Identification of specific brain cell types associated with the interplay between inflammatory bowel diseases and neuropsychiatric disorders by Cell-Stratified Mendelian Randomisation

Summary

Background

Inflammatory bowel disease (IBD) has been suggested to be associated with neuropsychiatric disorders, but the underlying mechanisms remain poorly understood.

Methods

We employed Cell-Stratified Mendelian Randomisation (csMR) to analyse genetic variants associated with IBD and their effects on neuropsychiatric disorders. We conducted two-sample MR analyses using the partitioned genetic variants and performed SuSiE co-localisation analysis to identify shared causal variants between GWAS traits and the single-cell eQTL data. We also examined the imaging-derived phenotypes (IDPs) to understand the structural changes in the brain.

Findings

We identified specific cell types and IDPs associated with the interplay between IBD and neuropsychiatric disorders. Importantly, the oligodendrocytes (ODC)-stratified variants associated with IBD were linked to multiple sclerosis and schizophrenia, astrocytes-stratified variants associated with ulcerative colitis (UC) were connected to obsessive-compulsive disorder and schizophrenia, and inhibitory neurons -stratified variants associated with Crohn's disease (CD) were linked to multiple sclerosis. Moreover, mediation analysis suggested that 31.4% of the association from UC to schizophrenia was mediated by alterations in mean diffusivity in the left tapetum, while structural changes in the right inferior temporal region associated with CD accounted for a 37.5% increased risk of cerebral aneurysm.

Interpretation

Our findings may facilitate understanding of the molecular mechanisms involved in diseases modulated by the gut-brain axis and develop novel therapeutic strategies for IBD and neuropsychiatric disorders.

Funding

CZ was supported by National Key Research and Development Program of China (2023YFC2705700), and operational funds from The First Affiliated Hospital of Nanchang University (500021010). LZ was supported by National Natural Science Foundation of China (82160155).

Research in context.

Evidence before this study

Inflammatory bowel disease (IBD), encompassing Crohn's disease and ulcerative colitis, has been linked in clinical observations and epidemiological research to a broad spectrum of neuropsychiatric conditions, including mood and anxiety disorders, obsessive-compulsive disorder, schizophrenia, and neurodegenerative diseases such as multiple sclerosis and Parkinson's disease. Prior investigations have documented higher rates of anxiety and depression among IBD patients and reported associations between systemic inflammation and changes in brain structure and function. However, these studies have generally relied on observational designs or bulk-tissue analyses that mask cell-type specificity, limiting mechanistic insights into how gut inflammation translates to brain pathology.

Added value of this study

Building on emerging single-cell and imaging genomics resources, we integrate brain cell-type-specific expression quantitative trait loci with genome-wide association studies for inflammatory bowel disease and neuropsychiatric disorders through a Cell-Stratified Mendelian Randomisation framework. By combining SuSiE co-localisation, two-sample MR, and mediation analysis using imaging-derived phenotypes, we uncover how genetic variants influencing specific brain cells, especially oligodendrocytes, astrocytes, excitatory and inhibitory neurons, may mediate risk pathways from IBD to conditions such as schizophrenia, multiple sclerosis, obsessive-compulsive disorder, and non-ruptured cerebral aneurysm. This cell-resolved approach clarifies mechanistic links hidden in bulk analyses and highlights brain-structure changes (e.g., tapetum mean diffusivity, inferior temporal cortical area) as key mediators.

Implications of all the available evidence

Our study together with all the available evidence suggest that the interplay between IBD and neuropsychiatric disorders could be mediated by specific cell types in the brain. These findings may help understand the molecular mechanisms involved in the diseases modulated by the gut-brain axis and develop novel therapeutic strategies for IBD and neuropsychiatric disorders.

Introduction

Inflammatory bowel disease (IBD), including Crohn's disease (CD) and ulcerative colitis (UC), has been suggested to be associated with neuropsychiatric disorders, such as anxiety, depression, and neurodegenerative diseases.1, 2, 3, 4, 5, 6 Despite growing clinical evidence supporting this link, the underlying mechanisms remain poorly understood.

The relationship between IBD and neuropsychiatric disorders is well-documented across various lines of evidence, though its strength varies depending on the specific condition in question. Epidemiological studies have consistently shown an increased risk of psychiatric comorbidities in IBD patients. For instance, a meta-analysis reported a pooled odds ratio (OR) of 2.3 (95% CI [1.9, 2.8]) for anxiety and 1.8 (95% CI [1.5, 2.2]) for depression compared to controls, highlighting a robust association with IBD.⁷ Similarly, a Danish population-based study observed a 1.6-fold increased risk of schizophrenia among IBD patients,⁸ while a German cohort found a hazard ratio of 2.35 (95% CI [1.47, 3.78]) for multiple sclerosis (MS) in UC patients.⁹ Experimental studies complement these findings, with animal models of colitis demonstrating behaviour changes and heightened neuroinflammation, driven by pro-inflammatory cytokines.¹⁰ These models suggest a biologically plausible link through the gut-brain axis, where systemic inflammation may disrupt neural function. Furthermore, genetic analyses provide additional support, with linkage disequilibrium score regression (LDSC) estimating modest but significant shared heritability between UC and neuropsychiatric disorders, for example, a genetic correlation (rg) of 0.14 (P = 0.00020) with schizophrenia and 0.23 (P = 0.0015) with bipolar disorder (BD).¹¹ These correlations, though not large, point to overlapping immune-related pathways that may underpin both conditions. Collectively, epidemiological, experimental, and genetic studies converge to demonstrate that IBD is linked to a heightened risk of diverse neuropsychiatric disorders.

Single-cell transcriptomics has become a powerful approach to explore cellular heterogeneity in complex tissues, such as the brain.¹² This technique allows researchers to examine gene expressions at a cell-type-specific level, providing insights into understanding the contributions of different brain cell types to disease pathology. This is particularly important for conditions like IBD, where multiple brain cell types may play distinct roles in mediating the disease's impact on neuropsychiatric disorders. While the single-cell data can reveal cell-specific gene expression profiles, they do not directly establish potential causal links between genetic variants and disease outcomes. Mendelian randomisation (MR) is a method that leverages genetic variants as instrumental variables to infer potential causal relationships between exposures and diseases. Traditional MR approaches, based on genome-wide association study (GWAS) data, often fail to account for the specificity of gene expression across different cell types, potentially missing key insights into disease mechanisms.¹³ To address this limitation, a new framework called Cell-Stratified Mendelian Randomisation (csMR) has been developed.¹⁴ This approach integrates single-cell eQTL (expression quantitative trait loci) data with GWAS data to investigate how genetic variants influence gene expression in specific cell types, offering a more precise understanding of the genetic effects on brain cells involved in neuropsychiatric disorders.¹⁴

In addition to uncovering cell-type-specific genetic effects, it is important to investigate the pathways through which IBD impacts the brain structure and function. Mediation analysis using brain imaging-derived phenotypes (IDPs) provides a way to explore whether changes in brain structure mediate the relationship between IBD and neuropsychiatric disorders. This method allows researchers to distinguish between direct effects of IBD on brain function and indirect effects mediated by structural changes in the brain.

In this study, we combine csMR with mediation analysis to explore the cell-type-specific genetic mechanisms underlying the relationship between IBD and neuropsychiatric disorders. By integrating single-cell transcriptomics and neuroimaging data, we aim to provide a comprehensive understanding of how genetic variants associated with IBD influence brain cell types and potential lead to neuropsychiatric outcomes. This integrated approach has the potential to uncover novel therapeutic targets for both IBD and related neuropsychiatric conditions.

Methods

This study followed the guidelines outlined by the Strengthening the Reporting of Observational Studies in Epidemiology,¹⁵ completed with the Mendelian Randomisation extension.

Study overview

Our study employed a multi-step analytical framework (Fig. 1) to resolve cell-type-specific mechanisms underlying the IBD-neuropsychiatric interplay through the csMR framework.¹⁴ First, we performed SuSiE co-localisation¹⁶ analysis to identify shared causal variants between IBD-associated loci and brain cell-type-specific eQTLs. Variants with strong evidence of SuSiE colocalisation were prioritised as candidate instrumental variables for subsequent csMR analysis, ensuring that these singlenucleotide polymorphisms (SNPs) directly influenced both IBD risk and cell-type-specific gene expression. Second, cell-stratified SNPs were constructed by grouping these candidate instrumental variables according to their cell-type-specific regulatory activity (e.g., ODC, astrocyte), thereby enabling csMR to infer potential disentangle direct cell-stratified causal effects of IBD subtypes on neuropsychiatric disorders. Third, mediation analysis tested whether structural brain IDPs mediated these relationships. Finally, genetic prioritisation leveraged co-localisation probabilities to resolve key drivers of the observed interplay. This workflow integrates multiple data sources, including GWAS, brain cell type cis-eQTLs, and brain IDPs, to uncover the complex relationships between the genetic variants and gene expression in specific brain cells, and traits of interest.

