AS showing relatively stronger musculoskeletal-interface enrichment, PsA showing stronger peripheral epithelial/hepatic enrichment, and EnA retaining a relatively strong intestinal component. ReA showed a broader and less focused distribution (Fig. 4e–h). These results provide a supportive spatial scaffold suggesting that the shared IBD–SpA genetic signal may be organized along a gut–immune core with subtype-dependent peripheral deployment.

Fig. 4.

Complementary spatial projection of shared IBD–SpA signals using gsMap. (a) Reference spatial map generated from the embryonic mouse atlas used in gsMap, shown without disease weighting, to visualize the developmental compartments used for downstream projections. This panel provides a developmental spatial scaffold for interpreting projected enrichments. (b–d) projection of IBD-, CD-, and UC-weighted gene sets onto the embryonic atlas, showing predominant enrichment in gut-, immune-, and adjacent mesoderm-related territories. (e–h) corresponding projections for AS, PsA, EnA, and ReA, illustrating heterogeneous spatial distributions across SpA subtypes. These patterns are presented as developmentally informed spatial contextualization

Transcriptome-wide association identifies convergent and subtype-specific genes

Gene-level mediators of shared risk were investigated using a combination of cross-tissue and single-tissue transcriptome-wide association studies. Cross-tissue TWAS identified 183 genes with regulatory effects shared across tissues, with the highest count observed for CD (119 genes), followed by IBD (115) and UC (54) (STable 16). Complementary single-tissue TWAS across 49 tissues identified 6,591 genes achieving adjust p < 0.05 in at least one tissue, including 4,388 genes for IBD (STable 17), 3,921 for CD (STable 18) and 2,761 for UC (STable 19). Integration of cross-tissue and single-tissue results yielded 95 high-confidence candidate genes for further investigation: 60 for IBD (Fig. 5), 61 for CD and 31 for UC (SFigure 5; STable 20). Among these, five genes (IL1RL1, PNKD, PEX13, ARPC2 and CXCR2) were consistently significant across all IBD–SpA trait combinations examined, defining a core gene set recurrently linking intestinal and joint phenotypes.

Fig. 5.

Cross-trait TWAS highlights shared expression-linked genes for IBD–SpA pairs. (a–d) TWAS heatmaps for IBD–AS, IBD–PsA, IBD–EnA and IBD–ReA across 49 GTEx tissues, showing that a recurrent set of genes (including ATG16L1, IL1RL1, ADCY3, PTPN2, ZFP36L2, REL, CXCR1/2) is predicted to be associated with both intestinal and musculoskeletal inflammation. Tissues are grouped by organ system to illustrate that many signals arise in digestive, immune/hematopoietic and barrier-related tissues, but several genes also show expression associations in blood, fibroblasts and vascular tissues

Beyond this core, several genes exhibited combination-specific associations. PAX8 and GPBAR1 were uniquely associated with IBD–EnA pairs, while MBD5 emerged specifically in IBD–ReA combinations. For CD–AS pairs, four genes (SEH1L, ASCC2, RTP5 and AC005104.3) showed exclusive positive associations. CD–EnA combinations revealed five specific associations (GTPBP1, ATP8A1, ERCC3, PITX2 and SMARCE1P6), whereas RP5-1039K5.16 and ING2-DT were uniquely associated with CD–PsA and CD–EnA, respectively. These patterns highlight a layered molecular structure in which a small set of shared genes coexists with subtype-specific regulatory programs tailored to distinct IBD–SpA combinations.

SMR-supported genes and pathway enrichment

To refine the transcriptomic candidates, we performed SMR analysis and summarized the pair-specific overlap between significant TWAS and SMR signals derived from the corresponding MTAG-weighted cross-trait analyses (STables 17–21). We identified 33 distinct genes overall, with overlap sets ranging from 8 to 22 genes per pair (STable 22). Specifically, the overlap sets comprised 19 genes for IBD–AS, 21 for IBD–EnA, 18 for IBD–PsA, 20 for IBD–ReA, 17 for CD–AS, 22 for CD–EnA, 12 for UC–EnA, 9 for UC–PsA, and 8 for UC–ReA. Across pairs, IL1RL1 was retained in all nine overlap sets; LINC01882 and CXCR2 in eight; PIM3, AC016747.3, SATB2, and AHSA2P in seven; and ATG16L1, CENPO, BLTP1, DNAJC27–AS1, ADCY3, and ZFP36L2 in six. Functional enrichment analysis of these genes revealed both conserved and pair-specific biological themes. Across multiple IBD–SpA combinations, the dominant shared signal was an inflammatory cytokine axis, including cytokine receptor activity, cytokine binding, immune receptor activity, cytokine–cytokine receptor interaction, and recurrent chemokine signaling (Fig. 6; SFigure 6-7). This core inflammatory program was independently supported by Reactome, which highlighted interleukin-33 signaling across all nine IBD–SpA pairs and additionally identified interleukin-1 family signaling and interleukin-37 signaling in CD–AS (STable 23).

