Genetic dissection of stool frequency implicates vitamin B1 metabolism and other actionable pathways in the modulation of gut motility

Abstract

Background

Genetic studies of stool frequency (SF), an indirect proxy for gastrointestinal transit, may reveal therapeutically tractable pathways relevant to IBS and other dysmotility disorders.

Objective

To identify genes and mechanisms involved in gut motility, providing a foundation for clinical translation.

Design

We performed a multiancestry genome-wide association study (GWAS) meta-analysis of SF in 268 606 European and East Asian individuals. Heritability and genetic correlations with other traits were estimated, and Mendelian randomisation was used to test causal relationships. GWAS signals were fine-mapped and functionally annotated to prioritise candidate genes and pathways. Findings implicating thiamine metabolism were followed-up with dietary interaction analyses in UK Biobank (UKB).

Results

SF heritability was comparable in Europeans (7.0%) and East Asians (5.6%). We observed strong genetic correlations with gastrointestinal and psychiatric disorders (r_g=0.18–0.47), and causal effects on IBS. Novel correlations with cardiovascular traits (r_g=0.12–0.14) were supported by drug signature enrichment analyses. We identified 21 independent loci, including 10 novel signals implicating bile acid synthesis (KLB) and cholinergic signalling (COLQ). Fine-mapping converged on vitamin B1 metabolism, highlighting single-variant causal effects at SLC35F3 (a thiamine transporter) and XPR1 (phosphate exporter essential for thiamine activation). In 98 449 UKB participants, thiamine intake was positively associated with SF (p<0.0001), and a combined SLC35F3/XPR1 genotype score significantly modulated this effect (p<0.0001).

Conclusions

We identify therapeutically tractable mechanisms involved in the control of gut motility, including a previously unrecognised role for vitamin B1. These findings warrant mechanistic and clinical studies to evaluate their translational potential in IBS and other dysmotility syndromes.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Alterations of gut motility are a hallmark of IBS, but the underlying mechanisms remain poorly understood, limiting the development of targeted, mechanism-based interventions.

• Stool frequency can be used as an indirect population-level proxy for gut motility, but its genetic determinants are not fully defined.

WHAT THIS STUDY ADDS

• A multiancestry genome-wide association study meta-analysis showed that stool frequency is partially heritable and shares genetic architecture with gastrointestinal, psychiatric and cardiovascular traits.

• Causal inference analyses demonstrated effects on IBS and implicated actionable pathways including bile acid and cholinergic signalling.

• Fine-mapping identified SLC35F3 and XPR1 as key loci, uncovering a previously unrecognised role for vitamin B1 metabolism in gut motility, and revealed a strong association between dietary vitamin B1 intake and stool frequency in the general population.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• These findings highlight tractable molecular mechanisms controlling gut motility, providing rationale for functional studies and clinical trials.

• They point to thiamine metabolism and other modifiable pathways as potential targets for personalised nutritional or pharmacological interventions in IBS and functional constipation.

• More broadly, they support the use of endophenotypes such as stool frequency for genetic discovery of IBS-relevant mechanisms.

Introduction

GI motility underlies food digestion, nutrient absorption and waste elimination, making it essential for human health and well-being.^{1 2} Its regulation depends on a multifactorial network of communication involving the gut–brain axis (GBA), the immune system and the gut microbiome, and is further influenced by external factors such as diet, physical activity and medications.³ Alterations in the control of motility and peristalsis represent a key pathogenic mechanism in IBS⁴ and other disorders of gut–brain interaction (DGBI), as well as in severe dysmotility conditions such as chronic idiopathic intestinal pseudo-obstruction.^{5 6} Because limited understanding of the underlying aetiological and pathophysiological processes still constrains therapeutic progress,^{1 7} genetic insight into the mechanisms of gut motility may support the development of novel treatment strategies. Indeed, a conceptual framework for studying endophenotypes of GI disease like gut motility has been previously proposed and successfully applied to testing individual genes for their involvement in the modulation of gut motor and sensory functions in patients with IBS and controls.^8–10

Endophenotypes (or intermediate phenotypes) are quantitative, measurable traits that closely reflect the physiological processes disrupted in disease (eg, blood pressure in cardiovascular disease (CVD)). Studying these traits can reduce complexity and increase the likelihood of identifying genes critical to the pathophysiological mechanism(s) underlying a specific condition.¹¹ Colonic transit time, a measure of gut motility highly relevant to IBS, can be directly measured in specialised tertiary centres,^{9 10} but this is not feasible in cohorts large enough for hypothesis-free genome-wide association studies (GWAS). Indirect assessment, however, can be obtained through questionnaire data on bowel habits including stool frequency (SF; number of bowel movements over time), collected as part of large-scale population-based biobank surveys. Although not a perfect proxy for gut motility, SF correlates with colonic transit, possibly more in Asians than Europeans,^{12 13} and enables the study of traits at both ends of the motility spectrum, such as diarrhoea and constipation.

