Interstitial cystitis: a phenotype and rare variant exome sequencing study

Summary

Background

Interstitial cystitis/bladder pain syndrome (IC/BPS) is a poorly understood and underdiagnosed syndrome of chronic bladder/pelvic pain with urinary frequency and urgency. Conflicting data exist regarding associated phenotypes, and little is known of its genetic aetiology.

Methods

We conducted a retrospective case–control analysis of electronic medical record (EMR) and retrospective case–control analysis of exome sequencing (ES) to identify phenotypes related to and rare variant risk factors for IC/BPS. Retrospective EMR data were collected from a national network of research sites in eMERGE-III. ES data were acquired from two research cohorts of individuals with IC/BPS and family members in addition to unaffected individuals stored at Columbia University Irving Medical Center. We determined odds ratios and P values for phenotypes associated with IC/BPS in eMERGE, odds ratios and P values for genes and gene-sets enriched for rare damaging variants in individuals with IC/BPS, and P values indicating overlap of genes enriched with rare damaging variants in individuals with IC/BPS with canonical gene sets.

Findings

Using the eMERGE data, we compared 193 individuals with IC/BPS to 99,482 individuals without and confirmed known phenotypic associations such as gastroesophageal reflux disease (odds ratio [OR] 2.3, P = 1.6 × 10⁻⁵, logistic regression) and irritable bowel syndrome (OR 8.5, P = 3.3 × 10⁻²¹, logistic regression). Notable associations, including anaphylactic shock (OR 8.8, P = 1.3 × 10⁻¹¹, logistic regression), intervertebral disc disorders (OR 2.6, P = 2.6 × 10⁻⁶, logistic regression), and laxity of ligament or hypermobility syndrome (OR 15.3, P = 2.0 × 10⁻⁵, logistic regression), were detected. In an exome ultra-rare variant gene-level collapsing analysis with 348 IC/BPS and 11,627 controls, no gene association reached study-wide statistical significance. Gene-set analyses extended the previously reported association with ATP2C1 and ATP2A2 (OR, 7.4; 95% CI, 1.6–26.4; P_adjusted = 0.033, Cochran–Mantel–Haenszel test [CMH]), implicated in Mendelian desquamating skin disorders but did not provide evidence for other proposed pathogenic pathways. Pathway analysis detected associations with “anaphase-promoting complex-dependent catabolic process” (P = 6.2 × 10⁻⁶, Fisher's exact tests [FET]), the “regulation of MAPK cascade” (P = 5.5 × 10⁻⁶, FET) and “integrin binding” (P = 1.1 × 10⁻⁹, FET).

Interpretation

Individuals with IC/BPS should be screened for related phenotypes, and previously unreported phenotypic associations with IC/BPS. Perturbations in biological networks for epithelial integrity and cell cycle progression in IC/BPS pathogenesis provide a roadmap for its future investigation.

Funding

This publication was supported by the Centers for Disease Control and Prevention of the U.S. Department of Health and Human Services (HHS) as part of a financial assistance award totaling $2,400,000 with 60 percent funding from the CDC/HHS (U01DP006634-01). The contents are those of the author(s) and do not necessarily represent the official views of, nor are they endorsed by, the CDC/HHS or the U.S. Government (CAB). National Institutes of Health grant 1K08HG012374 (JEM). New York-Presbyterian Samberg Scholar (JEM). Thrasher Early Career Research Award (JEM). NIH/NIDDK CAIRIBU Interactions Core U24-DK-127726 (CAB). NICHD Boston Children’s Hospital Intellectual and Developmental Disabilities Research Center Molecular Genetics Core Facility U54HD090255. NIDDK George M. O'Brien Urology Cooperative Research Centers Program U54DK104309 (AGG). The eMERGE Network was initiated and funded by NHGRI through the following grants: Phase III: U01HG8657 (Group Health Cooperative/University of Washington); U01HG8685 (Brigham and Women's Hospital); U01HG8672 (Vanderbilt University Medical Center); U01HG8666 (Cincinnati Children's Hospital Medical Center); U01HG6379 (Mayo Clinic); U01HG8679 (Geisinger Clinic); U01HG8680 (Columbia University Health Sciences); U01HG8684 (Children's Hospital of Philadelphia); U01HG8673 (Northwestern University); U01HG8701 (Vanderbilt University Medical Center serving as the Coordinating Center); U01HG8676 (Partners Healthcare/Broad Institute); and U01HG8664 (Baylor College of Medicine).

Research in context.

Evidence before this study

Interstitial cystitis/bladder pain syndrome (IC/BPS) is a chronic condition characterised by pelvic pain and urinary symptoms, with an unclear aetiology and no definitive biomarkers. Prior literature has suggested associations between IC/BPS and other conditions such as irritable bowel syndrome (IBS), anxiety, and gastroesophageal reflux disease (GERD). Genetically, a few Mendelian disorders and candidate gene studies have proposed potential contributors, including calcium transporter genes like ATP2A2 and ATP2C1. However, there has been limited large-scale, unbiased genomic investigation into rare variant contributions, comorbid phenotypes, and pathogenic pathways.

We considered published articles from 1990 until 2025 indexed in PubMed and Google Scholar as sources for this study. No study was excluded as IC/BPS is a heterogeneous disease and hypotheses that disagree with others may be proven accurate for a subset of individuals with IC/BPS. Start date of the research for this manuscript was July 30, 2003 and the end date was August 6, 2025. Search terms used were “interstitial cystitis” and “bladder pain syndrome”.

Added value of this study

This study represents an integrated analysis of phenotype and exome sequencing data in IC/BPS. Using phenome-wide analysis of over 100,000 individuals, it confirms known associations (e.g., GERD, IBS) and identifies additional comorbidities such as anaphylactic shock and intervertebral disc disorders. Rare variant exome sequencing in 348 IC/BPS cases versus nearly 12,000 controls affirms the role of ATP2A2 and ATP2C1 and uncovers pathways enriched for damaging variants, including cell cycle regulation (anaphase-promoting complex-dependent catabolic process), MAPK cascade regulation, and integrin-mediated signalling. The use of pathway-based analyses (GSEA and network-based heterogeneity clustering) enabled signal detection despite locus heterogeneity and limited sample size. Gene burden, family-based and trio analyses provided candidate genes, including PIEZO2 and DYX1C1 (DNAAF4). A relationship between Hunner lesions and the HLA-DQ2.5 haplotype was suggested.

Implications of all the available evidence

The findings reinforce the conceptual model that IC/BPS is a genetically heterogeneous disorder involving epithelial dysfunction, cell cycle dysregulation, and abnormal sensory processing. These results suggest that patients with IC/BPS may benefit from evaluation for comorbid epithelial and inflammatory conditions and motivate future work to functionally characterise implicated genes and pathways. The study also highlights the utility of integrating comprehensive phenotyping with rare variant sequencing and systems biology approaches to illuminate complex syndromes like IC/BPS. This evidence may support future biomarker development and targeted therapeutic strategies for affected individuals.

Introduction

Interstitial cystitis/bladder pain syndrome (IC/BPS) is a disorder of chronic pain including symptoms of urinary frequency, urgency, and bladder, suprapubic, and lower back discomfort or pain.^1,2 Prevalence estimates in the United States suggest 2.7%–6.5% of adult women and 1.9%–4.2% of men have IC/BPS.^1,3 However, IC/BPS is likely underdiagnosed due to the lack of biomarkers or known aetiology.³

The pathophysiology of IC/BPS is incompletely understood.⁴ The characteristic Hunner lesion is neither sensitive nor specific.⁵ Genetic factors contribute, evidenced by higher concordance rates among monozygotic twins and first-degree relatives.^6,7 Our group has previously identified Mendelian disorders within an IC/BPS cohort.⁸ Phenotype associations exist between IC/BPS and nociplastic pain conditions (e.g., irritable bowel syndrome), psychiatric disorders, and other medical comorbidities.9, 10, 11 We hypothesise that investigating clinical databases and larger cohorts of exome sequencing data will reveal additional associated phenotypes and rare variant risk factors.

Methods

Study design

We conducted a retrospective case–control study of phenotypic information from the Electronic Medical Records and Genomics (eMERGE-III) network and of exome sequencing data from two research cohorts of individuals with interstitial cystitis/bladder pain syndrome (IC/BPS). The study followed the STREGA reporting guideline for genetic association studies.

Ethics

Written informed consent from all participants was obtained. The BCH cohort was enrolled following informed consent under an IRB-approved protocol at Boston Children's Hospital (04-11-160). The MaGIC samples were received anonymised under an IRB-approved protocol at Columbia University Irving Medical Center (AAAS7948).

