Bidirectional Association between Atrial Fibrillation and Ovarian Cancer: Evidence from the UK Biobank and Mendelian Randomization.

As depicted in Fig. 1 , this analysis included 273,170 female subjects of the UKB at baseline. After the exclusion of subjects with missing data on BMI ( n = 1,458) and body fat mass ( n = 5,208), subjects with reported AF events ( n = 2,042) and ovarian cancer ( n = 613) and fallopian tube cancer ( n = 14) before the first assessment date were excluded. Flowchart of participant selection of the UKB cohort. OCA, ovarian cancer. Baseline characteristics of the study population are summarized in Table 1 . A total of 265,248 eligible individuals were included in the analysis (mean age: 56.29 years). During follow-up, 13,027 (4.9%) developed AF, and 1,256 (0.5%) were diagnosed with either ovarian cancer ( n = 1,123, 0.4%) or fallopian tube cancer ( n = 133, 0.1%). Among ovarian cancer cases, there were 69 endometrioid (0.03%), 633 serous (0.2%), and 110 mucinous (0.04%) tumors. The mean BMI and body fat mass were 27.1 kg/m 2 and 26.9 kg, respectively. At baseline, 160,323 (60.4%) women were postmenopausal, and 244,554 (92.2%) were of White ethnicity. A total of 239,938 participants reported current alcohol use (90.5%), and 23,646 participants (59.9%) were never smokers. With respect to medication history, 32,749 (12.3%) reported cholesterol-lowering medication use, 45,628 (17.2%) reported blood pressure–lowering medication use, and 19,230 (7.2%) reported hormone replacement therapy. Comorbidities included hypertension in 22.7%, diabetes in 0.4%, myocardial infarction in 0.8%, and heart failure in 0.2% of participants. Baseline characteristics among the UKB cohort ( N = 265,248). NOTE: Values are presented as mean (SD) for continuous variables and number (percentage) for categorical variables. OCA and FTC events are shown both in combination and separately by tumor site. OCA histologic subtypes include endometrioid, serous, and mucinous tumors. Abbreviations: FTC, fallopian tube cancer; OCA, ovarian cancer. A total of 265,248 eligible individuals were included in the analysis. During follow-up, 1,256 incident ovarian cancer cases and 13,027 incident AF cases were identified in the overall cohort. For the analysis of the ovarian cancer to AF association, the median follow-up duration was 5,399 days (approximately 14.8 years), whereas the median follow-up for the AF to ovarian cancer association was 5,417 days (approximately 14.8 years). In bidirectional analyses, both AF to ovarian cancer and ovarian cancer to AF associations were statistically significant ( Fig. 2A ). In the ovarian cancer to AF direction, women with ovarian cancer had an increased risk of new-onset AF (HR, 1.75; 95% CI, 1.43–2.14). Conversely, in the AF to ovarian cancer direction, new-onset AF was associated with a higher risk of developing ovarian cancer after multivariable adjustment (HR, 1.30; 95% CI, 1.05–1.61). In both AF to ovarian cancer and ovarian cancer to AF Cox regression models, Schoenfeld residual tests indicated modest departures from proportionality (global P < 0.05; Supplementary Table S2); these were addressed by time-stratified and stratified Cox analyses, which produced directionally consistent HRs. Bidirectional associations between AF and ovarian cancer and cumulative risk curves. A, Forest plots illustrating bidirectional associations between AF and ovarian cancer using Cox proportional hazards regression. The associations of ovarian cancer to AF and AF to ovarian cancer were each stratified by the interval between disease onset. P for interaction was statistically significant. All models were adjusted for age, ethnicity, menopause, BMI, body fat mass, use of cholesterol-lowering medication, use of blood pressure–lowering medication, hormone replacement therapy, smoking status, alcohol usage status, history of hypertension, diabetes, heart failure, and myocardial infarction. B, Cumulative risk of ovarian cancer in participants with and without AF. C, Cumulative risk of AF in participants with and without ovarian cancer. Solid lines represent estimated cumulative risk; dashed lines indicate 95% CIs. OCA, ovarian cancer. Bold text denotes statistically significant associations ( P < 0.01). When stratifying the ovarian cancer to AF association by the time interval between the date of attending the cohort and AF onset ( Fig. 2A ), the strongest effect estimate was observed in the group with the longest interval (>9 years, HR, 2.54; 95% CI, 1.39–4.64), and the P for interaction was statistically significant ( P 9 years (HR, 1.63; 95% CI, 1.13–2.34; P for interaction < 0.01). Cumulative incidence analyses demonstrated a bidirectional association between AF and ovarian cancer. Women with AF had a higher cumulative incidence of ovarian cancer compared with those without AF ( Fig. 2B ), and conversely, women with ovarian cancer had a higher cumulative incidence of AF than those without ovarian cancer ( Fig. 2C ). The proportional hazards assumption was examined using Schoenfeld residuals (R function cox.zph). As shown in Supplementary Table S2, modest violations were observed for AF status, age, and ethnicity