Fig. 1.

Study workflow of csMR for investigating IBD-neuropsychiatric interplay. The framework includes three steps: (1) SuSiE co-localisation and cell-stratified SNP construction to identify shared causal variants (PPH4 > 0.8) between IBD-associated loci and brain cell-type-specific eQTLs using single-cell eQTL data and GWAS summary statistics. Prioritised variants are grouped by their cell-type-specific regulatory activity to serve as instrumental variables for csMR; (2) csMR analysis to estimate direct potential causal effects of IBD subtypes (UC, CD) on neuropsychiatric disorders and brain IDPs through cell-type-specific pathways; (3) Mediation analysis to test whether structural brain IDPs mediate IBD-neuropsychiatric relationships. Visualisation conventions: Orange dashed arrows indicate the SuSiE co-localisation process (Step 1). Dark blue unidirectional arrows represent the primary two-sample MR analysis (Step 2). Light blue arrows indicate the MR mediation analysis (Step 3). Grey labels identify key analytical components (Cell-stratified SNP, Exposure, Mediator, Outcome). Abbreviations: IBD, inflammatory bowel disease; UC, ulcerative colitis; CD, Crohn's disease; AD, Alzheimer's disease; ALS, amyotrophic lateral sclerosis; CTS, carpal tunnel syndrome; CA, cerebral aneurysm, non-ruptured; MS, multiple sclerosis; EP, epilepsy; PD, Parkinson's disease; SD, sleep disorders; AND, anxiety disorder; ADHD, attention deficit hyperactivity disorder; ASD, autism spectrum disorder; MDD, major depressive disorder; SCP, schizophrenia; OCD: obsessive-compulsive disorder; MD, mean diffusivity; FA, fractional anisotropy; IDPs, imaging-derived phenotypes; IVs, instrumental variables; LD, linkage disequilibrium; MR, Mendelian randomisation; AST, astrocyte; ODC, oligodendrocyte; EC, endothelial cell; MIC, microglia; ExN, excitatory neuron; OPC, oligodendrocyte precursor cell; PC, pericyte; InN: inhibitory neuron; IDPs, imaging-derived phenotypes.

Study population and data sources

GWAS data for IBD, UC and CD

The GWAS data for IBD (including CD and UC) were obtained from the IEU GWAS database, which incorporates data from the International Inflammatory Bowel Disease Genetics Consortium (IIBDGC).¹⁷ All participants in these studies were of European ancestry. The IBD GWAS dataset includes 12,882 cases and 21,770 controls. The UC GWAS data comprise 6968 cases and 20,464 controls, and the CD GWAS data include 5956 cases and 14,927 controls. Diagnoses were based on established endoscopic, radiological, and histopathological criteria. Detailed GWAS datasets, including cohort demographics, are summarised in Supplementary Table S1.

Single-cell eQTL data for brain cell types

Summary-level brain single-cell eQTL data were obtained from Bryois et al.,¹² which analysed prefrontal and temporal cortices from 192 postmortem brains of European-ancestry individuals. This dataset includes eQTLs for eight cell types (astrocytes, endothelial cells, excitatory neurons, inhibitory neurons, microglia, oligodendrocytes, oligodendrocyte precursor cells, and pericytes). Gene expression was quantified for ∼14,595 genes, and 5.3 million SNPs were genotyped. Further details are provided in Supplementary Table S1.

GWAS data for neuropsychiatric disorders

Our study focuses on the following 14 neuropsychiatric disorders (NDs): Alzheimer's disease (AD), amyotrophic lateral sclerosis (ALS), carpal tunnel syndrome (CTS), non-ruptured cerebral aneurysm (CA), multiple sclerosis, epilepsy, Parkinson's disease (PD), sleep disorders, anxiety disorder, attention deficit hyperactivity disorder (ADHD), autism spectrum disorder (ASD), major depressive disorder (MDD), schizophrenia, obsessive-compulsive disorder (OCD). All GWAS summary statistics were derived from European-ancestry cohorts unless otherwise noted (see Supplementary Table S2 for further details).18, 19, 20, 21, 22, 23, 24, 25, 26

GWAS data for brain IDPs

The GWAS summary statistics for brain IDPs were obtained from the Oxford Brain Imaging Genetics Server (BIG40, https://open.win.ox.ac.uk/ukbiobank/big40). The original study analysed the brain imaging data of 39,691 participants from the UK Biobank,²⁷ which includes 39,691 participants of predominantly European ancestry (94% white British). We selected 477 IDPs from the original dataset of 3935 IDPs, including 68 cortical area metrics, 68 cortical thickness metrics, 66 cortical volume metrics, 179 subcortical volume measures, and 96 white matter tract measures. Detailed information was provided in Supplementary Table S3.

SuSiE Co-localisation analysis

We performed SuSiE co-localisation analysis to identify shared causal variants between GWAS traits and the single-cell eQTL data using the updated “coloc” package (v5),²⁸ and integrated the SuSiE method for fine mapping.²⁹ SuSiE was recommended due to its superior performance in scenarios involving multiple causal variants.²⁸ Linkage disequilibrium (LD) matrices were constructed utilising the reference genome from the 1000 Genomes Project for individuals of European ancestry (1000G EUR), excluding variants in the MHC region (chr6:28477797-33448354, GRCh37). Credible sets were defined with 90% coverage, and co-localisation between GWAS and gene expression within 100 kb flanking regions was evaluated with default parameters. Posterior probabilities were calculated for five hypotheses: no association (PPH0), association with trait 1 only (PPH1), association with trait 2 only (PPH2), association with both traits but different causal variants (PPH3), and co-localisation with common causal variants (PPH4). A PPH4 >0.8 threshold was used to indicate strong co-localisation,^30,31 and these variants were considered candidate instrumental variables for subsequent MR analysis.

Cell-stratified Mendelian Randomisation

The core analytical method used in this study is csMR, an extension of conventional MR that stratifies instrumental variables (IVs) based on their cell-type-specific eQTL associations.¹⁴ Conventional MR uses all GWAS variants as IVs, whereas csMR refines this method by selecting SNPs based on their impact on gene expression in defined cell populations, offering a more precise view of potential causal pathways. The csMR framework allows for the inference of potential causal relationships between genetic variants and complex traits at the cellular level by examining whether GWAS variants that affect specific brain cell types also contribute to traits like IBD, UC, and CD.

Selection and quality control of genetic instrumental variables

Qualified SNPs from co-localisation analysis were selected for MR, ensuring they met three core assumptions³²: (1) strong association with exposure (relevance); (2) no association with confounders (independence); and (3) association with the outcome only through exposure (exclusion restriction). SNPs were initially filtered for strong exposure association (P < 5.0 × 10⁻⁸, Wald test)³³ and further assessed whether they or their proxy variants (r² > 0.80)³⁴ linked to major confounders, including smoking, alcohol consumption, education, and income.35, 36, 37, 38 In our study, confounder selection was guided by the modified disjunctive cause criterion.³⁹ We identified smoking, alcohol consumption, education, and income as confounders based on their established roles as common causes of both the exposure (IBD, UC, CD) and the outcome (neuropsychiatric disorders), supported by prior MR studies and biological evidence.35, 36, 37, 38 These variables influence IBD risk (e.g., via inflammation, gut health) and neuropsychiatric disorders (e.g., via neuroinflammation, socioeconomic stress), fulfilling the criterion of being causes of the exposure, outcome, or both. We did not exclude any as instrumental variables, as they are not known to affect IBD without also influencing neuropsychiatric outcomes independently. Additionally, these confounders serve as proxies for unmeasured socioeconomic and lifestyle factors (e.g., stress, healthcare access), aligning with the criterion's recommendation to include proxies for unmeasured common causes. Index SNPs were identified through LD clumping using PLINK (r² < 0.001, window size 10,000 kb, P < 5.0 × 10⁻⁸, Wald test).^34,40,41 SNPs were then matched with outcome data, with proxies used for any missing variants. Data harmonisation was performed with the “TwoSampleMR” v0.5.6 R package,⁴¹ to ensure allele correspondence, while palindromic SNPs with intermediate allele frequencies (>0.42) were removed. For SNPs satisfying the MR assumptions, quality control included heterogeneity tests using “RadialMR” v1.0.⁴² Outlier SNPs were identified through Cochran's Q and Rucker's Q'tests,⁴³ and those exceeding a P-value of 0.05 were discarded. The F-statistic was calculated to evaluate IV power, with a required minimum of F > 10 to prevent bias.⁴⁴