Participants, phenotyping and phenotype analysis for eMERGE-III

Phenotype data were obtained through eMERGE-III.¹² The eMERGE III network comprises multiple academic medical centres and integrated health systems across the United States, encompassing diverse geographic regions, healthcare delivery models, and demographics. The included cohorts therefore reflect a broad cross-section of individuals engaged in clinical care within large biobank-linked electronic health record (EHR) systems. However, as these participants were enrolled primarily through institutional biobanks and not through population-based sampling, they are best considered clinically ascertained rather than fully representative of the general U.S. population.

eMERGE diagnoses were harmonised by converting ICD-10-CM codes to ICD-9-CM, the dominant coding system in eMERGE-III; ICD-10's granularity complicates reverse mapping.^13,14 The dataset included 102,739 individuals and 19,411 ICD-9 codes. Five codes (×999, NoD.x, NoI9, IMO0001, IMO0002) and individuals with unavailable or ambiguous sex were excluded. We performed two analyses to identify conditions co-morbid with IC/BPS (Supplementary Tables S1–S6). In the “primary” analysis, individuals with ICD-9 code 595.1 (“chronic interstitial cystitis”) entered at least twice were classified as individuals with IC/BPS (Fig. 1, Supplementary Tables S1–S3). In the “secondary” analysis, individuals with ICD-9 code 595.1 coded at least once were classified as individuals with IC/BPS (Supplementary Tables S4–S6). In both analyses, individuals without ICD-9 code 595.1 were classified as controls. ICD-9 codes were mapped to 1817 unique phecodes using mapCodesToPhecodes and addPhecodeInfo in the R PheWAS package.^14,15 ICD-9 code 595.1 was assigned phecode 592.13 (“chronic interstitial cystitis”). In the primary analysis, other phecodes were derived using the createPhenotypes function in the R PheWAS library using default settings including add.phecode.exclusions = T but modified min.code.count = 2 so that a minimum of two ICD-9 codes from the individual was required to be counted. In the secondary analysis, we modified min.code.count = 10 so that a minimum of 10 ICD-9 codes from the individual was required.

Fig. 1.

Phenotype analysis using eMERGE. “Primary” comorbidity analysis identifying phecodes in individuals with IC/BPS in eMERGE compared 193 individuals with IC/BPS to 99,482 individuals without IC/BPS (see methods). The association of having IC/BPS and carrying each of 1816 phecodes was assessed. Statistical association was tested with logistic regression (see Methods). Blue horizontal line indicates 0.05, the red horizontal line indicates 0.05/1816. Phecodes with P values < 2.8 × 10⁻⁵ are annotated. Full data available in Supplementary Table S1. Labels and data points for non-specific phecodes beginning in “Other”, Symptoms”, or “Functional” have been removed to improve readability. Some labels were excluded for clarity. See Methods for “primary” analysis description.

Participants, phenotyping, and informed consent for exome sequencing analysis

Two cohorts with IC/BPS were included for exome sequencing (Supplementary Figure S1). (A) The “BCH cohort” (n = 406) was previously described.⁸ Probands (n = 319) had an IC/BPS diagnosis from a U.S. or Canadian physician, with urinary urgency and pelvic/suprapubic/abdominal pain for ≥3 months in a 6-month period. Exclusion criteria included urinary tract infection in the prior 3 months, urinary tract abnormalities, or genitourinary cancers. Cases could be sporadic or familial. Comorbidities (i.e., Hunner lesions, anxiety, migraines) were extracted from medical records and questionnaires. Sex and gender identity was obtained from the electronic medical record and from self-report (Table 1). The remaining non-proband individuals (n = 87) included affected and unaffected family members. Of the 319 probands, 250 had exome sequencing which passed quality control and relatedness filtering (Supplementary Figure S1, see Methods). (B) The “MaGIC cohort” included DNA from 100 randomly selected female unrelated individuals from The Maryland Genetics of Interstitial Cystitis Study (MaGIC).¹⁶ Inclusion required age ≥13, IC/BPS diagnosis, ≥1 affected relative, living in the US or Canada and availability for contact. Exclusion criteria included Parkinson's disease, Multiple Sclerosis, Spina bifida, or another neurologic disease predating IC/BPS symptoms. Phenotypes (anxiety, irritable bowel syndrome [IBS], migraine) and sex were extracted from sample metadata. Gender identity was not collected.¹⁶ Samples were received anonymised and covered under IRB-approved protocol at Columbia University Irving Medical Center (AAAS7948). Of the 100 probands, 98 had exome sequencing which passed quality control and relatedness filtering (Supplementary Figure S1, see Methods).

Table 1.

Demographics of individuals in the BCH and MaGIC cohorts.

Geographic ancestry	BCH cohort			MaGIC cohort
	Total (% of population)	Females	Males	Total (100% females)
Latino	6 (2.4%)	6 (2.6%)	0 (0%)	3 (3.1%)
European	221 (88.4%)	202 (89%)	19 (82.6%)	64 (65.3%)
East Asian	2 (0.8%)	2 (0.9%)	0 (0%)	0 (0%)
Middle Eastern	13 (5.2%)	11 (4.8%)	2 (8.7%)	6 (6.1%)
Admixed	8 (3.2%)	6 (2.6%)	2 (8.7%)	25 (25.5%)
Total	250	227	23	98
Age of diagnosis	BCH cohort	Females	Males	MaGIC cohort
Min, 1st quartile, median, mean, 3rd quartile, max	8.0, 30.0, 43.0, 41.8, 52.25, 75	10.0, 30.0, 43.0, 41.9, 52.75, 75	8.0, 26.25, 39.0, 39.6, 48.0, 74	Unavailable

Demographic characteristics for probands in the BCH cohort (n = 250) and the MaGIC cohort (n = 98). Sex was obtained from the electronic medical record (eMERGE and BCH) self-report (BCH) or study metadata (MaGIC). Genetic ancestry was inferred from principal components and assigned using the Louvain clustering framework (see Methods). Age-of-diagnosis was available for 204 individuals in BCH (190 female) and unavailable for the MaGIC cohort. Values represent counts, percentages, or summary statistics as indicated.

Though some families had multiple affected members, only well-phenotyped probands were included in gene-level analyses. Combining the MaGIC and BCH cohorts created the “combined cohort” included 348 probands (98 MaGIC, 250 BCH) (Table 1). In the combined cohort of 348 analysed individuals (325 female), 43 individuals were diagnosed with anxiety (1 in BCH), 88 individuals with migraine (40 in BCH), 6 with Hunner lesions (6 in BCH), and 55 with IBS (0 in BCH).

Seven families had >1 affected and ≥1 unaffected member and analysed to identify familial genes/variants (see Family Enrichment Analysis). 13 trios with unaffected parents were used to identify de novo or damaging compound heterozygous genotypes (see De Novo Analysis).

The included BCH and MaGIC cohorts represent some of the largest well-phenotyped IC/BPS cohorts to date; however, they are not fully representative of the broader IC/BPS population. Both cohorts are enriched for individuals who have access to tertiary referral centres and/or are engaged with advocacy networks, which may bias toward individuals with more severe disease or greater healthcare access. In addition, ancestry in both cohorts is predominantly European, and socioeconomic diversity is limited relative to the U.S. population.

Exomes from 11,627 individuals (6573 females) from unrelated studies at CUIMC served as controls broadly described as “control”, “control with mild neuropsychiatric disease”, and “healthy family member.” No additional phenotyping data were available.

Next-generation sequencing data generation, harmonised alignment and variant calling

Exome sequencing (ES) of the “BCH cohort” was described previously.⁸ BCH cohort samples were sequenced at the Broad Institute (Cambridge, MA) using AgilentV6 capture kit; MaGIC cohort samples at Psomagen Inc. (Rockville, MD) using IDTERPv2 capture kit.

All case data were transferred to the Institute for Genomic Medicine (IGM) at CUIMC. Controls were either sequenced at or transferred to IGM. Control sequencing included 10,405 ES and 1222 genome sequencing (GS) samples. Multiple capture kits were used with 150 bp paired-end reads. Possible bias introduced by using ES from different capture kits (Supplementary Table S7) and GS in collapsing is corrected during coverage harmonisation.17, 18, 19

Harmonised alignment and variant calling

Individuals with IC/BPS (cases) and without (controls) were processed with the same bioinformatic pipeline for alignment and variant calling which has been used at the IGM to generate ES or GS for more than 120,000 individuals.20, 21, 22 IGM's analysis tool for annotated variants (ATAV)²³ platform was used to add Genome Aggregation Database (gnomAD) v.2.1 frequencies²⁴ and Transcript-inferred Pathogenicity (TraP) score.²⁵ The collapsing workflow utilises gene symbol matching on consensus coding sequence (CCDS release 20).²⁶ 18,766 CCDS v20 genes were used in the collapsing analysis.