A previous proof-of-concept study from our group demonstrated the utility of this approach via an SF GWAS meta-analysis in five cohorts of European ancestry (EUR), identifying pathways and cell types plausibly involved in the control of peristalsis.¹⁴ In this study, we expand this effort by combining GWAS results from larger datasets, including an East Asian (EAS) biobank.

Methods

A detailed description of the methods used in this study is reported in the online supplemental material, while computational tools, software and statistical approaches are summarised in online supplemental table 1.

Results

SF and characteristics of studied cohorts

SF was studied based on questionnaire data available for a total of 268 606 individuals from five EUR cohorts (N=167 966) and one EAS cohort (N=100 640), including biobanks from the Global Biobank Meta-Analysis Initiative.¹⁵ The demographics of these cohorts are reported in online supplemental table 2. SF ranged from 0.98 to 1.42 bowel movements per day across cohorts, and the prevalence of IBS and associated symptoms followed a U-shaped distribution across SF values, with constipation-predominant (IBS-C) and diarrhoea-predominant (IBS-D) subtypes at the opposite ends of the SF spectrum (online supplemental figure 1).

Extended SF GWAS meta-analysis in Europeans

Individual EUR GWAS were carried out using a common pipeline (online supplemental methods) and combined in a GWAS meta-analysis encompassing 7 879 955 high-quality single-nucleotide polymorphisms (SNPs) with no evidence of population stratification (as determined by linkage-disequilibrium score regression (LDSC); intercept=1.05). Of these SNPs, 3083 were detected at genome-wide significance level (p≤5×10⁻⁸) from 12 independent loci, including two novel associations (figure 1; table 1; online supplemental figure 2). Sex-stratified EUR GWAS meta-analyses did not disclose any novel signal compared with previous studies (online supplemental table 3).

Figure 1. Summary of SF GWAS results. Main findings are reported for ancestry-specific and multiancestry GWAS meta-analyses, together with annotation at respective genome-wide significant loci. From left to right: Manhattan plots of EUR, EAS and multiancestry GWAS, with prioritised candidate genes (or lead SNPs) indicated at each locus (novel loci are also indicated). EA, EAF and beta: effect alleles, their frequency and corresponding genetic effect estimates from EUR (blue) and EAS (orange) GWAS. Fine-mapping results for SF loci, annotated with the individual fine-mapped (causal) variants and their PIP. EAS, East Asian; EAF, effect allele frequency; EUR, European ancestry; GWAS, genome-wide association study; PIP, posterior inclusion probability; SF, stool frequency; SNP, single-nucleotide polymorphism.

Table 1. Stool frequency genome-wide significant loci identified through multiancestry GWAS meta-analysis.