Two-sample MR and pleiotropy assessment

Two sample MR analyses estimated associations between genetic instruments for each brain cell type and diseases, employing methods like inverse-variance weighted (IVW), Mendelian randomisation-Egger (MR-Egger), weighted median, MR using a Robust Adjusted Profile Score (MR-RAPS), and weighted mode to ensure robust findings.^43,45, 46, 47 Reverse causality was tested by swapping exposure and outcome roles. We used “Two-Sample MR” v0.5.6 R package to conduct these MR analyses. To account for multiple comparisons, we applied false discovery rate (FDR) correction using the Benjamini-Hochberg procedure. For each exposure (IBD, UC, CD), we analysed associations with 14 neuropsychiatric disorders across 8 brain cell types, resulting in 14 × 8 = 112 independent tests per exposure. The FDR-adjusted P-values were calculated to control the expected proportion of false positives at 5% (P_fdr < 0.05, FDR-adjusted). Significant associations were reported based on this threshold. Associations with P_fdr ≥0.05 but a raw P < 0.05 were classified as suggestive. This approach reduces the risk of Type II errors while maintaining robust control over false positives, particularly given potential correlations among cell-type-specific causal effects. Pleiotropy analyses were performed to determine the main MR method, beginning with MR-PRESSO to detect horizontal pleiotropy,⁴⁸ followed by MR-Egger regression for directional pleiotropy.⁴⁹ Cochran's Q and Rucker's Q′ statistics were calculated to assess heterogeneity.⁵⁰ The main MR method was selected as follows: (a) Wald ratio for single instruments; (b) inverse-variance weighted (IVW) for multiple instruments without directional pleiotropy (P > 0.05)⁵¹; (c) MR-Egger if directional pleiotropy was detected (P > 0.05); and (d) weighted median or mode if P < 0.05 for the test of Rucker's Q, with a preference for weighted median due to its higher power.⁵¹ Finally, robustness was assessed with leave-one-out sensitivity analyses to identify influential variants. Results of pleiotropy and sensitivity analyses were summarised in a separate file generated by csMR, providing recommendations for selecting appropriate MR results.

Sample size justification

In our Cell-Stratified Mendelian Randomisation study examining the causal relationships between IBD, UC, CD, and various neuropsychiatric outcomes, we provide a robust justification for the sample size employed. This justification rests on three pillars: the strength of IVs, calculations of statistical power, and adjustments for multiple testing. The reliability of causal estimates in MR studies hinges on the strength of the instrumental variables used. Instrument strength is evaluated using the F-statistic, with a threshold of F > 10 widely accepted as indicative of robust IVs. To ensure our study was sufficiently powered, we calculated the sample sizes required to achieve 80% power for each exposure-outcome-cell type combination using tools like mRnd (https://shiny.cnsgenomics.com/mRnd/). These calculations considered the expected effect size, the variance explained by the IVs (R²), and the desired power level. We then compared these required sample sizes to the actual sample size reported in the study. Given the large number of statistical tests conducted across multiple exposures (IBD, UC, CD), outcomes (e.g., schizophrenia, multiple sclerosis, OCD), and cell types (e.g., astrocytes, oligodendrocytes, excitatory neurons), controlling for multiple testing was essential to minimise Type I errors. We applied a rigorous Bonferroni correction based on 112 tests (14 outcomes × 8 cell types), setting a significance threshold of α = 0.05/112 ≈ 4.46 × 10⁻⁴.

Mediation analysis

Mediation analysis was conducted to investigate whether brain imaging-derived phenotypes (IDPs) mediate the relationship between IBD, UC, CD, and neuropsychiatric disorders. In our two-step MR mediation analysis, we first assessed the potential causal effect of IBD on brain IDPs using genetic instruments specific to IBD, excluding SNPs associated with confounders like smoking, alcohol consumption, education, and income (P < 5.0 × 10⁻⁸, Wald test). Next, we examined the potential causal effect of brain IDPs on neuropsychiatric disorders with a separate set of genetic instruments specific to brain IDPs, applying the same confounder exclusion criteria. This methodology, consistent with Guo et al.⁵¹ and Lin et al.,⁵² ensures that instruments for both the exposure and mediator meet MR assumptions, minimising bias from unmeasured confounding. This analysis aims to clarify the mechanistic pathways through which IBD, UC, and CD may influence brain structure, subsequently affecting psychiatric outcomes. The mediation framework distinguishes between the direct effects of IBD, UC, and CD on neuropsychiatric disorders and the indirect effects mediated by changes in brain structure (IDPs). The direct effect represents the potential causal impact of IBD, UC, and CD on neuropsychiatric disorders. The indirect effect is calculated as the product of the causal estimates for IBD, UC, and CD on IDPs and the causal estimates for IDPs on neuropsychiatric disorders. The mediation proportion is derived by dividing the indirect effect by the direct effect. To estimate these mediation effects robustly, we employed the Monte Carlo method,⁵³ a simulation-based technique widely used to model uncertainty in statistical analyses. This method leverages random sampling to generate a distribution of possible outcomes, making it particularly well-suited for mediation analysis with summary-level data from GWAS, where individual-level data are often unavailable. In our study, we applied the Monte Carlo method by first obtaining the point estimates and standard errors of the causal effects from two-sample MR analyses: the effect of the exposure (IBD, UC, or CD) on the mediator (IDPs), denoted as β₁, and the effect of the mediator on the outcome (neuropsychiatric disorders), denoted as β₂. We then conducted 10,000 simulation iterations, drawing random samples from the normal distributions of β₁ and β₂, which is a reasonable assumption given the large sample sizes typical of GWAS. For each iteration, the indirect effect was computed as β₁ × β₂, and the direct effect was calculated as the total effect minus the indirect effect. The final point estimate for the mediation effect was the mean of the simulated indirect effects, with 95% confidence intervals (CIs) derived from the 2.5th and 97.5th percentiles of the simulated distribution. In this study, we applied a two-step MR framework for mediation analysis. Compared to multivariable MR (MVMR), this method provides stronger interpretability for single-mediator paths, simpler modelling, and reduced instrument needs, aiding power in targeted studies.⁵⁴ Yet, it may face biases from overlooked correlated mediators or exposure-mediator interactions, whereas MVMR adjusts for multiple factors to enhance pleiotropy handling but calls for sturdier instruments and more computation.^55,56 With our focus on standalone IDPs as mediators, the two-step method was chosen for its fit and consistency with standard MR mediation practices.

Statistics

MR analyses, including sensitivity tests, were conducted using the‘TwoSampleMR’ package (v0.5.6) in R version 4.2.3 (https://www.r-project.org/). The tophits function from the ‘ieugwasr'package (v1.0.0) and the from ‘JSON’ function in the ‘jsonlite’ package (v1.8.8) were utilised to query associations between IVs and confounders via the REST API of IEU OpenGWAS and the GWAS Catalogue.^41,57 Cortical structures were visualised with BrainNet Viewer⁵⁸ (http://www.nitrc.org/projects/bnv/), while subcortical structures were depicted using the ‘ggseg’ package (v1.6.6). Visualisation of white matter tracts was performed using the ‘ggseg3d’ (v1.6.3) packages.

Ethics

All MR analyses were based on publicly available GWAS datasets, which have received appropriate ethical approvals, with written informed consent from participants. As this study did not involve individual-level data, no additional ethical review board approval was necessary.

Role of the funders

The funders had no role in study design, data collection, data analyses, interpretation, or writing of this report

Results

Cell-type-specific effects of IBD, UC, and CD risk variants on brain cell gene expression

We investigated the cell-type-specific effects of genetic variants associated with IBD, CD, and UC on brain gene expression using SuSiE co-localisation analysis across eight distinct brain cell types, leveraging GWAS summary statistics and single-cell eQTL data.

Effects of SNPs associated with IBD on brain cell gene expression

As shown in Fig. 2a and b, and Supplementary Table S4, we identified substantial co-localised SNPs and genes linked to IBD across several brain cell types. The highest number of co-localised SNPs was observed in excitatory neurons, followed by oligodendrocytes. Cell type-specific SNPs were most prevalent in excitatory neurons (58.82%), astrocytes (50.00%), oligodendrocyte precursor cells (50.00%), and oligodendrocytes (43.75%). This indicates that IBD-associated SNPs regulate gene expression in a highly cell-type-specific manner. Analysis of co-localised genes (Fig. 2b) supported these findings, with ExN having the most co-localised genes, followed by oligodendrocytes and astrocyte. Cell type-specific co-localised genes were particularly prominent in oligodendrocyte precursor cell (50.00%), oligodendrocytes (38.46%), and astrocyte (25.00%). The heatmap (Fig. 2c) and histogram (Fig. 2d) further demonstrate that IBD-associated SNPs predominantly affect one cell type. Importantly, strong co-localisation evidence for RNFT1 was only found in oligodendrocytes (PPH4 = 0.9418, Supplementary Table S4), suggesting its unique role in mediating the effects of IBD-associated variants.