The Transcript-inferred Pathogenicity (TraP) score is a method to identify pathogenic non-coding variants (specifically synonymous and intronic variants) in genic regions by evaluating their potential to damage a gene's transcripts. The TraP model is a random forest model which evaluates 20 selected features to predict changes in splicing or transcription. The model is trained on a set of pathogenic synonymous variants and benign variants (de novo mutations from healthy individuals). The final TraP score ranges between zero and one (0–1) and represents the fraction of decision trees that classify a variant as pathogenic. Below 0.459 is enriched for benign variants, ≥0.459 and < 0.93 is possibly damaging, ≥0.93 is considered probably damaging.

Sample Quality Control (QC) and Removal of Related Individuals

Cases and controls passed identical quality control (Supplementary Figure S1). Inclusion required ≥90% of CCDS release 20 covered at ≥ 10×, ≤2% contamination (VerifyBamID²⁷), and ≥85%/80% overlap with dbSNP for SNVs/indels.²⁸ Samples with discordance between self-declared and sequence-derived sex were excluded to prevent phenotype-genotype mismatch. Related individuals (second-degree or closer) were filtered using KING, prioritising cases and better-covered samples.²⁹ The BCH cohort included 406 data samples. Of these, 381 completed the bioinformatic pipeline. Of these, 300 were identified as individuals who had been enrolled as probands into studies (see eMethods “Proband versus family-based analyses”). 2 were excluded for sex mismatch suggesting sample swap, 3 were excluded for unclear metadata, 1 was excluded for high contamination. After removing additional related or duplicate samples, 250 samples remained. For the MaGIC cohort, 100 data samples completed the bioinformatic pipeline. 2 were excluded for unclear metadata leaving 98 samples.

Combined, 348 (325 female) samples (250 from BCH, 98 from MaGIC) were compared to 11,627 (6573 female) unaffected individuals for a rare variant gene, gene-set, pathway, and exome wide association study (ExWAS) analyses.

Collapsing analysis

Collapsing analysis tested associations between IC/BPS and rare variant carrier status at the gene or gene-set level (Fig. 2, Supplementary Figures S3–S5, Supplementary Tables S8–S15).17, 18, 19^,30 Each model defines “qualifying variants” (QVs) which are considered to have an equivalent effect using criteria such as external/internal allele frequency, variant effect, and in silico predictions (Supplementary Table S8).¹⁷ Steps included: i) select high-quality unrelated cases and controls enrolled broadly as “healthy family members”, “control mild neuropsychiatric disease” or “control”, ii) cluster sample by geographic ancestry, iii) harmonise coverage within cluster, iv) define QV criteria, v) assign 0/1 indicator for QV present in the gene, or gene-set, vi) test association with CMH test and vii) visualise results. This approach aggregates rare variants of the same effect by gene or gene-set while controlling for ancestry.

Fig. 2.

Ultra-Rare Damaging Missense and Loss-of-Function Model with Unrelated Probands. The quantile–quantile plots for the protein-coding genes with at least one case or control carrier of an ultra-rare predicted loss-of-function or predicted damaging missense variant. All variants are ultra-rare (i.e., allele frequency of less than 0.05% in internal case and control by cluster, absent in external reference cohorts, allele frequency <0.002% in IGM database). P values were generated from the exact two-sided Cochran-Mantel-Haenszel (CMH) test by gene by cluster to indicate association between carrier status and case/control status. No gene achieved study-wide significance P < 5.4 × 10⁻⁷ after Bonferroni correction indicated by dashed line. Top ten case enriched genes are labelled. The green lines represent the empiric 95% confidence interval created by permutations (n = 1000). Red indicates case-enriched while blue indicates control enriched. Included effects = stop-gain, frameshift, splice donor, splice acceptor, missense. All missense variants REVEL > 0.5. 331 individuals with IC/BPS compared to 6516 controls. Full data available in Supplementary Table S11.

Clustering

Clustering was performed as previously described.^18,19,30 Principal component analysis (PCA) on predefined variants was used to capture population substructure.31, 32, 33, 34 The Louvain method identified clusters from the first six PCA components.^30,35 Geographic ancestry was assigned using a pre-trained neural-network with generated probability estimates (European, African, Latino, East Asian, South Asian and Middle Eastern). A 95% probability cut-off was used for the ancestry groups, and “Admixed” samples were those that did not reach 95% for any of group (Supplementary Figure S2A). The assigned ancestries were used only to visually inspect using the Uniform Manifold Approximation and Projection (UMAP) the overlap between predicted ancestries and clusters created by the Louvain method.36, 37, 38

Clusters with ≥15 cases and ≥15 controls were used (Figs. 2 and 3, Supplementary Figures S2–S5, Supplementary Tables S9–S15). Cluster sizes ≥20 or ≥30 have been utilised previously.^18,19,30 We utilised 15 cases to include cluster 5 with 18 individuals with IC/BPS (Supplementary Figure S2). Of 348 IC/BPS individuals, 331 (309 female) were included in analysed clusters. Of 11,627 controls, 6516 (3424 female) were included in analysed clusters. Genomic inflation (Figs. 2 and 3, Supplementary Figures S3–S5) were assessed. A European ancestry cohort was formed from clusters 0, 2, and 6, including 293 individuals with IC/BPS (276 female) and 3392 controls (1804 female) for network-based heterogeneity analysis (Table 2, Supplementary Table S16, Supplementary Figures S6 and S7).

Fig. 3.

Rare variant exome wide association study. The quantile–quantile plots for variants in protein-coding genes with at least one case or control carrier. All variants had an allele count of at least 10 in eligible bases after coverage harmonization. P values were generated from the exact two-sided Cochran-Mantel-Haenszel (CMH) test by gene by cluster to indicate a different carrier status of cases in comparison to controls. No variant achieved study-wide significance P < 5.4 × 10⁻⁷ after Bonferroni correction indicated by dashed line. Top ten case enriched variants are labelled. Red indicates case-enriched while blue indicates control enriched. The green lines represent the empiric 95% confidence interval created by permutations (n = 1000). 331 individuals with IC/BPS compared to 6516 controls. Full data available in Supplementary Table S18.

Table 2.

Gene clusters and pathway enrichment detected by Network-based Heterogeneity Clustering (NHC).

Genes	Cluster P value	GO pathway
ANAPC1 ANAPC7 CDC20 CDC23 FOXO1 NEK2 PPARGC1A SKIL SMAD3 TRIM33	8.6e-08	anaphase-promoting complex-dependent catabolic process (6.2e-06)
EGF ERBB2 ERBB4 FGF10 FGFR1 FGFR4 IGF1R NRG2 PIK3R3	0.00043	peptidyl-tyrosine modification (1.2e-07) positive regulation of MAPK cascade (5.5e-06) transmembrane receptor protein tyrosine kinase activity (3.2e-07) Transmembrane receptor protein kinase activity (1.2e-06)
BCAR1 FLNC ITGA2 ITGA6 ITGA9 ITGB1 ITGB3 LAMA5 TLN2	0.0042	integrin-mediated signalling pathway (2.0e-10) cell adhesion mediated by integrin (4.2e-06) Integrin binding (1.1e-09)

Damaging ultra-rare variant model used as input where sample–gene combination included if gene harboured a variant in the damaging ultra-rare model (Fig. 2) for both individuals with and without IC/BPS. Table includes enriched gene clusters (P < 0.05) which are associated with GO pathways detected by NHC. This analysis was performed on a single cluster of European geographic descent with 293 individuals with IC/BPS and 3392 individuals without. Full data in Supplementary Table S16.

Coverage harmonisation

To reduce bias from unequal coverage (e.g., ES versus GS, capture kits), we applied coverage harmonisation by cluster (Supplementary Figure S2B). All samples passed QC for ≥90% of CCDS release 20 covered at > 10×. Only genomic sites with >10× coverage were considered. Sites were removed if the absolute case–control difference in ≥10× coverage exceeded 7%. For gene and gene-set collapsing, only sites ≥80% of cases and ≥80% of controls >10× coverage were included. For ExWAS, the threshold was ≥90% coverage in both groups.