Lead SNP	CHR	BP	EA	OA	EAF (EUR/EAS)	EUR GWAS			EAS GWAS			Multiancestry GWAS		Nearest gene	Causal SNP	PIP	nVars
						P value	BETA	SE	P value	BETA	SE	P value	PHET-ANC
Newly discovered
rs12022782	1	180 869 516	A	G	0.485/0.569	1.2E−06	−0.014	0.003	1.3E−03	−0.028	0.009	4.2E−08	1.1E−01	XPR1	rs12022782	0.741	2
rs12407945	1	234 398 692	T	C	0.179/0.054	4.2E−08	−0.021	0.004	5.4E−01	−0.011	0.018	2.5E−07	6.1E−01	SLC35F3	rs12407945	0.926	1
rs73142301	3	15 459 152	T	C	0.099/0.173	2.4E−06	−0.022	0.005	1.1E−10	−0.071	0.011	1.2E−14	4.6E−05	COLQ	–	–	6
rs71606161	4	39 398 356	C	A	0.812/0.669	4.9E−03	0.010	0.004	2.4E−08	0.050	0.009	3.0E−09	3.3E−05	KLB	–	–	41
rs3839589	6	1 772 218	GTGTAA	G	0.283/0.492	2.8E−08	0.018	0.003	1.0E−02*	0.022*	0.008*	8.9E−08*	5.1E−01*	GMDS	–	–	6
rs9460537	6	20 632 110	T	C	0.166/0.392	9.0E−06	−0.017	0.004	1.5E−04	−0.033	0.009	3.9E−08	8.8E−02	CDKAL1	–	–	62
rs2581260	8	77 493 526	A	T	0.163/0.255	8.4E−04	0.013	0.004	9.0E−08	0.052	0.010	2.3E−09	1.9E−04	ZFHX4-AS1	rs2581260	0.899	11
rs4870765	8	122 259 446	C	G	0.260/0.347	1.9E−06	0.015	0.003	2.7E−05	0.038	0.009	1.8E−09	1.9E−02	–	–	–	6
rs62033403	16	53 822 237	G	A	0.395/0.122	3.1E−03	0.009	0.003	2.1E−07	0.066	0.013	1.5E−08	8.4E−06	FTO	–	–	45
rs292828	19	58 545 133	C	T	0.914/0.459	1.6E−03	−0.016	0.005	2.3E−07	−0.044	0.009	1.1E−08	4.6E−03	ZSCAN1	–	–	17
Previously discovered (ref 14) and confirmed at genome-wide significance (p≤5E−08)
rs11240503	1	205 484 073	A	G	0.301/0.661	5.8E−10	0.019	0.003	4.4E−03	0.027	0.009	8.1E−11	4.3E−01	CDK18	rs11240503	0.552	2
rs10079941	5	122 381 363	A	G	0.648/0.909	2.1E−08	0.016	0.003	8.8E−02	0.026	0.015	3.6E−08	5.3E−01	SNX24	–	–	30
rs12700026	7	2 562 612	C	A	0.110/0.028	3.6E−09	0.027	0.005	3.0E−01	−0.026	0.025	1.6E−08	3.7E−02	LFNG	rs12700026	0.999	1
rs4556017	7	100 632 790	T	C	0.852/-	3.1E−09	0.024	0.004	–	–	–	–	–	MUC12	rs4556017	0.956	1
rs1159724	8	71 999 821	T	A	0.629/0.130	1.0E−07	0.016	0.003	1.5E−03	0.041	0.013	4.6E−09	5.6E−02	–	–	–	182
rs10832353	11	15 091 463	T	C	0.270/0.168	7.7E−09*	0.02*	0.004*	1.2E−10	0.073	0.011	7.8E−16	2.0E−06	CALCB	–	–	29
rs12273363	11	27 744 859	C	T	0.207/0.011	5.9E−23	−0.034	0.004	4.8E−01	−0.028	0.040	5.7E−22	9.3E−01	BDNF	rs12273363	0.552	3
rs56074694	12	66 405 866	A	G	0.337/0.043	6.7E−10	0.018	0.003	6.3E−01	0.010	0.021	4.8E−09	7.25E−01	–	rs56074694	0.999	1
rs4237908	12	98 371 501	A	G	0.448/0.393	6.5E−09	−0.016	0.003	3.9E−04	−0.031	0.009	9.3E−11	1.15E−01	–	–	–	7
rs4767325	12	115 924 093	T	C	0.507/0.122	2.9E−08	−0.016	0.003	1.9E−03	−0.039	0.013	1.7E−09	6.90E−02	–	–	–	45
rs62073098	17	44 316 449	C	A	0.224/0.047	7.8E−12	0.023	0.003	7.1E−01	0.008	0.021	6.4E−11	4.90E−01	ARL17B	–	–	3
Previously discovered (ref 14) and detected at suggestive significance (p≤5E−06)
rs13162291	5	154 369 987	A	G	0.192/0.552	7.9E−07	0.018	0.004	2.6E−01	0.010	0.009	2.7E−06	3.87E−01	KIF4B	–	–	–
rs62482222	7	100 197 866	A	G	0.135/0.104	5.4E−08	0.023	0.004	6.8E−01*	0.006*	0.014*	4.2E−06*	4.60E−01*	FBXO24	–	–	–
rs5757162	22	38 993 370	T	C	0.287/0.389	1.2E−07	0.016	0.003	1.0E+00	0.000	0.009	8.4E−07	7.14E−02	FAM227A	–	–	–

Positive beta (genetic effects) means higher stool frequency. The best candidate gene for each locus is reported. Only causal variants with more than 50% probability are reported. Loci containing genome-wide significant signals are reported in bold.

^*The p value, beta and SE are reported for its best proxy in high linkage disequilibrium. The rs4556017 variant either proxies in LD were not found in the EAS GWAS.

BP, base pairs; CHR, chromosome; EA, effect allele; EAF, effect allele frequency; EAS, East Asian; EUR, European; GWAS, genome-wide association study; LD, linkage disequilibrium; nVars, variants in credible set; OA, other allele; P_HET-ANC, p value for heterogeneity correlated with ancestry; PIP, posterior inclusion probability; SNP, single-nucleotide polymorphism.

The estimated h²_SNP of SF in Europeans was 7%, in line with previous findings.¹⁴ We then assessed genetic correlations (indicative of similar genetic architecture) between SF and 1461 other traits and diseases with GWAS data available, using the Complex Traits Genetics Virtual Lab (CTG-VL) platform (online supplemental methods). A total of 164 traits returned significant correlations (P_FDR<0.05) across GI, cardiovascular, musculoskeletal, neurological and psychiatric domains (figure 2 and online supplemental table 4). Significant genetic correlations were also observed for a wide variety of pain-related traits and their medication proxies (figure 2). Of note, the strongest correlation across medications was detected for chlorphenamine (r_g=0.38), an antihistamine.

Figure 2. Genetic correlations between SF and other traits. Significant (P_FDR<0.05) genetic correlations selected from relevant domains are shown, ranked based on r_g values (plus SE, reported as horizontal line). Positive correlations indicate shared genetic factors contributing to both traits, while negative correlations indicate that genetic predisposition to SF may be associated with reduced risk of the other trait. Full results are reported in online supplemental table 4. CVD, cardiovascular disease; DivD, diverticular disease; FDR, false discovery rate; GERD, gastroesophageal reflux disease; SF, stool frequency.