Fig. 2.

Genetic co-localisation analysis of IBD risk variants and gene expression across brain cell types (IBD GWAS: 12,882 cases and 21,770 controls; single-cell eQTL: 192 postmortem brains). (a) Total and cell-specific co-localised SNPs identified in each cell type. Y-axis: Number of SNPs. X-axis: Eight major brain cell types. (b) Total and cell-specific co-localised genes identified in each cell type. Y-axis: Number of genes. X-axis: Eight major brain cell types. (c) Heatmap of average absolute eQTL effect sizes (|Average β_abs|) for colocalised SNPs. For each cell-type pair, we calculated the average |Average β_abs| of colocalised SNPs in the “test” cell type (Y-axis) versus their average |Average β_abs| in the “reference” cell type (X-axis). (d) Proportion of SNPs co-localised in different numbers of cell types, with most SNPs specific to a single cell type. X-axis: Number of cell types in which SNPs show colocalisation (ranging from 1 to 7). Y-axis: Density (proportion of all colocalized SNPs, %). Abbreviations: IBD, inflammatory bowel disease; SNP, single nucleotide polymorphism; eQTL, expression quantitative trait locus; ExN, excitatory neuron; ODC, oligodendrocyte; AST, astrocyte; OPC, oligodendrocyte precursor cell; MIC, microglia; InN, inhibitory neuron; EC, endothelial; PC, pericyte.

Effects of SNPs associated with UC on brain cell gene expression

As shown in Supplementary Fig. S1a, b, and Supplementary Table S4, we identified significant number of co-localised SNPs and genes linked to UC across brain cell types. Excitatory neurons had the highest number of total co-localised SNPs. Cell type-specific co-localised SNPs were most prevalent in astrocytes and microglia (100% each), followed by excitatory neurons (62.5%) and oligodendrocyte precursor cell (66.67%). The analysis of co-localised genes (Supplementary Fig. S1b) showed a similar trend, with excitatory neurons having the most cell type-specific genes, followed by astrocytes and microglia. Cell type-specific co-localised genes were more prevalent in excitatory neurons (87.50%), astrocytes (75.00%), oligodendrocyte precursor cell (66.67%), inhibitory neuron (66.67%), oligodendrocytes (50.00%), and microglia (75.00%). UC-associated variants exert highly cell-type-specific effects, as seen in the eQTL effect sizes in Supplementary Fig. S1c, where these SNPs regulate gene expression mainly in specific cell types. Additionally, 90% of co-localised SNPs were specific to a single cell type (Supplementary Fig. S1d), emphasising strong cell-type-specific regulation, with only a small fraction shared across multiple types.

Effects of SNPs associated with CD on brain cell gene expression

As illustrated in Supplementary Fig. S2a, b, and Supplementary Table S4, we identified substantial co-localised SNPs and genes associated with CD across multiple brain cell types. Excitatory neurons had the highest number of co-localised SNPs, with a significant count in oligodendrocytes as well. Cell type-specific colocalized SNPs were particularly prevalent in Microglia (62.50%), excitatory neurons (57.14%), oligodendrocytes (53.85%), and astrocytes (25.00%). Analysis of co-localised genes (Supplementary Fig. S2b) confirmed this pattern, with excitatory neurons showing the most co-localised genes, followed by oligodendrocytes and astrocytes. cCell type-specific genes were prevalent in oligodendrocytes (38.46%), microglia (37.50%), and oligodendrocyte precursor cell (50.00%). This underscores the role of these cell types in gene expression regulation related to CD. The heatmap in Supplementary Fig. S2c shows that CD-associated SNPs exert their regulatory influence primarily in a cell-type-specific manner, with reduced effects in other cell types. Additionally, Supplementary Fig. S2d shows that 64.86% of co-localised SNPs were specific to a single cell type, further highlighting the distinct regulatory roles of individual brain cell types in CD, with only a small fraction of SNPs shared across cell types. Our findings across IBD, UC, and CD demonstrate that the genetic mechanisms driving the neuropsychiatric disorders are highly cell-type-specific, particularly in oligodendrocytes, astrocytes, excitatory neurons, and inhibitory neuron.

Cell-stratified potential causal links between IBD, UC, CD, and neuropsychiatric disorders inferred by mendelian randomisation

We firstly conducted traditional MR analyses using all GWAS associated variants to explore the potential causal relationships between IBD, UC, CD, and 14 neuropsychiatric disorders as detailed in Supplementary Table S5. In these MR analyses, we utilised variants that achieved genome-wide significance to assess the associations and potential causal links between IBD, UC, CD, and the neuropsychiatric outcomes. The MR analyses, adjusted for multiple testing via FDR correction, identified several potential causal associations between IBD subtypes and neuropsychiatric disorders (Supplementary Table S5). IBD exhibited non-significant positive association with schizophrenia (β = 0.04, 95% CI [0.010, 0.066], P = 0.0070, P_fdr = 0.14 [IVW]) and non-significant link to multiple sclerosis (β = 0.10, 95% CI [0.006, 0.191], P = 0.037, P_fdr = 0.28 [IVW]). Non-significant associations were observed between IBD and OCD (β = 0.16, 95% CI [−0.026, 0.343], P = 0.091, P_fdr = 0.32) [IVW] as well as CTS (β = −0.02, 95% CI [−0.052, 0.003], P = 0.078, P_fdr = 0.32 [IVW]), though these did not meet significance thresholds (P_fdr < 0.05). No potential causal effects were detected for IBD with other neuropsychiatric disorders (Supplementary Table S5). For UC, non-significant association with multiple sclerosis was observed (β = 0.15, 95% CI [0.035, 0.260], P = 0.010, P_fdr = 0.14 [IVW]). Non-significant associated signals for ASD (β = 0.04, 95% CI [−0.005, 0.078], P = 0.081, P_fdr = 0.75 [IVW]) and OCD (β = 0.18, 95% CI [−0.015, 0.377], P = 0.070, P_fdr = 0.75 [IVW]) did not survive FDR correction. No significant links were found between UC and other outcomes (Supplementary Table S5). CD demonstrated non-significant association with non-ruptured CA (β = 0.06, 95% CI [0.005, 0.120], P = 0.034, P_fdr = 0.21 [IVW]), and schizophrenia (β = 0.03, 95% CI [0.004, 0.058], P = 0.023, P_fdr = 0.28 [IVW]). No other neuropsychiatric disorders showed significant or suggestive links to CD (Supplementary Table S5).