Qualifying variant definition

Each collapsing model is defined by a “qualifying variant” (QV) (Supplementary Table S8). All variants meeting QV definition are considered equivalent.^17,39 External frequency filters in gnomAD could be either “ultra-rare” (absent from public databases and allele frequency <0.002% in the >120,000 samples in the IGM) or “flex” (MAF <0.1%). For the “flex” models, allele frequencies were filtered at a population specific level using gnomAD exomes. For gnomAD exomes, populations included afr, amr, asj, eas, sas, fin, and nfe. For gnomAD genomes, the MAF filter was applied to the full population. For ultra-rare models, variants were excluded with an internal allele frequency >0.05% applied to the combined case–control call set by cluster after excluding one allele.⁴⁰ For flex models, variants were excluded with an internal allele frequency >1% applied to the combined case–control call set by cluster after excluding one allele.^19,30,41, 42, 43

We define two allele frequency bands for gene-based collapsing. “Ultra-rare” indicates absent from gnomAD and allele frequency <0.002% in the entirety (>120,000 samples) of the database for the Institute for Genomic Medicine. “Flex” indicates allele frequency <0.1% in gnomAD Genome (global) and Exome (population specific, see Methods). We define one allele frequency band for ExWAS (see Methods). “Rare” indicates allele frequency <1% in gnomAD Genome (global) and Exome (population specific, see Methods).

Variants with proportion expression across transcripts (pext) ≤ 50% of the gene's max pext were removed in the ultra-rare and flex models as they are less likely to affect translated mRNA.⁴⁰ All predicted loss-of-function (LOF) variants (stop gain, frameshift, splice acceptor, and splice donor variants) were filtered with Loss-Of-Function Transcript Effect Estimator (LOFTEE) to remove likely false-positives.²⁴

Gene-sets

We tested whether QVs were enriched among cases in curated gene-sets derived from known pathways, phenotypes, and expression profiles (Supplementary Tables S14 and S15).¹⁷ Gene-sets were selected a priori based on established literature and biologically plausible mechanisms relevant to IC/BPS:

• 1.Genes previously implicated in Mendelian diagnoses among individuals with IC/BPS: ATP2C1 (MIM: 604384), DCAF8 (MIM: 615820), SIX5 (MIM: 600963), ENAM (MIM: 606585), ATP2A2 (MIM: 108740). We tested the hypothesis that the prior association of IC/BPS with select Mendelian disease would remain.⁸
• 2.Congenital anomalies of the kidney and urinary tract (CAKUT): 41 genes from the Invitae CAKUT panel (Test code: 434,341).⁴⁴ We tested the hypothesis that IC/BPS is associated with subclinical changes in genitourinary anatomy.⁸
• 3.Epithelial structure maintenance: 31 genes from GO:0010669 excluding MUC2 (MIM: 158370). We tested the hypothesis that IC/BPS is a disorder of epithelial maintenance dysfunction because of its prior association with disease of epithelial dysfunction.⁸
• 4.Morphogenesis of an epithelial sheet: 59 genes from GO: 0002011 excluding CLASP2 (MIM: 605853), MIR221 (MIM: 300568). We tested the hypothesis that IC/BPS is a disorder of epithelial maintenance dysfunction because of its prior association with disease of epithelial dysfunction.⁸
• 5.Desquamating dermatologic disorders: 79 genes from GeneDx dermatologic disorders panels (https://www.genedx.com/tests/Dermatology/dermatologic-disorders∼c9693). Diseases included Darier Disease (MIM: 124200), Hailey–Hailey Disease (MIM: 169600), Epidermolysis Bullosa (Dystrophic) (MIM: 131850, 131750, 226600), Pachyonychia Congenita type 1 (MIM: 167200), Epidermolysis Bullosa Simplex (MIM: PS131760), Epidermolytic Ichthyosis (MIM: PS607602, PS113800), Superficial Epidermolytic Ichthyosis (MIM: PS607602, PS113800), Bullous Ichthyosiform Erythroderma (MIM: PS607602, PS113800), Pachyonychia Congenita type 2 (MIM: PS167200), Epidermolytic palmoplantar keratoderma (MIM: PS144200), Epidermolysis Bullosa Junctional Type (MIM: PS226650), Generalised Atrophic Benign Epidermolysis Bullosa (GABEB), Herlitz Junctional Epidermolysis Bullosa, Mitis Junctional Epidermolysis Bullosa, Non-Herlitz Junctional Epidermolysis Bullosa, and Focal Dermal Hypoplasia (MIM; 305600). We tested the hypothesis that IC/BPS was associated with genes related to desquamating dermatologic disorders because of its prior association with Hailey–Hailey and Darier disease.⁸
• 6.GO chemical and genetic perturbations pathways involving both ATP2C1 and ATP2A2. Gene-sets include BLALOCK_ALZHEIMERS_DISEASE_DN (1237 genes), DIAZ_CHRONIC_MYELOGENOUS_LEUKEMIA_UP (1390 genes), ENK_UV_RESPONSE_KERATINOCYTE_DN (480 genes), LU_EZH2_TARGETS_DN (378 genes), MULLIGHAN_MLL_SIGNATURE_2_DN (274 genes). Gene-sets were extracted from the Molecular Signatures Database (MSigDB) using msigdbr.45, 46, 47 We tested the hypothesis that the association with epithelial dysfunction and desquamating dermatologic disorders are driven primarily by ATP2C1 and ATP2A2 which are the causative genes Hailey–Hailey and Darier disease.⁸
• 7.Bladder expression: 2257 genes expressed more than 50 transcripts per kilobase million (TPM) in the bladder per The Genotype-Tissue Expression (GTEx) database.⁴⁸ We tested the hypothesis that IC/BPS was related to genes expressed in the bladder.
• 8.Bladder development: 22 genes curated from literature.49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63 We tested the hypothesis that IC/BPS was related to genes expressed during bladder development.
• 9.Pain intensity: 44 genes identified by genome wide association study (GWAS).⁶⁴ Given the association of IC/BPS with nociceptive disorders, we tested the hypothesis that IC/BPS was related to genes previously associated with pain.
• 10.Major depression: 43 genes identified by GWAS.⁶⁵ Given the association of IC/BPS with psychiatric disorders, we tested the hypothesis that IC/BPS was related to genes previously associated with depression and anxiety.9, 10, 11
• 11.Anxiety: 11 genes identified by GWAS.⁶⁶
• 12.Irritable bowel syndrome (IBS): 7 genes identified by GWAS.⁶⁷ Given the association of IC/BPS with IBS, we tested the hypothesis that IC/BPS was related to genes previously associated with IBS.⁹
• 13.4 genes associated with IC/BPS in recent animal models.68, 69, 70, 71 We tested the hypothesis that IC/BPS is associated with genes implicated in animal models.
• 14.Osteoarthrosis/osteoarthritis: 51 genes identified by GWAS.⁷² We identified osteoarthrosis as a co-morbid condition with IC/BPS (see Results). We tested the hypothesis that the two conditions shared genetic architecture.
• 15.8 genes identified as harbouring de novo variants or rare compound heterozygous pairings in 13 probands (see Results). We tested the hypothesis that these genes carried ultra-rare but variants in other individuals with IC/BPS.

For gene-set #1, we also performed subgroup analyses for individuals with IC/BPS and with IBS, with anxiety, with migraines and with any of the 3 co-morbidities.

Gene-set enrichment and pathway analyses

We utilised network-based heterogeneity clustering (NHC, updated February, 2024) to identify gene clusters linked by biological and protein–protein interactions (PPIs).^73,74 NHC detects physiological homogeneity within a genetically heterogeneous dataset by testing case/control enrichment of gene and protein networks. NHC then performs pathway enrichment analyses (Supplementary Table S16). For NHC, we used a gene list per individual from the single European ancestry cluster combining clusters 0, 2 and 6 (Supplementary Figure S2B, see eMethods “Coverage Harmonisation”) harbouring an ultra-rare damaging variant (Fig. 2, Table 2, Supplementary Tables S11 and S16, Supplementary Figures S6 and S7, see eMethods “Ancestry clustering”). NHC was performed with parameters: --edge 0.99 --hub 100 --merge 0.5 --mode 2 --network Y --boost N. Cluster enrichment threshold was P < 0.01 (logistic regression) using three ancestry principal components via NHC; gene-set enrichment used FDR <1 × 10⁻⁵ (Fisher's exact tests [FET]) via NHC. NHC tested enrichment for gene-sets/pathways from MSigDB Hallmark,⁴⁷ KEGG (805), Reactome (1670), WikiPathway (791), GO Biological Process (7375), and GO Molecular Function (1753).