SF GWAS in East Asians

A total of 6 795 752 high-quality SNPs were included in the EAS SF GWAS, which did not show evidence of population stratification (LDSC intercept=1.04). In this analysis, 173 genome-wide significant SNP associations were detected, mapping to three independent loci: one in common with EUR and previously reported,¹⁴ and two novel and only detected at nominal significance in Europeans (figure 1; table 1).

SF estimated h²_SNP was 5.6% in East Asians, which is slightly lower but not significantly different from Europeans. A high genetic correlation was also observed when comparing GWAS data from EUR and EAS ancestries (r_g=0.88), indicating similar genetic architectures. While EAS-specific GWAS studies are generally under-represented and not available in the CTG-VL platform (online supplemental methods), testing the 164 most correlated EUR GWAS meta-analyses returned significant findings for 30 of these traits also for EAS SF. Among these, top correlations were observed for IBS, GERD, diverticular disease (DivD) and anxiety and depression disorders (online supplemental figure 3).

Multi-ancestry SF GWAS meta-analysis

A multiancestry GWAS meta-analysis (total N=2 68 606) was conducted on 4 583 016 high-quality markers using the Meta-Regression of Multi-Ethnic Genetic Association (MR-MEGA) software to model the genetic effects of EUR and EAS ancestries combined. A total of 479 genome-wide significant markers from 18 independent loci were detected by these means, covering 9/12 signals from the EUR GWAS, all 3 signals detected in East Asians, and 7 additional associations only detected in this combined analysis (figure 1; table 1; online supplemental figures 2 and 4). Hence, in total, 21 GWAS signals were observed in this study, detected either in the EUR, EAS or multiancestry GWAS analyses. Eight of these signals showed heterogeneity of allelic effects correlated with ancestry (P_HET-ANC<0.05), and some markers gave rise to significant associations only in Europeans, possibly due to largely discordant allele frequencies in East Asians (table 1). Overall, 10 novel SF loci were identified via ancestry-specific and multiancestry GWAS meta-analyses (figure 1), almost doubling the number of loci previously reported¹⁴ and highlighting ancestry-specific genetic factors in the determination of SF.

Mendelian randomisation analyses

To assess causative links between SF and other genetically correlated GI disorders, Mendelian randomisation (MR) analyses were performed under different models using independent instrument variables from the current SF GWAS, and DivD, IBS and haemorrhoids (HEM) GWAS performed on non-overlapping population samples (online supplemental methods; figure 3). Adding onto previous MR findings across these conditions,^{16 17} our analyses revealed bidirectional causative effects between SF and DivD. Of note, SF also showed predisposing effects on IBS but not vice versa. Finally, HEM exerted significant negative effects (protection) on the genetic risk of SF, possibly due to its known association with constipation, a trait at the opposite end of the spectrum compared with (increased) SF studied here. Horizontal pleiotropy did not appear to affect these results, as demonstrated by Mendelian randomization pleiotropy residual sum and outlier (MR-PRESSO) analyses (figure 3) and specific Egger horizontal pleiotropy tests (all returning p>0.05).

Figure 3. Bi-directional MR analyses of SF and GI disorders. The causal relationship between SF and selected GI conditions is reported based on MR results in both directions (SF as exposure (left) or outcome (right)), with respective OR and 95% CIs (x-axis) according to different MR methods (y-axis), including MR-PRESSO in order to control for horizontal pleiotropy. Significance levels and the number of IVs used in each analysis are also reported. DivD, diverticular disease; HEM, haemorrhoids; IVs, instrument variables; IVW, inverse-variance weighted; MR, Mendelian randomisation; MR-Egger, Mendelian randomization with Egger regression; MR-PRESSO, Mendelian randomization pleiotropy residual sum and outlier; N.S., not significant; SF, stool frequency.

Gene set enrichment analyses

Positional and expression-quantitative trait loci (eQTL) mapping of EUR, EAS and multiancestry GWAS meta-analyses was performed with the Functional Mapping and Annotation (FUMA) tool to annotate candidate genes for each GWAS signal, resulting in a total of 197 protein-coding genes mapping at the 21 loci. A gene-set enrichment analysis (GSEA, online supplemental methods) confirmed and expanded previous findings in relation to sensory perception, neuropeptide signalling pathway and ion channel activity (online supplemental figure 5; online supplemental table 5).¹⁴ However, additional enriched pathways and mechanisms were identified, possibly due to the 10 additional loci detected here, including regulation of sensory perception of pain, forebrain development and glycosaminoglycan biosynthetic process among Gene Ontology Biological Process terms. Finally, Human Phenotype Ontology screening revealed significant associations with bowel incontinence, schizophrenia and pulmonary conditions, among others.

Fine-mapping and functional annotation of novel SF GWAS loci

Phenome-wide association study (PheWAS) analyses were first carried out to gain biological insight from the 10 novel associations identified in this study (online supplemental methods). Previous GWAS findings from the same regions were detected for seven loci, mostly related to cardiovascular, GI and metabolic traits (online supplemental figure 6 and table 6). Then, in order to identify the most likely causative SNP(s), fine-mapping was performed for EUR and multiancestry specific signals (respectively using FINEMAP and MR-MEGA analytical tools, online supplemental methods). These analyses returned a credible set of SNPs for all loci, three of which were fine-mapped to a single causative variant with high confidence (posterior inclusion probability (PIP) ≥50%) (figure 1; table 1; online supplemental figure 2), namely rs12407945 (PIP=0.93) from the EUR GWAS, and rs2581260 (PIP=0.90) and rs12022782 (PIP=0.74) from the multi-ancestry GWAS.