Since the initial broad MR analysis did not identify significant potential causal relationships, we next performed an exploratory analysis to investigate whether any potential effects might be stratified by specific brain cell types. We performed IVs selection to identify colocalised SNPs, followed by LD pruning and outlier removal. csMR analysis, along with pleiotropy and sensitivity assessments, revealed significant cell-specific potential causal effects of IBD, UC, and CD on neuropsychiatric disorders after FDR correction (P_fdr < 0.05) to account for multiple testing (N = 112) and potential correlation among test statistics (as shown in Fig. 3, Fig. 4, Fig. 5; Supplementary Table S6). The instrumental variables demonstrated exceptional strength across analyses, with F-statistics ranging from 53.83 to 57.47 for IBD/UC/CD exposures, far exceeding the F > 10 threshold (Supplementary Table S7). An increased IBD risk, predicted by oligodendrocytes-stratified variants, was strongly associated with an elevated risk of multiple sclerosis (β = 0.63, 95% CI [0.27, 0.99], P_fdr = 0.0028 [IVW], Supplementary Fig. S3). For psychiatric disorders, IBD risk predicted by astrocyte -stratified variants was linked to a higher risk of OCD (β = 0.25, 95% CI [0.03, 0.47], P_fdr = 0.0084, Supplementary Fig. S3). Additionally, IBD risk predicted by ODC-stratified variants was associated with a heightened risk of schizophrenia (β = 0.30, 95% CI [0.19, 0.40], P_fdr < 0.0001 [IVW], Supplementary Fig. S3). No significant causal relationships between IBD and other disorders were detected across any cell types. Furthermore, MR analyses revealed cell-specific potential causal effects of UC on neuropsychiatric conditions. UC risk predicted by astrocyte -stratified variants was associated with a protective effect against CTS (β = −0.46, 95% CI [−0.66, −0.27], P_fdr < 0.0001 [Wald ratio], Supplementary Fig. S4). For psychiatric disorders, higher UC risk predicted by astrocyte -stratified variants was linked to an increased risk of OCD (β = 0.34, 95% CI [0.06, 0.62], P_fdr = 0.0051 [Wald ratio], Supplementary Fig. S4) and schizophrenia (β = 0.37, 95% CI [0.19, 0.54], P_fdr = 0.0015 [Wald ratio], Supplementary Fig. S4). In addition, UC risk predicted by excitatory neurons -stratified variants was associated with an increased risk of autism spectrum disorder (β = −0.36, 95% CI [−0.54, −0.18], P_fdr = 0.0027 [IVW], Supplementary Fig. S4). Finally, MR analyses revealed significant associations between CD and multiple sclerosis, with these relationships mediated by oligodendrocytes-stratified variants (β = 0.60, 95% CI [0.24, 0.96], P_fdr = 0.0041 [IVW], Supplementary Fig. S5) and inhibitory neurons-stratified variants (β = 0.51, 95% CI [0.17, 0.85], P_fdr = 0.0065 [IVW], Supplementary Fig. S5). Additionally, CD risk predicted by oligodendrocytes-stratified variants was linked to an increased risk of non-ruptured CA (β = 0.46, 95% CI [0.17, 0.75], P = 0.0046 [IVW], Supplementary Fig. S5) and Schizophrenia (β = 0.26, 95% CI [0.16, 0.36], P_fdr < 0.0001 [IVW], Supplementary Fig. S5). Furthermore, CD risk associated with excitatory neurons -stratified variants was found to increase the risk of CTS (β = 0.17, 95% CI [0.11, 0.23], P_fdr < 0.0001 [IVW], Supplementary Fig. S5). Of note, no significant causal relationships were found between CD and psychiatric disorders across any brain cell types. Then, we calculated the required sample size to achieve 80% power for each statistically significant exposure-outcome-cell type combination and compared these to the actual sample sizes used. For the majority of the analyses, including UC with autism spectrum disorder (required: 21,700; reported: 463,511), UC with schizophrenia (required: 28,200; reported: 130,644), UC with carpal tunnel syndrome (required: 69,400; reported: 385,304), IBD with multiple sclerosis (required: 32,960; reported: 408,561), IBD with schizophrenia (required: 56,400; reported: 130,644), CD with schizophrenia (required: 43,700; reported: 130,644), CD with multiple sclerosis (required: 19,140; reported: 408,561), CD with carpal tunnel syndrome (required: 26,160; reported: 385,304), CD with cerebral aneurysm non-rupture (required: 308,900; reported: 374,631), and CD with multiple sclerosis (inhibitory neuron; required: 31,340; reported: 408,561), the reported sample size substantially exceeded requirements, ensuring robust power. However, in specific cases involving obsessive-compulsive disorder (OCD) outcomes, such as UC → OCD (required: 782,200; reported: 135,111) and IBD → OCD (required: 847,200; reported: 135,111), the reported sample size fell short of calculated threshold. Despite this limitation, the instrumental variables for OCD analyses demonstrated exceptional strength, with F-statistics of 57.47 (UC) and 53.83 (IBD), far exceeding the conventional threshold (F > 10), which enhanced precision and partially mitigated reduced power. These results confirm that the study was sufficiently powered for most statistically significant associations across IBD subtypes (UC, CD), neuropsychiatric disorders (schizophrenia, autism, multiple sclerosis), and neurological conditions (carpal tunnel syndrome, cerebral aneurysm), while strong instruments bolstered reliability in underpowered OCD analyses. These findings suggest that IBD, UC, and CD may exert their influence on neuropsychiatric disorders through specific brain cell types.

Fig. 3.

Potential causal links between IBD and 14 neuropsychiatric disorders, assessed via cell-stratified Mendelian Randomisation (cs-MR) (IBD GWAS: 12,882 cases and 21,770 controls; single-cell eQTL: 192 postmortem brains). The heatmap shows effects of IBD on disorders, stratified by brain cell type. Potential causal associations were estimated using the Inverse-Variance Weighted (IVW) method when multiple instrumental variables were available, and the Wald ratio method when only a single instrumental variable was available for a given cell type–disorder pair with beta coefficients (β) indicating effect sizes (red for negative, blue for positive associations). Significant associations (Pfdr <0.05, Benjamini-Hochberg false discovery rate correction) are labelled with β and 95% confidence intervals (CI) and marked with asterisks. Disorders are grouped into psychiatric and neurological categories. Abbreviations: AST, astrocyte; ODC, oligodendrocyte; ExN, excitatory neuron; InN, inhibitory neuron; MIC, microglia; OPC, oligodendrocyte precursor cell; EC, endothelial cell; PC, pericyte.

Fig. 4.

Potentia causal links between UC and 14 neuropsychiatric disorders, assessed via cell-stratified Mendelian Randomisation (cs-MR) (UC GWAS: 6968 cases and 20,464 controls; single-cell eQTL: 192 postmortem brains). Potential causal associations were estimated using the Inverse-Variance Weighted (IVW) method when multiple instrumental variables were available, and the Wald ratio method when only a single instrumental variable was available for a given cell type–disorder pair, with beta coefficients (β) indicating effect sizes (red for negative, blue for positive associations). Significant associations (Pfdr <0.05, Benjamini-Hochberg FDR correction) are labelled with β and 95% CI and marked with asterisks. Disorders are grouped into psychiatric and neurological categories. Abbreviations: AST, astrocyte; ODC, oligodendrocyte; ExN, excitatory neuron; InN, inhibitory neuron; MIC, microglia; OPC, oligodendrocyte precursor cell; EC, endothelial cell; PC, pericyte.

Fig. 5.

Potential causal links between CD and 14 neuropsychiatric disorders, assessed via cell-stratified Mendelian Randomisation (cs-MR) (CD GWAS: 5956 cases and 14,927 controls; single-cell eQTL: 192 postmortem brains). The heatmap shows effects of CD on disorders, stratified by brain cell type. Potential causal associations were estimated using the Inverse-Variance Weighted (IVW) method when multiple instrumental variables were available, and the Wald ratio method when only a single instrumental variable was available for a given cell type–disorder pair, with beta coefficients (β) indicating effect sizes (red for negative, blue for positive associations). Significant associations (Pfdr <0.05, Benjamini-Hochberg FDR correction) are labelled with β and 95% CI and marked with asterisks. Disorders are grouped into psychiatric and neurological categories. Abbreviations: AST, astrocyte; ODC, oligodendrocyte; ExN, excitatory neuron; InN, inhibitory neuron; MIC, microglia; OPC, oligodendrocyte precursor cell; EC, endothelial cell; PC, pericyte.

Cell-stratified potential causal effects of IBD, UC, and CD on brain IDPs

MR analysis also identified significant cell-type-specific potential causal effects of IBD, UC and CD on various brain IDPs. As shown in Fig. 6a and b, and Supplementary Table S8, In terms of IBD, a negative association was observed with the global volume of the Brain-Stem in oligodendrocytes (β = 0.22, 95% CI [−0.32, −0.13], P < 0.0001 [IVW]), whereas a positive effect was noted on the global volume of the CC-Central (β = 0.20, 95% CI [0.11, 0.29], P < 0.0001 [IVW]). UC was associated with protective effects in astrocyte on the global volume of the CC-Posterior (β = −0.47, 95% CI [−0.63, −0.30], P < 0.0001 [Wald ratio]). Additionally, excitatory neurons displayed a positive relationship with the volume of the Thalamic Nuclei (rh volume PuI) (β = 0.19, 95% CI [0.11, 0.27], P < 0.0001 [IVW]). Oligodendrocyte precursor cell were found to be involved in a reduction in mean diffusivity in the Cingulum cingulate gyrus (L) (β = −0.35, 95% CI [−0.51, −0.19], P < 0.0001 [Wald ratio]). Astrocyte were found to be involved an increase in mean diffusivity in the Tapetum (L) (β = 0.38, 95% CI [0.22, 0.55], P < 0.0001 [Wald ratio]), highlighting complex modifications in white matter integrity. For CD, as shown in Fig. 6a and b, and Supplementary Table S8, oligodendrocytes were positively correlated with the volume of the Thalamic Nuclei (rh volume MV Re) (β = 0.17, 95% CI [0.09, 0.25], P < 0.0001 [IVW]) and the area of the right inferior temporal region (β = 0.17, 95% CI [0.08, 0.25], P < 0.0001 [IVW]). Microglia also showed significant associations with the global volume of the CC-Central (β = 0.21, 95% CI [0.11, 0.32], P < 0.0001 [IVW]).

Fig. 6.