We used fast gene-set enrichment analysis (fGSEA, R package fgsea) to identify biological pathways (Supplementary Figures S8–S11) enriched in 459 genes with case enrichment (P < 0.1, CMH) from the ultra-rare damaging collapsing model (Fig. 2). Enrichment in hallmark gene-sets from Molecular Signature Database (MSigDB) were reported if the pathway was significantly associated (P < 0.05, fGSEA) after multiple comparisons adjustment by fGSEA.^47,75,76

Exome wide association study (ExWAS)

All tested variants were “rare” (MAF <1%) in both gnomAD genome and exome and no effect filtering was applied (Supplementary Table S8). Enrichment of variants in cases was tested using the CMH test (Fig. 3, Supplementary Tables S17 and S18).^19,30,41, 42, 43 Variants with allele count <10 were excluded. At a cluster level, the internal allele frequency was allowed to exceed 1% only by one allele.^19,30,41, 42, 43 Analysis method was identical to collapsing (see “Collapsing Analysis” in methods) but variant (not gene or gene-set) 0/1 status was based on absence/presence of the variant as opposed to any QV within a gene or gene-set (step v).

Family enrichment analysis

Qualifying variants (QVs) were heterozygous variants in all affected and absent from all unaffected family members (Supplementary Tables S19 and S20). We identified gene-variant effect pairs (Supplementary Table S20) observed in ≥2 families, where QVs had consistent effects (e.g., missense, LOF). QVs required: QUAL >30, GQ ≥ 20, minimum 3 alternate reads, and gnomAD MAF <1%.

De novo analysis

Thirteen probands had unaffected parents available for trio analysis (Supplementary Table S21), performed on the Codified Genomics platform (San Diego, CA). Genes harbouring de novo variants contributed to gene-set analysis for ultra-rare damaging collapsing analyses (Supplementary Tables S14 and S15).

Variants were phased with parental sequences to identify de novo and compound heterozygous variants. Variants were filtered for genotype quality (GQ > 20) and read depth (>10×). Combined Annotation Dependent Depletion (CADD) was used for in-silico predictions of the functional effect of missense mutations and SpliceAI for presumed splice variants. ‘Qualifying variants’ were defined as those that satisfied each of the following criteria: i) de-novo ii) allele frequency <10 in gnomAD for de novo variants and <0.02 for compound heterozygous variants (http://gnomad.broadinstitute.org/), iii) missense or loss of function or frameshift or affecting a splice-site and iv) if missense, in silico predicted damaging by SIFT,⁷⁷ Polyphen-2 or a CADD score > 10.77, 78, 79 American College of Medical Genetics and Genomics (ACMG) guidelines were followed to quantify the pathogenicity of qualifying variants.⁸⁰

Copy number variant detection from exome sequencing

Copy number variants (CNVs) were identified from exome data using the eXome-Hidden Markov Model (XHMM) following the tutorial provided at https://zzz.bwh.harvard.edu/xhmm/tutorial.shtml.⁸¹ We filtered CNV calls for genomic disorders as previously described (Supplementary Tables S22 and S23) using R (v.4.3.1).⁸² The analysis was run in batches for two different target interval files: IDTERPv1 and Agilent V6. Identified CNVs were annotated using AnnotSV version 3.4.2.⁸³

HLA typing and associations with Hunner lesions

Class I and II HLA typing was performed using DRAGEN V4.2.7 HLA caller. 138 unique alleles were identified across HLA genes A, B, C, DQA1, DQB1, and DRB1 (Supplementary Table S24). There were 6 individuals with IC/BPS and Hunner lesions. Three were in each of European ancestry clusters 0 and 2, which included 119 and 62 non-Hunner individuals with IC/BPS, respectively.

Proband versus family-based analyses

We performed two family-based analyses: (A) a family-based analysis to examine variants present in affected family members but absent in unaffected family members (Supplementary Tables S19 and S20) and (B) a de novo analysis in which we examined variants absent from either parent (Supplementary Table S21) which included affected probands and unaffected parents. For all other analyses, any individual related to the proband was removed (second-degree or closer, see “Sample Quality Control (QC) and Removal of Related Individuals” in Methods). This prevents spurious associations for rare inherited variants. By including only individuals who were primarily enrolled in IC/BPS studies (as opposed to “affected” family members), we included the most well-phenotyped individuals. This might prevent inclusion of an individual with a non-diagnosed condition other than IC/BPS which explained the individual's symptoms. This approach would bias towards null in these analyses by excluding potentially affected individuals.

Statistics

All statistical analyses were performed in R version 4.3.1.⁸⁴ Sample size was not determined a priori; all available and eligible samples from eMERGE, BCH and MaGIC cohorts were included. Formal power calculations were not performed as the study utilised all available, previously sequenced IC/BPS samples from two established cohorts; sample size was thus determined by cohort availability rather than a priori power targets. This approach is standard for exploratory rare variant analyses, where available sample size is substantially constrained by disease prevalence. No randomisation or blinding was performed because this was an observational study using existing clinical and sequencing data, and no experimental intervention was applied. Inclusion and exclusion criteria were prespecified and applied uniformly.

Statistical associations between IC/BPS status and phecodes in eMERGE were determined in the following manner. Logistic regression tested associations using the R PheWAS library between phecodes and chronic IC/BPS, adjusting for sex, age, site, total ICD-9 codes, and nine ancestry principal components (Supplementary Table S1). The library uses predefined “control” groups for each phecode. PheWAS plots were created with phewasManhattan function (Fig. 1). A Bonferroni-corrected statistical significance threshold was applied at 2.8 × 10⁻⁵ (0.05/1816 phecodes) (Fig. 1, Supplementary Tables S1 and S4). There are possible unaccounted for confounders in the eMERGE analysis which were not available to us for statistical correction including socioeconomic status (e.g., insurance type, median income from residence, education level), other elements of healthcare utilisation (e.g., number of visits, length of time of EHR follow-up), and traits not fully captured by phecodes (e.g., smoking status, body mass index, or alcohol or other substance use) among many others.

To evaluate differences in eMERGE between individuals with IC/BPS and without, we conducted statistical tests across covariates. Binary sex differences were assessed using the chi-square test of independence. Student's t-test was used to compare age with results reported as mean ± standard deviation. ICD count variable (a proxy for healthcare utilisation or comorbidity burden) differences were assessed using the Wilcoxon rank-sum test, with results summarised as medians and interquartile ranges. Additionally, we evaluated whether the distribution of cases and controls varied significantly across participating recruitment sites using a chi-square test of independence. For each test, two-sided P values were reported. Results were summarised in tabular form for “primary” (Supplementary Tables S2 and S3) and “secondary” (Supplementary Tables S5 and S6) eMERGE analyses.

Statistical associations between IC/BPS status and individual genes (or gene-sets) were determined in the following manner. For each gene (or gene-set), we counted the number of cases and controls with a QV within the gene (or within any gene in the gene-set) per ancestry cluster (Supplementary Figure S2B) and applied a two-sided CMH test to assess QV association with case/control status while controlling for ancestry cluster membership.^19,30,41,43

Statistical associations between IC/BPS status and individual variants (ExWAS) were determined in the following manner. For each variant in the call set, we counted the number of cases and controls with the variant per ancestry cluster (Supplementary Figure S2B) and applied a two-sided CMH test to assess variant association with case/control status while controlling for ancestry cluster membership, For gene and variant association, we controlled multiple comparison testing using a Bonferroni-adjusted threshold of P < 5.4 × 10⁻⁷ (0.05/[18,766 CCDS genes × 2 non-synonymous models + 55,607 variant sites in ExWAS]). Our intent was to control the study-wise family-wise error rate (FWER) across all primary inferential claims we make—whether at the gene level (collapsing) or the single-variant level (ExWAS). Because we interpret discoveries from either analysis as equally primary, we pre-specified a single family of hypotheses comprising both sets of tests and used a Bonferroni threshold of α study-wise = 0.05/(18,766 genes × 2 models + 55,607 variants) = 5.37 × 10⁻⁷. Any correlation among tests makes the choice of Bonferroni conservative and biases towards the null hypothesis.

Treating gene-level and single-variant analyses as two separate families is also reasonable in some contexts (e.g., when they are framed as distinct secondary aims). If one adopts that framing, the gene-level threshold would be 0.05/(18,766 × 2) = 1.33 × 10⁻⁶ and the variant-level threshold 0.05/55,607 = 8.99 × 10⁻⁷. Our combined threshold (5.37 × 10⁻⁷) is more stringent than either of these, so our approach errs on the side of reducing false positives across the study's primary results.