The marker fine-mapped with highest confidence (PIP=0.93) in the EUR GWAS, rs12407945, is a non-coding SNP strongly associated with eQTLs for a thiamine transporter gene, the solute carrier family 35 member F3 (SLC35F3), influencing its expression in multiple tissues including brain and GI-related tissues (online supplemental figure 7). Another variant mapped with high confidence, rs2581260 (PIP=0.90), has been associated with HEM in previous large GWAS meta-analyses (online supplemental figure 6 and table 6),¹⁸ and maps in a regulatory region within intron 1 of the poorly characterised Zinc Finger Homeobox 4 antisense messenger RNA (mRNA) (ZFHX4-AS1). However, this SNP does not appear to be associated with any eQTLs; hence, the mechanism behind the statistical observation remains elusive. The third variant fine-mapped at high resolution, rs12022782 (PIP=0.74), is a non-coding variant associated with eQTLs in various tissues for four genes (online supplemental figure 7A), including the phosphate exporter SLC53A1 encoded by the XPR1 gene, which shows eQTLs in the oesophagus mucosa (online supplemental figure 7B). Of note, markers from this locus have also been previously associated with systolic blood pressure via GWAS (online supplemental figure 6 and table 6).¹⁹

Several other interesting candidate genes were mapped to loci for which fine-mapping did not result in a single causative variant (figure 1, table 1), including the fibroblast growth factor coreceptor klothoβ (KLB) from the rs71606161 EAS-specific locus, and the collagen like tail subunit of asymmetric acetylcholinesterase (COLQ) gene from locus rs73142301. The lead variant rs71606161 and its linkage disequilibrium (LD) proxies all show eQTLs for KLB in the Genotype-Tissue Expression (GTEx) Portal (online supplemental figure 7A). Indeed, the effect allele (C) associated with higher SF appears to reduce KLB expression, which would lead to increased bile acid synthesis and accelerated colonic transit.¹⁰ Out of three protein-coding genes mapped to the rs73142301 locus, COLQ appears by far the best candidate due to its involvement in cholinergic transmission in the enteric nervous system,²⁰ and the presence of eQTLs in several tissues (online supplemental figure 7A). This locus has also been associated with a higher risk of DivD in previous GWAS analyses (online supplemental figure 6 and table 6).²¹

Testing vitamin B1-SLC35F3/XPR1 interaction in the modulation of SF

The convergence of two fine-mapped SF GWAS loci on genes involved in thiamine metabolism (SLC35F3—a thiamine (vitamin B1) transporter, and XPR1—an exporter of inorganic phosphate essential for converting thiamine into its active form thiamine pyrophosphate (TPP)) prompted us to investigate the vitamin B1 pathway as a potential modulator of gut motility. Analysis of dietary data from 98 449 UK Biobank participants (online supplemental methods) revealed a significant association between thiamine intake and gut motility, with higher thiamine dietary levels correlating with increased SF (p<0.0001; figure 4). Notably, inclusion of SLC35F3 and XPR1 genotype (summarised as a cumulative polygenic score of 0–4 effect alleles) as an interaction term significantly improved the explanatory power of the model (likelihood ratio test: p<0.0001; Δ deviance=12.06; Δ Akaike Information Criterion=−14.9). These findings suggest that the relationship between thiamine intake and SF is modulated by genetic variation in these two genes, with the combined effect alleles exerting the strongest (negative) influence on SF (figure 4).

Figure 4. Correlation between SF and dietary thiamine intake, and effect of SLC35F3/XPR1 genotypes. Average SF is reported across dietary thiamine percentile distribution, together with the generalised linear model prediction curve (black line) and its 95% confidence intervals (grey area). Data are also presented upon stratification for SLC35F3/XPR1 polygenic scores (dotted lines), in order to highlight the influence that effect allele genotypes have on the relationship between SF and thiamine intake. SF, stool frequency.