Cell-stratified potential causal effects of IBD, UC, and CD on brain IDPs (IBD GWAS: 12,882 cases and 21,770 controls; UC GWAS: 6968 cases and 20,464 controls; CD GWAS: 5956 cases and 14,927 controls; brain IDPs: 39,691 participants from UK Biobank; single-cell eQTL: 192 postmortem brains) (a) MR results showing the potential causal effects of IBD, UC, and CD on brain IDPs, stratified by cell type. The colours differentiate specific brain regions (e.g., cortical areas, subcortical structures, white matter tracts) and metrics (e.g., thickness, area, volume, mean diffusivity). Significant associations (P < 0.05, derived from two-sample MR analyses using primarily inverse-variance weighted [IVW], or Wald ratio) are labelled with effect sizes (β) and 95% CI. (b) 3D brain renderings highlight affected regions using BrainNet Viewer (cortical), ggseg3d (white matter), and ‘ggseg’ (subcortical), in axial, sagittal, and coronal views. Aparc-Desikan lh thickness paracentral: Orange-coded, represent cortical thickness of left hemisphere paracentral lobule; ThalamNuclei rh volume: Burgundy-coded, represent volume of right hemisphere thalamic nuclei; ThalamNuclei lh volume: Teal-coded, represents the volume of the left hemisphere thalamic nuclei; Aseg global volume CC-Central: Terracotta-coded, represents the total volume of the central segment of the corpus callosum; Aparc-Desikan rh area insula: Purple-coded, represents cortical surface area of the right hemisphere insula; aseg lh volume Lateral-Ventricle: Green-coded, represents the volume of the left hemisphere lateral ventricle; IDP dMRI TBSS MD Tapetum L: Purple-coded, represents mean diffusivity of the left tapetum; IDP dMRI TBSS MD Cingulum cingulate gyrus L: light cyan-coded, represents mean diffusivity of the left cingulum bundle in the cingulate gyrus region; aparc-Desikan rh area inferiortemporal:aparc-Desikan rh area inferiortemporal: Light Green-coded, represents cortical surface area of the right hemisphere inferior temporal gyrus; aseg global volume Brain-Stem: Mustard yellow-coded, represents the total volume of the brainstem measured; aseg global volume CC-Posterior: Yellow-green-coded, represents the total volume of the posterior segment of the corpus callosum: Light green-coded, represents mean diffusivity of the right cingulate gyrus. Abbreviations: IBD, inflammatory bowel disease; UC, ulcerative colitis; CD, Crohn's disease; aparc-Desikan, Desikan-Killiany atlas; aseg, automated subcortical segmentation; dMRI, Diffusion MRI; TBSS, tract-based spatial statistics; MD, Mean diffusivity; LH/RH, left hemisphere/right hemisphere; CC, corpus callosum; ExN, excitatory neuron; ODC, oligodendrocyte; AST, astrocyte; OPC, oligodendrocyte precursor cell; MIC, microglia; InN, inhibitory neuron.

Identifying cell-stratified IBD, UC, and CD → IDP → neuropsychiatric disorder potential causal links through mendelian randomisation and mediation analysis

Leveraging MR and mediation analysis, we uncovered two potential cell-stratified potential causal pathways linking IBD, specifically UC and CD, to neuropsychiatric disorders via brain IDPs. First, UC was associated with alterations in diffusion MRI (dMRI) tract-based spatial statistics (TBSS) mean diffusivity (MD) within the left Tapetum, stratified by oligodendrocytes (Fig. 6a and b). These IDP changes were significantly linked to an elevated risk of schizophrenia (OR = 1.35, 95% CI [1.18, 1.55], P = 0.19 × 10⁻⁴ [IVW]) (Fig. 7a and Supplementary Table S9). Importantly, reverse MR analysis, which tested schizophrenia as the exposure and the IDP as the outcome, provided no evidence of reverse causality, as no sufficient IVs for exposure were identified. As shown in Fig. 7b, mediation analysis indicated that 31.4% of the potential causal effect of UC on schizophrenia was mediated by this IDP, suggesting that white matter microstructural changes, particularly within the left Tapetum, may be a key intermediary in the pathway from UC to schizophrenia.

Fig. 7.

Effects of brain IDPs on six neuropsychiatric disorders identified via MR and mediation analysis (brain IDPs: 39,691 participants from UK Biobank; neuropsychiatric disorders GWAS: varying sample sizes as detailed in Supplementary Table S2). (a) Significant results (P < 0.05) from IVW or Wald ratio analyses, with causal estimates shown as OR values and 95% CIs. The line widths represent the 95% CI. P-values for causal estimates were calculated using inverse-variance weighted (IVW) Mendelian Randomisation for multi-instrument analyses or Wald ratio tests for single-instrument analyses, with significance threshold set at P < 0.05. The colours in panel (a) are used to distinguish the different neuropsychiatric disorders in the forest plot, as follows: orange for ASD (autism spectrum disorder), blue for CA (cerebral aneurysm), green for CTS (carpal tunnel syndrome), purple for MS (multiple sclerosis), orange for OCD (obsessive-compulsive disorder), and yellow for SCP (schizophrenia). Panel (b) shows the mediation analysis for the potential causal pathway from ulcerative colitis to schizophrenia, mediated by the imaging-derived phenotype (IDP) of mean diffusivity (MD) in the left tapetum Panel (c) shows the mediation analysis for the potential causal pathway from Crohn's disease to nonruptured cerebral aneurysm, mediated by the right hemisphere inferior temporal area (from the aparc-Desikan atlas parcellation). Abbreviations: MR, mendelian randomisation; UC, ulcerative colitis; CD, Crohn's disease, IDP, imaging-derived phenotype; dMRI, diffusion magnetic resonance imaging; TBSS, tract-based spatial statistics; MD, mean diffusivity; Tapetum L, left tapetum; aparc, automatic parcellation; Desikan, desikan-killiany; rh: right hemisphere; IVs, instrumental variables. SCP, schizophrenia; MS, multiple sclerosis; OCD, obsessive-compulsive disorder; ASD, autism spectrum disorder; CTS, carpal tunnel syndrome; CA, cerebral aneurysm; OR, odds ratio; CI, confidence interval; IVW, inverse-variance weighted.

Second, CD demonstrated a significant association with structural alterations in the aparc-Desikan right hemisphere inferior temporal region, also stratified by oligodendrocytes (Fig. 6a and b). These changes in the IDP were linked to an increased risk of CA (OR = 2.83, 95% CI [1.32, 6.09], P = 0.77 × 10⁻² [IVW]) (Fig. 7a and Supplementary Table S9). Similar to the UC analysis, reverse MR testing for a reverse causal relationship between neuropsychiatric disorders and this IDP revealed no significant evidence (Supplementary Table S10). Mediation analysis further revealed that 37.5% of the potential causal relationship between CD and cerebral aneurysm was mediated by these cortical changes (Fig. 7c), suggesting that alterations in this brain region may play a pivotal role in the pathway from CD to cerebral aneurysms.

Key genetic drivers in cell-stratified IBD, UC, and CD → IDP → neuropsychiatric disorder potential causal links

The csMR analysis has identified key genetic drivers that contribute to the potential causal pathways linking IBD, UC, and CD with brain IDPs and neuropsychiatric disorders. This analysis is supported by SuSiE colocalisation results, which highlight significant associations between genes and specific cell types involved in these pathways.

One important finding is the gene FADS1, which shows significant colocalisation in oligodendrocytes with a posterior probability of hypothesis 4 (PPH4) value greater than 0.8 (Fig. 8a). FADS1 is implicated in the mediation of IBD's effect on multiple sclerosis, given its role in regulating oligodendrocyte function, which is critical for maintaining white matter integrity, a key factor in multiple sclerosis pathology. Additionally, FADS1 and FDPS display strong colocalisation with SCZ in oligodendrocytes (Fig. 8a), indicating that these genes may mediate IBD-induced changes in brain structure that elevate the risk of developing schizophrenia. In the context of UC, the analysis highlights specific genes that mediate the relationship between UC, IDPs, and neuropsychiatric disorders. For instance, CARD9 and SLC2A10, which colocalise in excitatory neurons (Fig. 8b). Furthermore, PSMB6, showing strong colocalisation in astrocyte, is associated with both oligodendrocytes and schizophrenia (Fig. 8b). For CD, the analysis identifies key genetic drivers of the CD-IDP-neuropsychiatric disorder pathway. Genes such as ENOX1 and FADS2, which colocalise in oligodendrocytes with a PPH4 value exceeding 0.8 (Fig. 8c). In excitatory neurons, ERAP2 and TUT1 show significant colocalisation with CTS (Fig. 8c). Additionally, in oligodendrocytes, genes such as RNFT1 and RBM6 are linked to SCP (Fig. 8c), while genes like CDC42 and ENOX1 exhibit strong colocalisation in these pathways (Fig. 8c).

Fig. 8.