For gene-sets, we controlled for multiple comparison testing by applying a false discovery rate (FDR) adjustment to the gene-set P values. Specifically, we used R's stats::p.adjust with method = ‘fdr’ (Benjamini–Hochberg procedure) on the vector of unadjusted P values.

To evaluate the distribution of statistical associations of genes and variants, we created quantile–quantile (QQ) plots.¹⁷ The synonymous model was used as a negative control (Supplementary Figures S3 and S4). Each QQ plot compared observed versus expected P values for all genes or variants in a model. Empirical (permutation-based) expectations were generated by permuting case/control labels within clusters while maintaining the gene-by-sample (or variant-by-sample) matrix.^19,30 For each permutation, P values were calculated using the CMH test, preserving cluster membership. This was repeated 1000 times per model. For each permutation the P values were ordered. The mean of each rank-ordered estimate across the 1000 permutations (i.e., the average 1st order statistic, the average 2nd order statistic, etc.) represented the empirical estimates of the expected ordered P values. The genomic inflation factor λ based on the permutation-based expected P values was estimated using regression.^85,86

Statistical associations between IC/BPS status and pathways were determined using NHC and fGSEA. (A) NHC first identifies gene clusters from gene harbouring a QV (defined by ultra-rare damaging variants) among cases (see Network-based Heterogeneity Clustering [NHC] in eMethods). Once the clusters are identified, the associations of IC/BPS status with harbouring a QV in the cluster is tested using logistic regression adjusting for three ancestry PCs. Gene clusters with PC-adjusted case–control P values < 0.01 were considered significantly enriched. Once gene clusters are identified, enrichment of curated pathways was tested. For each gene cluster, we tested enrichment of pathways from KEGG and Reactome using Fisher's exact test (FET), declaring significance at P < 1 × 10⁻⁵. We similarly tested enrichment for Gene Ontology biological process (8992 terms) and molecular function (2812 terms) categories using FET with the same P value < 1 × 10⁻⁵ threshold. (B) fGSEA generated P values for gene-set enrichment using the preranked fGSEA framework. Briefly, genes were ranked by P values from the ultra-rare damaging variants model (Fig. 2), and an enrichment score was computed for each pathway based on the distribution of its genes along this ranked list. The null distribution of the ES was estimated empirically by repeatedly sampling random gene-sets of the same size with permutations shared across pathways, and the nominal P value was calculated as the proportion of random gene-sets with an ES at least as extreme as the observed ES. Nominal enrichment P values from fGSEA were adjusted for multiple testing across all pathways using the Benjamini–Hochberg procedure to control the false discovery rate.

P values from fGSEA could vary by run due to its permutation-based method. To assess robustness to fGSEA's stochastic permutations, we quantified run-to-run variability by repeating the pathway analysis 1000 times with identical inputs and settings. For each run, we collected adjusted P values across all eligible pathways and summarised their distributions (Supplementary Figure S9). We also recorded, per run, the top-ranked pathway by adjusted P and evaluated how often “Transport of small molecules” ranked first; under a null in which each of the 23 eligible pathways is equally likely to be top, we computed a two-sided binomial probability. As a negative control with less significantly associated genes, we repeated the same 1000-run procedure on a matched-size gene-set (n = 463) from the ultra-rare damaging collapsing model defined by case enrichment (odds ratio >1) and P > 0.22 (CMH), which yielded 35 eligible Reactome pathways; we summarised adjusted P distributions and minimum observed adjusted P across runs (Supplementary Figures S10 and S11).

Statistical associations between HLA typing and Hunner lesion status among individuals with IC/BPS was tested in the following manner. Logistic regression was used to test associations between each HLA allele and Hunner lesion status, adjusting for sex and six ancestry principal components. Each allele was tested independently using the R glm () function with a binomial family. The log-odds estimate and standard error were extracted, and the odds ratio (OR) and corresponding 95% confidence interval (CI) were calculated using the exponential of the estimate and ±1.96 times its standard error.

Role of funders

The funders had no role in study design, data collection, data analysis, interpretation, or writing of the report.

Results

Phenome-wide co-morbidity analysis

We performed a phenome-wide comorbidity analysis across 1816 phecodes in the Electronic Medical Records and Genomics (eMERGE) network dataset^12,14 (Fig. 1, Supplementary Tables S1–S3). Our primary analysis (case inclusion required coding at least twice for IC/BPS) compared 193 individuals (177 female) with IC/BPS to 99,482 individuals (53,357 female) without IC/BPS. We validated prior comorbidities, including dysuria (odds ratio [OR] 22.5, P = 6.8 × 10⁻³², logistic regression),^1,2 polyuria (OR 38.0, P = 5.5 × 10⁻⁴³, logistic regression),^1,2 irritable bowel syndrome (OR 8.5, P = 3.3 × 10⁻²¹, logistic regression),⁸⁷ oesophagitis (OR 2.3, P = 1.8 × 10⁻⁵, logistic regression),¹⁰ gastroesophageal reflux disease (OR 2.3, P = 1.6 × 10⁻⁵, logistic regression),¹⁰ and allergic rhinitis (OR 3.5, P = 1.7 × 10⁻¹⁰, logistic regression).⁸⁸ Notable associations included anaphylactic shock (OR 8.8, P = 1.3 × 10⁻¹¹, logistic regression),⁸⁹ intervertebral disc disorders (OR 2.6, P = 2.6 × 10⁻⁶, logistic regression),⁹⁰ and laxity of ligament or hypermobility syndrome (OR 15.3, P = 2.0 × 10⁻⁵, logistic regression).⁹¹ Prior phenotypic associations such as chronic pain (OR 3.4, P = 0.00024, logistic regression)⁹ and anxiety (OR 1.9, P = 0.0033, logistic regression),⁴ among others, were suggestive. In our secondary analysis, we included more cases (requiring coding for IC/BPS at least once) but were more conservative in identifying comorbidities (requiring coding at least 10 times) among cases and controls (Supplementary Tables S4–S6); we compared 360 individuals (316 female) with IC/BPS to 99,482 individuals (53,224 female) without IC/BPS. Previously unreported associations were identified such as Barrett's esophagus (OR 11, P = 1.9 × 10⁻⁵, logistic regression) and osteoarthrosis/osteoarthritis (OR 3.2, P = 1.1 × 10⁻⁵, logistic regression). For both analyses, we report comparisons between cases and controls for the covariates and eMERGE recruitment sites (Supplementary Tables S2 and S3, S5 and S6).

Gene-based collapsing

We first explored gene-based collapsing to identify associations of case/control status with individual genes (Supplementary Table S8).^17,18,30 We created models with both ultra-rare (absent from control databases) and rare (minor allele frequency [MAF] < 1%) qualifying variants. We created models for damaging effects (damaging missense and predicted loss-of-function) and synonymous (presumed control). All models showed limited genomic inflation (ʎ < 1.1) indicating ancestry sub-structure matching (Fig. 2, Supplementary Figures S3–S5, Supplementary Tables S9–S12).

In our ultra-rare and rare damaging models, no gene association exceeded study-wide significance (Fig. 2, Supplementary Figure S5, Supplementary Tables S11 and S12). The top association was DNAAF4 (DYX1C1, MIM: 608706) for which 3 cases harboured ultra-rare functional variants and no controls harboured variants.⁹² We highlighted seven genes (Supplementary Table S13) in our ultra-rare damaging model which had more than one case carrier of a qualifying variant (QV) and no control carriers (5939 genes harboured multiple control carriers and no case carriers, data not shown). HTR7 (MIM: 182137) encodes a serotonin receptor implicated in visceral hypersensitivity in irritable bowel syndrome.⁹³ A more inclusive rare damaging model did not yield compelling associations (Supplementary Figure S5, Supplementary Table S12).

Gene-set analyses

Using gene-sets drawn from recently observed (Fig. 1, Supplementary Tables S1 and S4) and previously associated phenotypes, we tested for enrichment of ultra-rare damaging variants (Supplementary Tables S14 and S15, see Methods).^8,44,49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70^,72,94,95 We only detected statistical evidence in the combined cohort for the 5-gene-set implicated in our prior analysis of Mendelian disorders in IC/BPS(8) (OR, 7.4; 95% CI, 1.6–26.4; P_adjusted = 0.033, CMH). Although neither association was significant separately when adjusted for multiple comparisons, both trended towards significance (MaGIC cohort; OR, 13.5; 95% CI, 1.4–68.2; P_adjusted = 0.058, CMH; BCH cohort; CMH, OR, 5.3; 95% CI, 0.6–25.7; P_adjusted = 0.18, CMH). Importantly, the association was driven entirely by case variants in ATP2C1 (3 cases) and ATP2A2 (1 case). Two IC/BPS affected proband carriers were identified in the previously analysed BCH cohort and two additional carriers were found in the MaGIC cohort. There were 4 individuals in the combined BCH and MaGIC cohort with ultra-rare damaging variants in ATP2C1 or ATP2A2. There was no common co-morbidity pattern across individual 1 (anxiety, IBS), individual 2 (migraine, IBS, anxiety), individual 3 (migraine, Hunner lesion), and individual 4 (no comorbidities). An additional ATP2C1 ultra-rare damaging missense variant was not included because of inadequate representation of geographic ancestry (cluster 11, Supplementary Figure S2B).