Drug signature analysis

The translational potential of GWAS findings for drugs discovery was assessed using a two-step computational pipeline translating genetic variation associated with SF into a candidate gene expression signature that can be used to screen public expression databases for compounds able to mimic or oppose such signature (and thus ultimately expected to have stimulatory or inhibitory effects on gut motility) (online supplemental figure 8 and online supplemental methods).²² Application of this pipeline to SF GWAS data resulted in the identification of 831 chemical perturbagens (small molecules inducing gene expression changes in various conditions and cell lines) with significantly correlated or anti-correlated transcriptional profiles in gut and brain-related tissues (online supplemental table 7). Among these were several perturbagens with a known mechanism of action or disease target (N=538), including medications used to treat constipation and diarrhoea (pinaverium bromide and loperamide), psychoactive compounds targeting serotonin (metergoline and urapidil) or adenosine receptors (caffeine), acetylcholinesterase inhibitors used to treat delayed gastric emptying (physostigmine), as well as antihistamines (chlorphenamine and loratadine), thus providing bona fide validation of our computational approach. Remarkably, several drugs administered to treat hypertension and other cardiovascular conditions, including the antihypertensives verapamil, nimodipine and felodipine (calcium channel blockers) and ambrisentan (endothelin receptor antagonist), were identified. Indeed, an enrichment analysis based on the Anatomical-Therapeutic-Chemical drug classification system (online supplemental table 7) revealed calcium channel blockers from the cardiovascular system to be overrepresented among perturbagens (P_FDR=0.023), while protein kinase inhibitors from the antineoplastic and immunomodulating agents were the most enriched (P_FDR=4.88×10⁻¹²) (figure 5). Finally, 293 small molecules without a known target or mechanism of action were also among the significant perturbagens (including top signals from SMR000076922 and SMR000043984), suggesting novel translational opportunities for the modulation of gut motility.

Figure 5. SF drug signature-driven enrichment of ATC therapeutic classes. ATC therapeutic categories are reported, and those significantly enriched based on SF drug signature analyses are highlighted in green (Fisher’s exact test and FDR correction). ATC, Anatomical-Therapeutic-Chemical; FDR, false discovery rate; SF, stool frequency.

Discussion

We report here a multiancestry GWAS meta-analysis of SF in 268 606 EUR and EAS individuals, followed by functional annotation to gain mechanistic insight. By these means, we (1) provide novel data on the genetic architecture of this trait and its similarities with other phenotypes across ancestries, (2) identify three fine-mapped causal variants/genes and seven additional association signals from ten loci not previously reported and (3) highlight actionable pathways and drug targets for potential therapeutic exploitation. Among these, a central novel finding is the identification of vitamin B1 (thiamine) metabolism as a plausible regulator of gut motility.

Two newly implicated loci, fine-mapped to single causative variants affecting SLC35F3 and XPR1 expression, converge on thiamine transport and utilisation, highlighting vitamin B1 metabolism for the first time in relation to its modulatory effects on gut motility. Here we were able to show (1) that dietary intake of thiamine (vitamin B1) positively correlates with SF and (2) that this relationship is modified by genetic variation in SLC35F3, a transporter regulating thiamine intracellular trafficking and availability, and XPR1, an exporter of inorganic phosphate essential for the conversion of vitamin B1 into its active form TPP. The effect alleles from both genes appear to negatively modulate the interaction between dietary thiamine and SF and are associated with eQTLs affecting mRNA expression at both loci. This may therefore impact the intracellular levels of thiamine and phosphate, the components necessary to produce active TPP (ultimately responsible for the biological effects on SF), thus impacting cofactor availability within metabolically competent compartments such as the mitochondria.²³ Future studies can investigate these interactions at a mechanistic level.

The discovery of thiamine as a modulator of gut motility is an important novel finding, which finds only scarce prior in previous scientific literature: in line with our results, one population-based study detected an inverse correlation between dietary intake of thiamine and chronic constipation.²⁴ At the same time, thiamine deficiency has been linked to GI disorders, in particular the so-called GI beriberi characterised by impaired gastric function and gut dysmotility.²⁵ While thiamine is mainly obtained from the diet, future research may explore whether targeted nutritional interventions, such as thiamine supplementation, can alleviate disordered gut motility and IBS symptoms in genetically susceptible individuals, thereby supporting a personalised approach to disease management. Moreover, solute carriers like SLC35F3 and SLC53A1 represent ideal therapeutic targets, as evidenced by current and novel drugs in clinical trials that inhibit, enhance or circumvent transporter function to treat a wide range of diseases.²⁶ Of note, fedratinib, a kinase inhibitor used to treat myelofibrosis, was found to inhibit another thiamine transporter (SLC19A2) and cause Wernicke’s encephalopathy, another form of thiamine deficiency also characterised by constipation.²⁷ This compound anticorrelated with SF in our drug signature analyses, demonstrating the effectiveness of our computational strategy to discover novel therapeutic opportunities.