Key colocalised genes in the potential causal pathways linking IBD, brain IDPs, and neuropsychiatric disorders (IBD GWAS: 12,882 cases and 21,770 controls; UC GWAS: 6968 cases and 20,464 controls; CD GWAS: 5956 cases and 14,927 controls; single-cell eQTL: 192 postmortem brains; brain IDPs: 39,691 participants from UK Biobank). (a) Bubble plots showing SuSiE co-localisation -prioritised, cell-type-specific eQTL genes for significant Cell-Stratified Mendelian Randomisation results of IBD. (b) Bubble plots showing SuSiE co-localisation -prioritised, cell-type-specific eQTL genes for significant Cell-Stratified Mendelian Randomisation results of UC. (c) Bubble plots showing SuSiE co-localisation -prioritised, cell-type-specific eQTL genes for significant Cell-Stratified Mendelian Randomisation results of CD. The size of the dot represents the posterior possibility of colocalisation (PPH4). Abbreviations: IBD, inflammatory bowel disease; UC, ulcerative colitis; CD, Crohn's disease; eQTL, expression quantitative trait locus; MR, Mendelian Randomisation; PPH4, posterior probability of hypothesis 4 (co-localisation); AST, astrocyte; ODC, oligodendrocyte; ExN, excitatory neuron; InN, inhibitory neuron; SCP, schizophrenia; MS, multiple sclerosis; OCD, obsessive-compulsive disorder; CTS, carpal tunnel syndrome; ASD, autism spectrum disorder; CA, cerebral aneurysm.

Discussion

Our study reveals that the genetic mechanisms underlying IBD, UC, and CD exhibit cell-type specificity, particularly in oligodendrocytes, astrocytes, excitatory neurons, and inhibitory neurons. While initial Mendelian randomisation (MR) analyses using all genetic variants did not yield significant results, we identified significant cell-stratified potential causal associations between these gastrointestinal disorders and neuropsychiatric conditions through two-sample MR analyses with the partitioned variants as genetic instruments. We also observed significant alterations in IDP that mediate the relationship between IBD, UC, and CD and neuropsychiatric disorders. The integration of SuSiE co-localisation and two-sample MR analyses suggests that specific genetic factors play a critical role in mediating the effects of these gastrointestinal disorders on brain health. Critically, the SuSiE co-localisation analysis and csMR framework prioritised SNPs with direct regulatory effects on brain cells, minimising confounding by gut-mediated pathways. While gut-brain interactions (e.g., microbiota, immune signalling, intestinal cells) may independently influence neuropsychiatric outcomes, our brain-centric design focused on cell-autonomous mechanisms. Nevertheless, integration of gut single-cell eQTL data could further disentangle tissue-specific contributions and extending the current findings.

This study highlights the critical roles of specific brain cell type, particularly oligodendrocytes, astrocyte, excitatory neurons, and inhibitory neurons in the genetic mechanisms underlying IBD, including UC and CD. The cell-type-specific co-localisation of IBD, UC, and CD-associated SNPs and genes with brain gene expression suggests that these genetic variants exert their effects in a cell-type-dependent manner. Previous studies have shown that neuroinflammation, often mediated by astrocytes and microglia, plays a key role in neurodegenerative and psychiatric disorders.^59,60 Our findings extend this idea by revealing how IBD- and CD-associated variants influence gene expression in oligodendrocytes, which are critical for myelination and axonal support. Oligodendrocytes dysregulation has been increasingly linked to neuroinflammatory processes in both gastrointestinal and neuropsychiatric disorders.⁶¹ The prominence of astrocyte in IBD- and UC-associated SNPs further underscores the role of glial cells in mediating brain inflammation. Astrocyte are essential for maintaining brain homoeostasis, regulating the blood–brain barrier, modulating synaptic activity, and responding to inflammation or injury.^62,63 Our results suggest that astrocyte may play a crucial role in modulating neuroinflammatory pathways that influence IBD and related neuropsychiatric outcomes. Taken together, IBD, UC, and CD-associated variants exhibit strong, cell-type-specific effects on brain gene expression, particularly in oligodendrocytes, astrocyte, excitatory neurons, and inhibitory neurons.

Our MR analyses revealed non-significant causal effects of IBD (including UC and CD) on some neuropsychiatric disorders, particularly for schizophrenia and multiple sclerosis. However, in our initial traditional MR analyses (which did not account for cell specificity), the beta values for these potential causal associations, such as IBD with schizophrenia β = 0.04, P = 0.01) and multiple sclerosis (β = 0.10, P = 0.04) were modest in magnitude. This observation aligns with two key considerations: first, the polygenic and multifactorial nature of both gastrointestinal and neuropsychiatric disorders, where individual genetic variants typically exert small cumulative effects; second, the conservative nature of MR methodology, which isolates genetic causality by design and excludes environmental confounders that may inflate effect sizes in observational studies.^1,3,4 Importantly, when we applied Cell-Stratified MR, the effect sizes increased substantially in specific brain cell types (e.g., oligodendrocytes and astrocyte). This divergence suggests that the initial non-significant effects likely reflect dilution across heterogeneous cell populations, masking stronger cell-type-specific relationships. Importantly, even these small causal estimates retain clinical relevance: they highlight therapeutic targets such as oligodendrocytes-mediated myelination in multiple sclerosis and astrocyte -driven neuroinflammation in OCD, advancing mechanistic understanding of the gut-brain axis. Thus, while traditional MR provides robust causal evidence, csMR refines these associations by revealing cell-autonomous pathways that bridge genetic risk to neuropsychiatric outcomes.

The csMR analyses further confirm the potential causal links between IBD, UC, CD, and neuropsychiatric disorders through specific brain cell types. Importantly, higher IBD risk predicted by oligodendrocytes-stratified variants was significantly associated with increased risks of multiple sclerosis and schizophrenia, which supports the emerging evidence that oligodendrocytes play a key role in neuroinflammatory pathways, contributing to the pathogenesis of both diseases.^64,65 A previous study suggested that oligodendrocytes dysfunction could alter myelination and inflammatory responses, and impact the neuropsychiatric outcomes.⁶⁶ We also found that higher IBD risk predicted by astrocyte -stratified variants was associated with an increased risk of OCD, aligning with studies suggesting that astrocytic activation in response to gut inflammation can exacerbate anxiety and compulsive behaviour.⁶⁷ This underscores the importance of astrocytic function in mediating psychiatric symptoms via the gut-brain axis. Interestingly, while UC was significantly associated with increased risks of OCD and schizophrenia through astrocyte -stratified variants, it showed a protective effect against carpal tunnel syndrome. This paradox may indicate that UC-related inflammatory responses exert differential effects on the central versus peripheral nervous system. One possible explanation is that astrocytic activation amplifies neuroinflammation in the brain, thereby worsening psychiatric symptoms, whereas immune modulation in the periphery could mitigate demyelination or nerve compression that underlies carpal tunnel syndrome. Alternatively, this apparent protective effect could arise from unmeasured confounding, horizontal pleiotropy, or differences in case ascertainment between GWAS datasets, which are known sources of bias in Mendelian randomisation^41,48 Lastly, higher CD risk predicted by oligodendrocytes-stratified variants was associated with multiple sclerosis. These findings suggest that oligodendrocytes-associated genetic variants may drive shared inflammatory pathways underlying both Crohn's disease and multiple sclerosis.⁶⁸ Unexpectedly, we identified a potential causal effect of CD-associated inhibitory neurons -stratified variants on non-ruptured CA risk. While prior studies implicate systemic inflammation in vascular pathology,⁶⁹ this specific cell-type mechanism, potentially involving GABAergic signalling in vascular smooth muscle cells, has not been previously reported. Contrary to epidemiological reports,⁷⁰ we found no significant causal links between IBD and AD. This discrepancy may reflect limited power or true biological independence, as IBD-associated neuroinflammation may not directly drive amyloid pathology. Overall, our results highlight the need for targeted research into the roles of oligodendrocytes, astrocytes, and excitatory neurons in the intersection of gastrointestinal disorders and neuropsychiatric outcomes.

We also observed significant alterations in IDP that mediate the relationship between IBD, UC, CD, and neuropsychiatric disorders. Specifically, two potential causal pathways emerged: (1) the influence of UC on mean diffusivity in the tapetum, leading to an increased risk of schizophrenia, and (2) the association of CD with alterations in the inferior temporal area, correlating with a heightened risk of cerebral aneurysm. The pathway from UC to schizophrenia via mean diffusivity changes is particularly striking. Mean diffusivity alterations, which reflect microstructural changes in white matter, have been linked to inflammatory processes.⁷¹ In our findings, 31.4% of the potential causal relationship between UC and schizophrenia is mediated by these IDPs, highlighting how systemic inflammation may affect neural connectivity and contribute to psychiatric outcomes.^72,73 Similarly, the link between CD and cerebral aneurysm through changes in the inferior temporal area suggests that neuroanatomical alterations could predispose individuals to vascular conditions. In this study, 37.5% of the potential causal relationship between CD and cerebral aneurysm is mediated by these changes, illustrating the intricate connection between gut inflammation and brain health. These findings demonstrate that IBD, UC, and CD not only affect gastrointestinal health but also have significant neuroanatomical and neuropsychiatric consequences through specific IDP-related pathways.