Fig. 6.

Pathway enrichment of shared IBD–SpA genes. GO (BP/CC/MF) and KEGG analyses of genes jointly prioritised in IBD–AS, IBD–PsA, IBD–EnA and IBD–ReA pairs show recurring enrichment for dendritic-cell migration/chemotaxis, regulation of haematopoiesis, chromosome/membrane-associated complexes, and chemokine/cytokine receptor activities, together with KEGG terms such as cytokine–cytokine receptor interaction, chemokine signaling and bacterial/epithelial defence pathways

Translational annotation of prioritized genes

To further assess the translational relevance of the prioritized genes, we performed drug–gene interaction analysis. Among the TWAS–SMR-supported genes, 10 genes showed reported interactions with approved drugs, yielding 50 approved drug–gene pairs in total (STable 24). These included CXCR2 (7 approved interactions), ADCY3 (rimonabant, dronabinol, and cannabidiol), ACVR2A (sotatercept and dibotermin alfa), PIM3 (gefitinib), ATG16L1 (adalimumab), ZFP36L2 (acesulfame and saccharin), SF3B1 (olaparib and etoposide), NDUFS1 (metformin hydrochloride), and RPL9 (ataluren). Although these interactions do not constitute independent validation of gene causality, these interactions provide supportive translational annotation for a subset of prioritized genes.

Immune cell–specific transcriptomic associations refine effector genes

To further refine the cellular contexts in which candidate genes may act, we performed cell-level transcriptome-wide association analysis using expression models from 29 immune cell types. This analysis identified 2,058 unique genes with significant transcriptional associations with IBD or its subtypes under SpA comorbidity conditions in at least one immune cell type (STable 25). Integration of the tissue-level and cell-level transcriptomic results further distilled these signals into 20 genes (STable 26). The number of such tissue–cell overlap genes varied across trait pairs, comprising 13 genes for IBD–AS, 11 for IBD–PsA, 11 for IBD–EnA, 13 for IBD–ReA, 10 for CD–AS, 13 for CD–EnA, 4 for UC–PsA, 6 for UC–EnA, and 4 for UC–ReA. Across pairs, CXCR2 was retained in 8, IL1RL1 in 7, and ADCY3, ATG16L1, CENPO, SATB2, and ZFP36L2 each in 6. These genes were most frequently localized to monocyte/macrophage-related compartments, peripheral blood mononuclear cells, mucosal invariant T cells, natural killer cells, dendritic cells, and regulatory or effector-memory T-cell populations.

Directional causal relationships between IBDs and SpAs

To distinguish directional effects from shared genetic architecture, we performed bidirectional MR. At the multiple-testing-corrected threshold (p < 2.7 × 10^{− 3} for 18 tests), IVW dentified five significant forward associations: IBD → AS (OR = 1.44, p = 2.33 × 10^{− 4}), IBD → PsA (OR = 1.14, p = 2.03 × 10^{− 3}), CD → AS (OR = 1.36, p = 1.49 × 10^{− 3}), CD → EnA (OR = 1.54, p = 5.50 × 10^{− 1 8}), and UC → EnA (OR = 1.57, p = 1.20 × 10^{− 8}) (Fig. 7; STable 27). By contrast, IBD → ReA was only nominally significant, and no reverse-direction association survived multiple-testing correction. Sensitivity analyses refined these findings. IBD → AS remained robust across MR-Egger intercept, MR-PRESSO global, and leave-one-out analyses, whereas IBD → PsA, CD → AS, CD → EnA, and UC → EnA remained statistically significant but showed heterogeneity and/or outlier signals in MR-PRESSO global tests and Cochran’s Q statistics, and were therefore interpreted more cautiously (STables 28–29). Steiger directionality testing supported the forward direction for all five IVW-significant associations, including IBD → AS (112/112 SNPs), IBD → PsA (113/115), CD → AS (80/80), CD → EnA (79/79), and UC → EnA (57/57) (STable 30).

Fig. 7.