Strong candidate genes were also identified from the analysis of non-fine-mapped loci, including KLB and COLQ. Although not fine-mapped as the causative variant (possibly due to high LD across multiple associated SNPs in the region), the C allele from the rs71606161 locus is associated with eQTLs in multiple tissues and a reduced expression of the KLB gene coding for Klothoβ, a protein with a critical role in bile acid synthesis.²⁸ Of note, and in line with this finding, genetic variation resulting in decreased Klothoβ protein stability (rs17618244G, Arg728 allele) had already been associated with effects on gut motility measured as accelerated colonic transit in patients with IBS-D.¹⁰ Indeed, at the mechanistic level, reduced Klothoβ expression or function in hepatocytes negatively affects binding of fibroblast growth factor 19 (FGF19) to the corresponding FGF receptor, ultimately leading to an increased synthesis of bile acids and stimulation of colonic peristalsis.^28–30 COLQ is expressed in the colonic myenteric plexus, where it anchors acetylcholinesterase to the neuromuscular junction and contributes to acetylcholine receptor organisation, thereby shaping cholinergic neurotransmission that drives peristalsis.^{20 31 32} This pathway has repeatedly emerged in genetic studies of stool frequency and IBS, including fine-mapped signals affecting acetylcholinesterase (ACHE gene) expression (ref ¹⁴ and this study) and muscarinic-receptor-related pathways (adenylyl cyclase 2, ADCY2 gene).³³ In the context of current findings, vitamin B1 metabolism adds another layer since TPP is an essential cofactor for pyruvate dehydrogenase that produces acetyl-coenzyme A, the immediate substrate for acetylcholine synthesis in enteric neurons.³⁴ Reduced thiamine availability or impaired thiamine utilisation can therefore decrease acetylcholine production and weaken cholinergic tone. Altogether, genetic associations involving COLQ, cholinergic signalling and thiamine pathways point toward a unified mechanism in which altered enteric acetylcholine availability influences peristalsis and contributes to interindividual variation in gut motility and stool frequency.

In total, we report the identification of 21 individual association signals (table 1), nearly doubling the previous tally of SF loci.¹⁴ Leveraging current SF GWAS data and molecular insights from the literature, several genes from these loci have been fine-mapped and/or highlighted as the most likely causative candidates. Based on their known function(s), these genes are implicated in physiological pathways relevant to gut motility, with some converging onto shared underlying biological mechanisms (summarised in table 2).

Table 2. Genes identified in SF GWAS meta-analyses and their relevance to gut motility.

GWAS locus (tag SNP)	Gene symbol	Gene name	Gene function and potential mechanisms involved
Neurotransmission and synaptic regulation
rs73142301	COLQ	Collagen Like Tail Subunit Of Asymmetric Acetylcholinesterase	Anchors acetylcholinesterase at neuromuscular junctions and cholinergic synapses; variants may alter acetylcholine clearance, affecting smooth-muscle contraction and peristalsis.
rs11240503*	NFASC†	Neurofascin	Cell adhesion protein involved in axon guidance and formation of nodes of Ranvier; disruption may impair enteric neuronal conduction and coordination of motility patterns.
rs4556017*	ACHE‡	Acetylcholinesterase	Degrades acetylcholine at cholinergic synapses; altered expression or activity may shift excitatory neuromuscular signalling and change motility tone.
rs10832353	CALCB	Calcitonin Related Polypeptide Beta	Encodes β-CGRP, a neuropeptide involved in nociception and sensory motor reflexes; modulation of CGRP signalling may influence gut–brain communication and smooth muscle relaxation.
rs12273363*	BDNF	Brain Derived Neurotrophic Factor	Neurotrophin involved in neuronal growth, differentiation, survival and plasticity; alterations of its neurotrophic and neurotransmitter effects in the enteric nervous system may affect motility coordination.
rs62073098	CRHR1	Corticotropin Releasing Hormone Receptor 1	Mediates corticotropin-releasing hormone stress responses in the enteric nervous system and smooth muscles; activation enhances motility and visceral hypersensitivity.
Bile acids metabolism
rs71606161	KLB	KlothoB	Fibroblast growth factor coreceptor, binds to FGF19 contributing to regulation of bile acids synthesis; impaired signalling may disrupt bile-acid composition and thereby alter motility and colonic transit.
Epithelial and enterocyte function
rs11240503*	CDK18†	Cyclin Dependent Kinase 18	Expressed in BEST4+ enterocytes involved in pH sensing and electrolyte transport; altered activity may contribute to enterocyte and epithelial barrier dysfunction, affecting gut transit.
rs4556017*	MUC12‡	Mucin 12	Membrane-associated mucin contributing to barrier function, microbial interactions and mucosal hydration; variation may alter mucus thickness and integrity, impacting stool consistency and transit time.
rs12700026*	LFNG	Beta-1,3-N-Acetylglucosaminyltransferase Lunatic Fringe	Notch-modifying glycosyltransferase that regulates epithelial differentiation; disruption may shift secretory/absorptive lineage balance and affect motility through altered fluid handling.
Vitamin B1 metabolism
rs12022782*	XPR1	Xenotropic And Polytropic Retrovirus Receptor 1	Exporter of inorganic phosphate essential for the conversion of vitamin B1 into TPP; changes may affect production of thiamine derivatives, with downstream effects on cholinergic signalling, affecting gut motility.
rs12407945*	SLC35F3	Solute Carrier Family 35 Member F3	Transporter regulating thiamine intracellular trafficking and availability; reduced thiamine uptake will decrease TPP conversion, with downstream effects on cholinergic signalling.

^*Locus fine-mapped at single-variant resolution, corresponding to stronger evidence for a causative role of associated genes/mechanisms.

^†NFASC and CDK18 represent equally good candidates (associated with eQTLs) from the same fine-mapped locus rs11240503.

^‡ACHE and MUC12 represent equally good candidates (associated with eQTLs) from the same fine-mapped locus rs4556017.