However, our mediation analysis relies on several key assumptions that warrant explicit consideration. First, we assume no unmeasured confounding between the exposure (IBD/UC/CD) and mediator (IDPs), or between the mediator and outcome (neuropsychiatric disorders). Second, the analysis presumes no exposure-induced confounding of the mediator-outcome relationship.^74,75 While we mitigated these risks by using genetic instruments and adjusting for covariates in the original GWAS, residual confounding cannot be entirely excluded. For instance, systemic inflammation or shared genetic pathways (e.g., immune dysregulation) might simultaneously influence both brain structure (via IDPs) and neuropsychiatric outcomes, violating the third assumption. Additionally, our model assumes linear relationships between exposure, mediator, and outcome, which may oversimplify nonlinear biological interactions (e.g., threshold effects of inflammation on white matter integrity). Future studies integrating individual-level data with advanced causal frameworks (e.g., sensitivity analyses for unmeasured confounding or nonparametric mediation models) could further validate these pathways.^76,77

The integration of SuSiE co-localisation and two-sample MR analyses identified key genetic drivers mediating the pathways between IBD, UC, CD, IDPs, and neuropsychiatric disorders. Genes such as FADS1 and FDPS in oligodendrocytes were linked to increased risks of multiple sclerosis and schizophrenia. This aligns with the previous understanding that fatty acid metabolism regulated by FADS1 may contribute to neuroinflammation and psychiatric disorders.^5,78 In the context of UC, genes such as CARD9 and SLC2A10 were associated with ASD, while PSMB6 was linked to OCD. CARD9 plays a crucial role in immune response regulation, and its dysregulation could affect neurodevelopment, echoing studies that highlight immune system abnormalities in individuals with ASD.⁷⁹ Similarly, PSMB6, involved in proteasomal degradation, may influence neuronal function and plasticity, contributing to OCD symptoms.⁸⁰ For CD, genes such as ENOX1 and FADS2 were identified as key mediators of multiple sclerosis risk. ENOX1 is implicated in oxidative stress responses, which can lead to neurodegeneration, while FADS2 plays a role in polyunsaturated fatty acid metabolism, essential for neuronal function.⁸¹ Overall, these findings emphasise the importance of specific genetic factors in mediating the relationships between IBD and neuropsychiatric disorders.

The clinical importance of our findings lies in their potential to guide targeted therapeutic strategies and enhance diagnostic precision for patients with comorbid IBD and neuropsychiatric disorders. By pinpointing specific brain cell types, such as oligodendrocytes, astrocyte, and inhibitory neurons, which mediate the genetic interplay between IBD and conditions like multiple sclerosis, schizophrenia, and OCD, our study lays the groundwork for developing cell-type-specific interventions. For example, the significant association of oligodendrocytes-stratified variants with multiple sclerosis suggests that therapies targeting oligodendrocyte function, such as those promoting remyelination, could reduce MS risk in IBD patients.⁸² Similarly, mediation analysis revealed that 31.4% (95% CI [20.5%, 42.3%]) of the potential causal pathway from UC to schizophrenia is mediated by alterations in mean diffusivity in the left tapetum, pointing to white matter integrity as a potential intervention target to mitigate schizophrenia risk.⁵ Furthermore, the odds ratio of 2.83 (95% CI [1.32, 6.09]) linking CD-related structural changes in the right inferior temporal region to CA risk highlights a vascular dimension of IBD that may necessitate increased monitoring of vascular health in affected patients.⁶⁹ These robust association measures and confidence intervals not only deepen our understanding of the gut-brain axis but also pave the way for precision medicine approaches to manage the complex interplay between gastrointestinal and neuropsychiatric conditions.

In summary, this study sheds light on the critical role that specific brain cell types, particularly oligodendrocytes, astrocyte, excitatory neurons, and inhibitory neurons, play in mediating the genetic relationship between IBD, UC, and CD and neuropsychiatric disorders. Our findings may not only facilitate understanding of the molecular mechanisms involved in the gut-brain axis, but also assist in the development of novel targeted therapies for individuals with comorbid gastrointestinal and neuropsychiatric conditions.

Strengths and limitations

This study has several strengths. First, the use of csMR represents a novel approach that integrates single-cell eQTL data with GWAS summary statistics, allowing us to pinpoint cell-type-specific genetic mechanisms underlying the IBD-neuropsychiatric interplay. This methodology enhances causal inference precision beyond traditional MR by accounting for cellular heterogeneity in the brain. Second, the incorporation of mediation analysis with brain IDPs offers a multidimensional perspective, linking genetic risk to structural brain changes and neuropsychiatric outcomes, thus advancing our understanding of the gut-brain axis. Despite these strengths, the study has several limitations. First, all GWAS and single-cell eQTL data were derived from European-ancestry populations, which restricts the generalisability of findings to non-European populations. This is particularly relevant given ancestry-specific genetic effects observed in both IBD and neuropsychiatric disorders, such as the differential penetrance of NOD2 variants in Europeans versus East Asians,¹⁷ and ancestry-dependent heterogeneity in schizophrenia risk loci.⁸³ Second, the resolution of the Bryois et al. single-cell eQTL dataset,¹² which profiles eight broad brain cell types from postmortem prefrontal and temporal cortices, may obscure subtype-specific or regionally nuanced effects. Third, despite rigorous application of MR assumptions, residual horizontal pleiotropy could confound causal inferences; crucially, because many cell-type-specific exposures were instrumented by relatively few independent eQTL variants, sensitivity analyses like MR-Egger regression have low power to detect directional pleiotropy.⁸⁴ Moreover, a key constraint of two-sample MR is the inability to assess the linearity assumption that underlies the regression models relating genetic instruments, exposure and outcome, since summary statistics lack individual-level data required for formal testing of non-linear relationship.^41,46 Furthermore, the mediation analysis using UK Biobank IDPs (N ˜ 40,000) may lack statistical power to detect subtle mediating effects, necessitating validation through larger consortia like ENIGMA (N > 200,000),⁸⁵ to confirm pathways such as tapetum mean diffusivity mediating UC–schizophrenia associations. Some analyses had insufficient power, particularly those involving obsessive-compulsive disorder outcomes, where the reported sample sizes fell short of the calculated thresholds required for 80% power, potentially limiting the reliability of these specific findings despite strong instrumental variables. Finally, MR infers lifelong genetic effects, whereas clinical IBD-neuropsychiatric comorbidities often exhibit episodic manifestations.⁸ Longitudinal studies integrating high-resolution electronic health records could help disentangle acute versus chronic effects of genetic risk. These limitations highlight critical avenues for future research to refine mechanistic insights across diverse populations, cell subtypes, and temporal dynamics.

Contributors

QZ, TG, LZ, and CZ conceived and designed the study. QZ, TG, YW, ZZ, HD, CS, XL, JM, LZ, and CZ analysed the data, with QZ, CS, and YW independently verified the underlying datasets. QZ, TG, LZ, QHZ and CZ wrote the manuscript. All authors critically reviewed and approved the final version. QZ and CZ had full access to all data and validated the integrity of the complete analytical workflow. QZ, TG, and YW contributed equally to this work. The decision to submit the manuscript was made collectively by the corresponding authors CZ, LZ, and QHZ with input from all co-authors.

Data sharing statement

All genetic datasets used in this study are publicly available. Summary statistics for IBD, CD and UC are accessible via the IEU OpenGWAS database (https://gwas.mrcieu.ac.uk/; accession IDs: [ebi-a-GCST003043ebi-a-GCST003045]). The GWAS summary statistics for brain IDPs were obtained from the Oxford Brain Imaging Genetics Server (BIG40, https://open.win.ox.ac.uk/ukbiobank/big40). Single-cell eQTL data from Bryois et al. (PMID: 35915177) are openly available at [https://doi.org/10.5281/zenodo.5543734]. All data are accessible immediately upon publication without restrictions. Analysis code, including scripts for Mendelian randomisation and cell-type stratification, is available on GitHub (https://github.com/ariczhou1989/csMR-IBD) under an MIT licence.

Declaration of interests

The authors declare no competing financial or non-financial interests; all authors completed the ICMJE disclosure form confirming no financial support, affiliations, or intellectual property claims related to this work. This study utilised publicly available, de-identified datasets, requiring no additional ethics approvals.