Bidirectional Mendelian randomisation between IBD traits and SpA entities. Forest/point estimates for MR analyses testing IBD (overall), CD and UC as exposures and AS, PsA, EnA and ReA as outcomes, and the corresponding reverse tests. * p < 0.05; ** p < 0.01; *** p < bonferroni-corrected threshold for all MR comparisons. Forward MR from IBD—especially CD and UC—to AS, PsA and EnA showed the most consistent and robust effects across MR estimators, whereas reverse (SpA→IBD) effects were weaker or only nominal, reflecting both true asymmetry and the smaller instrument sets available for some SpA GWAS

Discussion

Clinical co-occurrence of IBD and SpAs has long suggested that these conditions are not two fully independent inflammatory syndromes, but manifestations along a broader spectrum. Previous genome-wide studies have shown that many immune-mediated diseases share risk loci [9, 11, 56]. However, most prior work has focused on cataloguing overlapping variants, with less emphasis on how this overlap is distributed across the genome, how it differs between specific disease pairs, or how it is realized in concrete tissue and cell contexts. The present study refines this view. Rather than treating IBD–SpA comorbidity as a single undifferentiated phenomenon, our framework decomposes it into several interpretable layers: a globally positive but heterogeneous genetic correlation; regional loci that contribute disproportionately to this overlap; convergent enrichment in immune- and barrier-related tissues and cell types; and, finally, a more modest directional component from intestinal inflammatory liability to selected SpA phenotypes.

A first major observation is that genetic sharing across the IBD–SpA axis is extensive but far from uniform. Bivariate LDSC showed significant genetic correlations for most IBD–SpA pairs, with EA and, to a lesser extent, AS showing the strongest links to IBD, especially to CD. This pattern mirrors clinical series in which intestinal inflammation most often co-occurs with axial SpA or EA, indicating that the genetic signal is capturing a recognized comorbidity rather than a statistical artefact [53, 54]. Local genetic-correlation analyses further suggested that these genome-wide signals are assembled from multiple LD-independent regions that vary in magnitude and, in some cases, in direction, and that may differ between IBD-derived and SpA-derived risk loci. However, because many regional patterns did not remain significant after multiple-testing correction, these findings are best interpreted as evidence of subtype-dependent regional heterogeneity rather than definitive subtype-specific asymmetry. Instead of supporting a single dominant pathway, this layered architecture is more compatible with a model in which a shared mucosal–immune background confers broad susceptibility, while pair-specific biological contexts shape the eventual clinical phenotype.

The cross-trait loci identified here are therefore not simply the expansion of locus counts, but the use of pair-specific cross-trait weighting to sharpen biologically coherent gut–joint signals. In line with prior cross-disorder analyses [57, 58], we recover shared immune loci, but the analysis goes further by showing that part of the gut–joint signal becomes visible only after incorporating information from the paired trait. In this sense, the key unit of inference in this study is not the single disease alone, but the weighted IBD–SpA pair. Several loci emerged only under such pair-specific cross-trait modelling, including the IBD–PsA variants rs2891409, rs34426521, and rs62324212. This indicates that part of the gut–joint axis lies below the discovery threshold of conventional single-disease GWAS and becomes detectable only when the comorbid context is modelled. Thus, the study helps bridge a gap between broad cross-disease clustering and locus-level evidence.

At the same time, multi-trait GWAS requires careful interpretation. Recent work on autoimmune pleiotropy has emphasized that effect directions can diverge between immune traits and that multi-trait methods can pull estimates toward the better-powered phenotype [59]. These considerations motivated our analytical design: trait pairs were first restricted to those with clear genome-wide genetic correlation; cross-trait analysis were then applied at the genome-wide level; and prioritized signals were subsequently evaluated using regional and locus-level evidence, and colocalization analyses. Importantly, the value of this framework lies not only in discovery, but in resolution: once pair-specific weighted signals are generated, they can be projected onto tissues, cell types, and transcriptomic programs that are more immediately interpretable biologically. Across LDSC-SEG, tissue-level TWAS, and immune-cell-informed transcriptomic models, immune-rich compartments, including peripheral blood, lymphoid tissues, synovial-fluid-related contexts, and mixed T-lineage and myeloid populations. These priorities are concordant with recent single-cell studies in IBD showing activated mucosal T-cell states together with inflammatory monocyte–macrophage programs, as well as synovial studies in SpA highlighting Th17/TRM17-like states, dysfunctional regulatory T cells, and activated myeloid milieus as dominant local actors [60].

By intersecting TWAS and SMR signals derived from the corresponding MTAG-weighted cross-trait analyses, we obtained compact overlap sets. These overlap sets varied in size and composition, reinforcing the idea that different comorbidity pairs are unlikely to be driven by exactly the same effectors. At the same time, several genes recurred across multiple pairs. In particular, IL1RL1 was retained across all nine pair-specific overlap sets, while LINC01882 and CXCR2 recurred across eight, and PIM3, AC016747.3, SATB2, AHSA2P, ATG16L1, CENPO, BLTP1, DNAJC27–AS1, ADCY3, and ZFP36L2 appeared repeatedly across multiple pairs. These genes should not all be interpreted as equivalent or fully validated causal mediators; rather, they represent a hierarchy of recurrent candidates supported by convergent transcriptomic evidence in a pair-specific context. This distinction is important, because the main contribution of this gene layer is to move from diffuse shared architecture toward a tractable, subtype-aware set of mechanistic hypotheses.