We applied the NHC algorithm to perform an unbiased network-based pathway analysis (Table 2, Supplementary Table S16, Supplementary Figures S6 and S7).⁷⁴ The most significantly associated Gene Ontology (GO) biological process pathway from the most significantly enriched gene cluster (cluster 11, Supplementary Figure S6A) was “anaphase-promoting complex-dependent catabolic process” (Table 2, GO: 0031145, P = 6.2 × 10⁻⁶, FET),⁹⁶ which includes SMAD3 (MIM: 613795). Additional noteworthy clusters included the EGF (MIM: 131530) and PIK3R3 (MIM: 606076) genes (Supplementary Table S16, Cluster 10, Supplementary Figure S6E), positive regulation of MAPK cascade (Table 2, GO: 0043410, P = 5.5 × 10⁻⁶, FET) and the integrin/focal adhesion networks (Table 2, Supplementary Table S16, Cluster 14, Supplementary Figure S6I). We also used fast preranked gene-set enrichment analysis (fGSEA) to detect pathways enriched in our ultra-rare variant analysis.⁷⁵ The “transport of small molecules” pathway (GO: 0006810, Reactome R-HSA-382551) was significantly enriched (Supplementary Figures S8–S11, P_adjusted = 0.042, fGSEA).⁹⁷ “Transport of small molecules” was the most significant pathway in all 1000/1000 runs (binomial P < 2.2 × 10⁻¹⁶). Its adjusted P value (fGSEA) was <0.05 in 10% of runs (100/1000, Supplementary Figure S9). In the negative-control analysis, across 1000 runs, no pathway achieved P_adjusted < 0.05 (fGSEA); the minimum adjusted p observed across all pathways and runs was 0.72 (fGSEA). These findings indicate that the enrichment signal for “Transport of small molecules” is robust to fGSEA's stochasticity and unlikely to be a chance result, whereas a similarly sized non-significant gene-set shows no pathway signal (Supplementary Figures S10 and S11).

Exome wide association study (ExWAS)

We performed a rare variant (MAF <1%) exome wide association study (ExWAS) (Fig. 3, Supplementary Tables S17 and S18). No variant associated exceeded study-wide significance. The variant most significantly associated with IC/BPS was a synonymous variant with a low (benign) Transcript-inferred Pathogenicity (TraP) score (0.41) in SLC9A5 [MIM: 600477] (1-48708137-C-A, 7 of 331 case carriers, 8 of 6516 control carriers; OR, 16.2; 95% CI, 4.8–54.1; unadjusted P = 6.6 × 10⁻⁶, CMH).²⁵ SLC9A5 promotes tumour growth and cell motility. It is not explicitly identified in our gene-set enrichment analysis, but it is part of the solute carrier protein superfamily of which six other members are implicated (i.e., SLC15A1, SLC18A1, SLC12A6, SLC35D1, SLC36A4, SLC7A1 in Supplementary Figure S8B).⁹⁸ The second most significant variant association is a SNV in CD68 [MIM: 153634] (17-7483627-A-G, 4 of 331 case carriers, 7 of 6516 control carriers; OR, 38.6; 95% CI, 7.7–168.1; unadjusted P = 2.0 × 10⁻⁵, CMH). It is a synonymous variant with a TraP score of 0.141.²⁵ CD68 encodes a scavenger receptor that is downregulated in Hunner lesions.⁹⁹ Although the small sample sizes leave the possibility that an adequately powered study would reveal these associations to be true-positives, the low (benign) TraP scores suggest that neither association will become significant even with larger sample sizes.

Families with multiple affected members

Eight genes harboured rare variants present in affected (and absent in unaffected) members in at least two families (Supplementary Tables S19 and S20). No genes were notable for previously identified Mendelian conditions associated with IC/BPS. PIEZO2 (MIM: 613629) encodes a mechanically activated ion channel Piezo2 implicated in visceral hypersensitivity in irritable bowel syndrome.¹⁰⁰

Trio analyses

We examined 13 available trios from the BCH cohort (Supplementary Table S21) for de novo variants and compound heterozygosity. No genes were notable for previously identified Mendelian conditions associated with IC/BPS. One de novo variant was in PLXNA4 (MIM: 604280), a candidate for host response and wound healing in the bladder urothelium.¹⁰¹ A gene-set analysis of these genes did not yield a significant association (Supplementary Tables S14 and S15).

Copy Number Variant (CNV) analysis

We reviewed deletions detected using XHMM.⁸¹ We identified one affected individual with a pathogenic CNV (Supplementary Table S22) associated with chromosome 16p11.2 duplication syndrome (MIM: 614671) which has also been identified in congenital anomalies of the kidney and urinary tract (CAKUT).¹⁰² Deletions associated with autosomal recessive conditions were detected but no additional pathogenic variants in the affected gene were detected (Supplementary Table S23).

HLA type analysis

We reviewed Class I and II HLA types for the BCH cohort (Supplementary Table S24). Logistic regression analysis between six individuals with Hunner lesions (n = 6) and those without (n = 181) suggested an association between Hunner lesions and the presence of the HLA alleles DQA1∗ 05:01 and DQB1∗ 02:01 (each with OR 22.75; 95% CI 2.07–249.66, P = 0.01, logistic regression, 5 out of 6 carried both) which is a known haplotype (HLA-DQ2.5).¹⁰³

Discussion

We conducted a phenome-wide comorbidity analysis, exome-wide analysis of ultra-rare and rare damaging variants in IC/BPS encompassing collapsing-based gene burden tests, single variant analyses, gene-set and network-based pathway analyses, family-based studies and CNV diagnostic evaluation to identify genetic risk factors for IC/BPS. The unbiased studies uncovered additional comorbidities and biological pathways in the aetiology of IC/BPS.

Phenotype analyses from eMERGE validated known associations while broadening the possibility of related phenotypes (Fig. 1). Confirmed associations included dysuria,^1,2 polyuria,^1,2 irritable bowel syndrome,¹¹ allergies,⁸⁸ GERD,^10,87 and myalgia.¹¹ Notable associations included anaphylactic shock,⁸⁹ intervertebral disc disorders,⁹⁰ and laxity of ligament or hypermobility syndrome.^91,104 Secondary previously unreported associations with oesophagus ulcers and Barrett's esophagus are likely downstream of GERD.¹⁰⁵ The association with Barrett's esophagus is significant due to its increased risk of cancer¹⁰⁵ providing additional evidence for a common pathophysiology based on epithelial dysfunction.⁸⁷ The association with osteoarthrosis/osteoarthritis is interesting due to overlap in treatments for the two conditions such as pentosan polysulfate sodium.¹⁰⁶

None of the signals in the gene-based collapsing analyses and ExWAS reached study-wide significance, reflecting the genetic heterogeneity of disease (Figs. 2 and 3). Our study thus nominates candidate genes including HTR7 (Supplementary Table S13). HTR7 encodes a 5-HT receptor implicated in visceral hypersensitivity in IBS which is associated with IC/BPS (Fig. 1).^9,93 The top association in gene-based collapsing, DNAAF4, is linked to dyslexia susceptibility and ciliary dyskinesia but also to familial urinary bladder cancer.107, 108, 109

In one family with multiple affected family members (Supplementary Table S20), we identified a rare segregating missense variant in PIEZO2, encoding a mechanosensitive ion channel implicated in urethra/bladder control of micturition.¹¹⁰ Piezo2 is upregulated in CYP-induced cystitis rats¹¹¹ and Piezo2 knockout mice experience incontinence and thickening of the bladder wall.¹¹² Targeting PIEZO2 has been considered as a therapeutic approach for IC-related bladder pain,¹¹¹ and a reduction of PIEZO2 expression was shown after administration of an antisense oligonucleotide treatment for IC/BPS.¹¹³

The analysis of the small number of available trios (Supplementary Table S21) yielded a de novo variant in PLXNA4, which has been linked to bladder epithelial biology in the context of IC/BPS.¹⁰¹ Plxna4⁺ urothelial cells have been shown to be a distinct superficial cell subtype in mouse, rat, and human bladder urothelium that is enriched for genes involved in immune signalling and wound repair. These cells localise to the outermost urothelial layer and are proposed to function as “sentinel” cells initiating host inflammatory and regenerative responses following infection or injury. Because impairment of this signalling network could plausibly contribute to the defective epithelial repair and chronic inflammation characteristic of IC/BPS, a damaging de novo PLXNA4 variant has a biologically plausible mechanism for linking altered urothelial homoeostasis to disease.