CGRP, calcitonin gene-related peptide; eQTL, expression-quantitative trait loci; FGF, fibroblast growth factor; GWAS, genome-wide association study; SF, stool frequency; SNP, single-nucleotide polymorphism; TPP, thiamine pyrophosphate.

At a broader level, and as reported in a previous survey,¹⁴ SF shows modest SNP heritability, estimated here around 7% from the analysis of multiple cohorts of individuals of EUR descent. This appears to be consistent across ancestries, as a similar estimate (5.6%) was obtained for EAS. Of note, this degree of heritability is not different from that observed for IBS, which ranges between 5.8% and 10.3% depending on various definitions,^{33 35 36} hence further supporting the adoption of such (heritable) endophenotypes for genetic research in DGBI and dysmotility syndromes. Indeed, even the number of association signals detected in the current GWAS meta-analysis (n=21) largely exceeds the number of IBS risk loci (n=6) identified in the largest GWAS meta-analysis including 53 400 patients and 433 201 controls.³⁵

Despite environmental and genetic differences between EUR and EAS cohorts, the genetic architecture of SF was nearly identical across ancestries (r_g=0.88) and highly similar to other GI conditions, particularly DivD and IBS. Overlapping GWAS signals and PheWAS annotations further linked SF with DivD and HEM, consistent with dysmotility as a shared hallmark.^{37 38} MR, which tests whether genetic susceptibility to one trait (exposure) also predisposes to another (outcome), indicated that DivD and HEM causally influence SF but in opposite directions: DivD increased, whereas HEM decreased SF. This aligns with GWAS evidence that enteric neurons and interstitial cells of Cajal, key regulators of peristalsis, mediate risk in both conditions,^{1 3 18 39} likely driving diarrhoea (higher SF) in DivD and constipation (lower SF, a recognised risk factor) in HEM. Conversely, when testing SF as exposure, MR results suggested causal effects on DivD and, most importantly, IBS. Genetic variation influencing SF thus also predisposes to IBS, underscoring the value of SF as an endophenotype for IBS gene discovery and providing a rationale for further investigating SF-related perturbagens and their molecular targets in IBS.

As previously shown,^{14 40 41} genetic correlations, GSEA and PheWAS analyses highlighted SF similarities with mood/anxiety disorders and other psychiatric conditions, as well as the involvement of intestinal sensory and neuronal functions via neuropeptide and ion channel signalling. This is not surprising, as these pathways are key to the communication along the GBA that is central to psychiatric comorbidities often observed in patients with IBS and other dysmotility syndromes.^{2 42 43} Strong evidence also supports genetic overlap between SF and cardiovascular traits. Genetic correlations revealed shared architecture with blood pressure, hypertension and angina, and several SF loci had previously been linked to cardiovascular phenotypes. Similar findings were reported for IBS defined according to consensus Rome III Criteria, particularly IBS-C, which shows significant genetic correlations with hypertension and heart disease.³³ Epidemiological studies further support this connection, as constipation and reduced defecation frequency are associated with higher risk of coronary artery disease, ischaemic heart disease, CVD mortality and major adverse cardiac events.^44–46

Our drug–gene expression analyses reinforced this link, with SF transcriptional signatures most closely resembling profiles of drugs used for hypertension and other CVDs (eg, verapamil, ambrisentan), and enrichment for calcium channel blockers. Histamine signalling also emerged from the data, with antihistamines such as chlorphenamine and loratadine showing strong genetic correlation and/or expression-based association with SF. While both ion channel function and histamine-related pathways have been previously linked to gut motility, IBS and visceral sensitivity,^47–52 these findings validate our drug–gene expression analysis approach and underscore that SF-correlated and anticorrelated drug perturbagens without known molecular targets offer promising leads for future investigation.

Finally, some limitations should be acknowledged: (1) to ensure adequate sample size for genetic analyses, we relied on SF derived from questionnaire data, which is only an indirect proxy for gut motility; (2) we did not include dietary, medication or lifestyle variables due to inconsistent availability across cohorts and to avoid significant loss of statistical power; (3) several SF-associated loci still lack unequivocal identification of causative gene(s), and experimental validation of proposed pathophysiological mechanisms is pending; (4) the therapeutic potential of most SF-associated perturbagens requires identification of their molecular targets and further evaluation of druggable pathways in preclinical and clinical settings. These aspects can be addressed in future studies and should stimulate further investigation building on the novel findings reported here.

In conclusion, this study demonstrates the value of quantitative endophenotypes such as SF for dissecting the genetic architecture of GI motility. We confirm and expand previous evidence implicating neuropsychiatric and cardiovascular pathways, neuropeptide and ion channel signalling and bile acid biosynthesis in the regulation of intestinal transit. Importantly, we uncover a previously unrecognised role for vitamin B1 (thiamine) in gut motility, highlighting a biologically plausible and modifiable axis with potential for nutritional or pharmacological intervention. Together, these findings provide a strong rationale for investigating and repurposing compounds that target genetically implicated, tractable pathways, warranting further mechanistic and clinical studies, particularly in the context of IBS and other dysmotility syndromes.

Data availability statement

Data are available in a public, open access repository.