The gsMap analysis adds a complementary spatial dimension, but it should be interpreted with appropriate caution. In this study, the spatial projection was not intended to function as a definitive atlas of adult inflammatory disease, but rather as a developmentally informed scaffold for asking whether independently prioritized signals show coherent anatomical organization. Under this interpretation, the observation that IBD-weighted signals cluster in gut- and immune-related territories, whereas SpA-weighted signals show more heterogeneous peripheral distributions, is broadly consistent with the gut–joint axis model. However, the translational relevance of this spatial layer remains provisional until validated in adult human disease-resolved datasets.

An important question raised by these findings is whether shared genetic background alone is sufficient to explain IBD–SpA comorbidity, or whether intestinal inflammatory liability also contributes directionally to selected SpA phenotypes. In the present analysis, bidirectional MR provided the strongest support for a forward effect of IBD on AS, which remained significant after sensitivity analysis. Additional forward associations involving IBD → PsA, CD → AS, CD → EnA, and UC → EnA were also statistically supported and directionally consistent, but showed evidence of heterogeneity and/or outlier structure in sensitivity analyses and therefore require more cautious interpretation. By contrast, reverse-direction associations did not survive multiple-testing correction and were less consistently supported by directionality testing. Taken together, these results support a predominantly forward component from intestinal inflammatory liability to selected SpA phenotypes, with the strongest evidence for IBD → AS, rather than equally robust evidence for a broad bidirectional causal model across all subtype pairs.

From a translational perspective, this distinction matters. The present findings do not immediately nominate a single therapeutic target for the entire IBD–SpA spectrum. Instead, they suggest that different weighted IBD–SpA pairs may point to different tractable effector pathways and candidate targets. Some genes are biologically appealing, while others may be more informative for subtype-specific biology than for broad therapeutic development. In this sense, the most immediate translational value of the study lies in prioritization: it narrows a large and heterogeneous shared-risk landscape into a ranked set of loci, genes, tissues, and experimental contexts for the next stage of mechanistic work. Pair-specific follow-up, for example, testing recurrent candidates in intestinal organoids or immune–epithelial co-culture systems may therefore be more informative than searching for a single universal gut–joint target.

Conclusion

Overall, this study suggests that IBD and SpA share a structured but heterogeneous genetic relationship. By integrating cross-trait, regional, and tissue/cell-context analyses, it refines the broad observation of gut–joint overlap into a more specific framework of where this overlap may arise and how it may differ across disease pairs. These findings provide a prioritized foundation for future mechanistic investigation.

Limitations

Several limitations should be considered when interpreting the findings. First, all genetic analyses and downstream contextual inferences were based primarily on European-ancestry datasets. Caution is needed when extending these results to non-European populations. Second, because harmonized age- or sex-stratified summary statistics were not consistently available across all IBD and SpA subtype datasets. We therefore could not directly evaluate whether the shared architecture differed by age or sex. Nevertheless, this limitation does not diminish the value of the prioritized loci, tissues, cellular contexts, and candidate genes as a reference for subsequent work. Third, the study integrated cross-trait mapping, transcriptomic prioritization, and multiple contextual resources to identify convergent biology. Although this approach improved biological interpretability, it does not by itself establish definitive causal mechanisms, and the prioritized loci and genes should be viewed as candidate effectors rather than confirmed causal drivers. In addition, the spatial component is best interpreted as a supportive, hypothesis-generating layer rather than a definitive map of adult human inflammatory disease. Functional follow-up studies, including intestinal organoid-based experiments, are being developed as part of our next-stage work. These considerations do not alter the central purpose of this work, which was to construct a structured, pair-specific framework that resolves broad gut–joint overlap into prioritized loci, tissues, cellular contexts, and candidate effectors for future mechanistic and translational investigation.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

Liangchun Wang: Methodology, Software, Data Analysis, Visualization and original draft preparation. Fangyan Li: Data Analysis, Visualization. Jianming Wang: Data Analysis. Hui Wang: Visualization. Yanqing Wang: Visualization, Conceptualization, Supervision, Writing and Editing.

Funding

This work is supported by Young Scientists Fund of the National Natural Science Foundation of China (82201383) and Shandong National Natural Science Foundation (ZR2025MS1247).

Data availability

We utilized GWAS summary statistics from the International Inflammatory Bowel Disease Genetics Consortium (IIBDGC). Summary statistics for spondyloarthritis (SpA) subtypes (SpAs) were obtained from FINNGEN, Yang Lab, and the GWAS Catalog. These datasets were selected based on comprehensive consideration of sample size, data quality, and the number of SNPs. Transcriptomic data were derived from the GTEx v8 release. Additional processed results supporting the findings of this study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

All authors have approved the submission.

Competing interests

We have no potential conflicts of interest to disclose.