Importantly, we confirmed the association with ATP2C1 and ATP2A2, which have been implicated in the desquamating skin disorders Hailey–Hailey and Darier disease (ATP2C1 and ATP2A2 are part of “IC Genes (#1)”, see Supplementary Table S14). ATP2C1 and ATP2A2 are expressed in the urothelium, suggesting the calcium transporter-related epithelial injury observed in these syndromes may also extend to the bladder urothelium and produce the symptoms of IC/BPS.^8,87 It should be emphasised that single gene associations with IC/BPS in this manuscript (e.g., DNAAF4, HTR7, PIEZO2, PLXNA4) have evidence from small numbers of cases. This is in contrast to genes such as ATP2C1 and ATP2A2 which were previously associated with IC/BPS and have additional evidence in this larger cohort.⁸

In the setting of genetic heterogeneity, pathway-based tests (Table 2, Supplementary Figure S8, Supplementary Table S16) yielded multiple additional candidate genes and pathways. SMAD3, which is included in the most enriched gene cluster, has been implicated as a therapeutic target for interstitial cystitis¹¹⁴ and has previously been implicated in bladder cancer.¹¹⁵ Additional genes within the cluster (ANAPC7 [MIM: 606949], CDC20 [MIM: 603618], and SKIL [MIM: 165340]) are linked to bladder cancer,116, 117, 118 and FOXO1 is tied to urothelial fibrosis.¹¹⁹ Other top pathways included the positive regulation of MAPK cascade, the integrin/focal adhesion networks (Table 2, Supplementary Table S16 Cluster 14), and the “small molecule transport” pathway (Supplementary Figure S8) which includes ATP2C1 and ATP2A2. These genes and pathways provide additional avenues for investigation in larger human cohorts and model organisms. Finally, preliminary analysis of HLA genotypes and individuals with Hunner lesions suggested an association with HLA-DQ2.5 haplotype, associated with susceptibility to celiac disease,¹⁰³ which supports an immunogenetic contribution to disease pathogenesis. This analysis was particularly underpowered and should be interpreted as an exploratory analysis and with caution.

We acknowledge the multiple lines of genetic evidence presented in this manuscript do not always converge onto a single genetic explanation for IC/BPS. This prompts a few possible explanations: (1) There is significant genetic heterogeneity underlying the diagnosis IC/BPS. This suggests that there may be subtypes that are genetically distinct but are grouped together in this cohort. Separating them in this manuscript was challenging given the sample size. (2) Small sample size increases the risk of Type II error.

Strengths

This study has several strengths. It combines a large eMERGE-III EHR cohort and two deeply phenotyped IC/BPS exome cohorts, which allows assessment of phenotypic associations in more than 100,000 individuals and rare variant burden in 348 unrelated cases with nearly 12,000 controls. The genomic analyses use a harmonised rare variant framework with ancestry clustering, coverage harmonisation, and stringent quality control to increase robustness of the findings. Convergent results across multiple analytic layers provide complementary biological insight and reduce dependence on any single method. Replication of ATP2C1 and ATP2A2 variation in the expanded cohort supports consistency with prior IC/BPS evidence, while broader network-level analyses highlight epithelial, mechanosensory, and immune pathways that may contribute to disease biology.

There were several limitations of this study. Full phenotype data were not available, potentially reducing our ability to capture the complete clinical spectrum of IC/BPS. We focused exclusively on rare variant risk factors and did not assess common variant risk factors thereby missing a possible genetic contributor. Our eMERGE case size is underpowered, reflecting challenges in studying IC/BPS which has a low prevalence, frequent under-coding, and delayed clinical referrals. Because ICD codes were not further verified with manual inspection of the medical record, the phenotype analysis is subject to non-differential misclassification. These factors bias towards null. Our eMERGE analyses remain susceptible to potential uncorrected confounding variables. Our IC/BPS exome sample size is small which will limit statistical power to detect rare variant associations. No gene or variant reached study-wide statistical significance; thus, the potential roles of candidate genes should be interpreted with caution. Despite modest sample sizes, our gene-level tests successfully identified biologically coherent calcium-handling genes that converged independently in pathway analyses, suggesting that our analytical framework effectively captures meaningful genetic associations even in smaller cohorts. The HLA analysis was particularly underpowered and should be interpreted as exploratory and with caution. The BCH and MaGIC cohorts may not fully represent the broader IC/BPS population. As they were recruited primarily through tertiary care centres and advocacy networks, there may be biases related to geography, healthcare access, socioeconomic status, and ancestry, with an overrepresentation of individuals of European descent. Our analysis was driven primarily by individuals of European ancestry (Supplementary Figure S2B). Larger and multi-ancestry cohorts will provide additional genetic insight.

Integrated signals across analytic levels identify three convergent axes of IC/BPS biology. Although no single gene or variant achieved study-wide significance in isolation, triangulating across our collapsing analyses, family-based enrichment, ExWAS, network/pathway enrichment, CNVs, and HLA revealed coherent convergence on (i) epithelial Ca²⁺ handling/barrier integrity, (ii) mechanosensation-MAPK/integrin signalling, and (iii) mucosal autoimmunity. The only a priori gene-set that replicated was our ATP2A2/ATP2C1-containing set, nominating Ca²⁺ pump biology and epithelial integrity as a shared risk axis across cohorts. Family-level recurrence of PIEZO2 and network-based heterogeneity clustering that highlighted MAPK cascade regulation and integrin/focal-adhesion modules point to a mechanotransduction program linking urothelial stretch sensing to downstream signalling. Preliminary HLA-DQ2.5 enrichment in individuals with Hunner lesions implicates antigen presentation pathways typical of mucosal autoimmunity. Together, these signals argue that IC/BPS genetic risk aggregates into interacting epithelial, sensory-mechanotransductive, and immune pathways rather than a single locus effect.

The phenotype associations encourage clinicians caring for individuals with IC/BPS to be vigilant about comorbid conditions including disorders of allergy, the oesophagus, the epithelium, joint laxity, and vertebral anomalies. Gene-set analyses identified predicted-damaging rare variants in genes for monogenic forms of desquamating skin disorders and evidence for a common pathway for those genes. Furthermore, this study supports the association of IC/BPS with several pathways such as SMAD3, cell cycle and integrin signalling as potential pathogenic mechanisms of disease. In addition, our study indicates that a larger sample size would continue to better delineate the genetic architecture of IC/BPS.

Contributors

Conceptualization: JEM, AGG, CAB.

Data curation: JEM, AM, SKM, AK, MV, MW, CW, HH, YL, KK, EE, CAB.

Formal Analysis: JEM, AM, SKM, MV, AK.

Funding acquisition: JEM, AGG, LK, CAB.

Investigation: JEM, AM, SKM, CAB.

Methodology: JEM, AM, SKM, JB, NH, MV, LK, RL, RA, AK, WS, KEB, CAB.

Project administration: JEM, MW, EE, CAB.

Visualization: JEM, AM, SKM, CAB.

Resources: AGG, CAB.

Software: JEM, AM, SKM.

Supervision: AGG, CAB.

Writing—original draft: JEM, SKM, CAB.

Writing—review & editing: JEM, SKM, EE, RL, RA, AGG, CAB.

J.E.M. and C.A.B. had full access to all the data in the study and verified the underlying data.

All authors read and approved the final manuscript.

J.E.M., A.G.G., and C.A.B. made the decision to submit the manuscript.

Data sharing statement

The Genotype-Tissue Expression (GTEx) Project was supported by the Common Fund of the Office of the Director of the National Institutes of Health, and by NCI, NHGRI, NHLBI, NIDA, NIMH, and NINDS. The data used for the analyses described in this manuscript were obtained from: the GTEx Portal on 5/4/2023.

Declaration of interests

AGG has received grants from Natera and has served on advisory boards for Natera through a service agreement with Columbia University. AG has also served on advisory boards for Actio Biosciences and Novartis, and has stock options for Actio Biosciences. All other authors declare they have no competing interests.