in the AF to ovarian cancer model. Violations were also observed for ovarian cancer status, cholesterol-lowering medication, blood pressure–lowering medication, hypertension, and myocardial infarction in the ovarian cancer to AF model. Global tests for both models indicated mild departures from proportionality ( P < 0.05). These deviations were addressed through time-stratified and stratified Cox sensitivity analyses, which yielded consistent HR estimates, supporting the robustness of the main results. Given the significant association between ovarian cancer and new-onset AF, we further investigated this link by histologic subtype ( Fig. 3 ). This analysis revealed that the impact was mainly driven by serous ovarian cancer (HR, 1.83; 95% CI, 1.40–2.39). Risk of new-onset AF by ovarian cancer histologic subtype and in subgroups defined by participant characteristics. The figure shows the associations between different histologic subtypes of ovarian cancer (endometrioid, mucinous, and serous) and the risk of developing AF, as estimated by Cox proportional hazards regression models. It also depicts the relationship between ovarian cancer and new-onset AF within various population subgroups, including individuals with hypertension, myocardial infarction, smoking, alcohol use, menopause, and White ethnicity. Values represent numbers or HRs with 95% CIs. Adjusted covariates were the same as in Fig. 2 . Bold text denotes statistically significant associations ( P < 0.01). As presented in Fig. 2A , a history of ovarian cancer was identified as a risk factor for developing AF. The robustness of this finding was confirmed through subgroup analyses, which showed directionally consistent associations ( Fig. 3 ). Notably, the risk was significant in participants both with (HR, 1.64; 95% CI, 1.20–2.24) and without (HR, 1.84; 95% CI, 1.41–2.38) a history of hypertension. The association remained significant when stratified by history of myocardial infarction, occurring in participants with (HR, 4.25; 95% CI, 1.57–11.49) and without (HR, 1.71; 95% CI, 1.39–2.10) this condition. Similarly, a significant risk was evident among both never smokers (HR, 1.93; 95% CI, 1.47–2.53) and previous smokers (HR, 1.76; 95% CI, 1.29–2.39). Notably, the P values for interaction were nonsignificant across all these subgroups. As demonstrated in Fig. 4A , in order to further clarify the relationship between AF and ovarian cancer, we employed two-sample MR analysis and revealed a significant association between AF and an elevated risk of ovarian cancer ( Fig. 4A ; OR, 1.07; 95% CI, 1.02–1.13; P value = 0.01, after removing pleiotropic SNPs with the MR-PRESSO method). This association persisted across ovarian cancer subtypes, including low-grade serous ovarian cancer ( Fig. 4A ; OR, 1.19; 95% CI, 1.01–1.39; P value = 0.03), endometrioid ovarian cancer ( Fig. 4A ; OR, 1.12; 95% CI, 1.01–1.24; P value = 0.03), and a combination of high- and low-grade serous ovarian cancer ( Fig. 4A ; OR, 1.05; 95% CI, 1–1.11; P value = 0.05). Leave-one-out analyses indicated no significant influence of any single genetic variant on the overall causal estimates for ovarian cancer (Supplementary Fig. S1A), endometrioid ovarian cancer (Supplementary Fig. S1B), low-grade serous ovarian cancer (Supplementary Fig. S1C), and combined high- and low-grade serous ovarian cancer (Supplementary Fig. S1D). Although the genetic association between AF and ovarian cancer was statistically significant, the effect size was small (OR, 1.05), indicating only a very modest absolute increase in risk. We provide a step-by-step estimate of the population-level impact in the Supplementary Materials and Supplementary Methods. MR analysis of a genetic relationship between AF and ovarian cancers using two different AF-GWAS datasets. This figure demonstrates MR analysis (inverse variance weighted method) of a genetic relationship between AF and ovarian cancers using GWAS datasets ( A ) ebi-GCST006414 and ( B ) ebi-GCST006061. *, Result after removing pleiotropic SNPs with the MR-PRESSO method. Values are ORs (95% CIs). Bol text denotes statistically significant associations ( P < 0.01). However, as depicted in Supplementary Table S3, considering potential pleiotropy for ovarian cancer, MR-Egger analysis (OR, 0.97; 95% CI, 0.89–1.07; P value = 0.58) and weighted median analysis (OR, 0.96; 95% CI, 0.90–1.03; P value = 0.27) did not support previous findings. The MR-PRESSO result suggested potential pleiotropy (MR-PRESSO global test P value = 0.23). Exclusion of two pleiotropic SNPs (rs1563304 and rs1906615) identified by MR-PRESSO rendered the results statistically significant (Supplementary Table S3, Excluded, OR, 1.07; 95% CI, 1.02–1.13; P value = 0.01). Additional MR analysis showed no evidence for a genetic effect of ovarian cancer liability on AF risk across all methods (Supplementary Table S4). To address potential differences between observational and genetic estimates, we conducted an exploratory comparison between the cohort-derived HR (HR, 1.30; 95% CI, 1.05–1.61) and the MR-derived OR (OR, 1.05; 95% CI, 1.00–1.11). On the log scale, the difference was not statistically significant (Z = 1.90, P value = 0.057). A sensitivity analysis excluding pleiotropic SNPs (MR OR, 1.07; 95% CI, 1.02–1.13) likewise showed no significant difference when compared with the cohort HR (Z = 1.73, P value = 0.084). These results (Supplementary Table S5) indicate no meaningful discrepancy between Cox regression and MR estimates, both consistently supporting a modest, directionally concordant association between AF liability and ovarian cancer risk. In addition, we validated these findings using an alternative AF GWAS dataset as the exposure (Supplementary Table S6; Fig. 4B ). Consistent genetic liability to AF was associated with ovarian cancer ( Fig. 4B ; OR, 1.05; 95% CI, 1.00–1.11; P value = 0.05) and a combination of high- and low-grade serous ovarian cancer ( Fig. 4B ; OR, 1.06; 95% CI, 1.00–1.13; P value = 0.04). Using the MR-PRESSO method, we detected two pleiotropic SNPs (rs242557 and rs55985730) for ovarian cancer. Even after removing these pleiotropic SNPs, the association remained statistically significant (Supplementary Table S6, Removed, OR, 1.05; 95% CI, 1.00–1.10; P value = 0.04). Utilizing PhenoScanner version 2, 51 out of 90 initial SNPs were identified as not associated with potential confounding factors. MR analysis with these 51 SNPs for ovarian cancer revealed a significant association between AF and ovarian cancer (OR, 1.07; 95% CI, 1.00–1.16; P value = 0.04), without evidence of pleiotropy for ovarian cancer (MR-Egger P value = 0.79; Supplementary Table S7). The MR-Steiger test confirmed the genetic effect direction (all P _steiger < 0.01; Supplementary Table S8). Detailed information on the SNPs used in our analysis is provided in Supplementary Table S9. Using the MR-TRYX approach, we further investigated horizontal pleiotropy between AF and ovarian cancer, as well as their association. Radial MR identification of eight outlier SNPs for ovarian cancer were identified. Among the 4,835 phenotypes accessible on MR-Base, 161 candidate traits exhibited a significant connection with outliers at a threshold of P < 5 × 10 −8 for ovarian cancer. Following least absolute shrinkage and selection operator regression, nine traits were prioritized in the context of ovarian cancer: neuroticism, feeling hurt, basal metabolic rate, leg fat-free mass (right), leg fat-free mass (left), leg predicted mass (right), leg predicted mass (left), and sensitivity/hurt feelings. We adjusted the exposure–outcome association for detected pleiotropic pathways, minimizing the influence of outliers (Supplementary Table S8). After adjusting for all outliers, the effect estimate remained consistent with the original IVW estimate (both OR > 1; P value 1; P value < 0.05). Furthermore, after the removal of outliers, overall heterogeneity was notably reduced by 19.38% (Q = 95.07) for ovarian cancer Details about the candidate traits related to each outlier are available in Supplementary Table S9. Through MR analysis utilizing common risk factors of AF and ovarian cancer subtypes, we selected three risk factors (BMI, childhood BMI, class II obesity) and explored potential vertical pleiotropic (mediating) pathways through these factors. In MVMR, after adjustment for genetically predicted BMI, childhood BMI, and class II obesity, the correlation between genetically predicted AF and ovarian cancer completely disappeared (all P value > 0.05, Supplementary Fig. S2), suggesting that BMI and obesity may serve as potential mediators in the association between AF and ovarian cancer. The process of discovering high-risk mutual genes of both AF and ovarian cancer is shown in Supplementary Fig. S3A. SNPs were gleaned from GWAS using the MR method among AF and four subtypes of ovarian cancer. Then, MR-TRYX analysis was conducted to identify potential horizontal pleiotropic pathways in the relationship between AF and subtypes of ovarian cancers (Supplementary Tables S10 and S11). In sensitivity analyses using MR-TRYX and MVMR, we observed that BMI and obesity may serve as potential mediators between AF and ovarian cancer. Adjustment for genetically predicted BMI and obesity attenuated the AF–ovarian cancer association (Supplementary Fig. S2, all P > 0.05). Full details are provided in Supplementary Results (Supplementary Tables S10 and S11; Supplementary Fig. S2). A comprehensive meta-analysis using GWAS summary data for AF and the four ovarian cancer subtypes was conducted to investigate shared genetic risk loci. The identification of common genetic variants was achieved through two distinct gene mapping methodologies using the tool FUMA: (i) location-based mapping, which relied on genomic proximity, and (ii) eQTL mapping, predicated on gene expression patterns across diverse tissues. Through FUMA analysis, a set of 1,400 candidate SNPs ( P < 0.05) were identified associating with AF, including 30 lead variants ( P < 5 × 10 −8 ) at 30 risk loci exhibiting significant associations with both AF and ovarian cancer. Additionally, 144 mapped genes based on these results were identified (Supplementary Fig. S3B). Tissue enrichment analysis revealed that the majority of these genes were predominantly expressed in cardiac tissue (Supplementary Fig. S3C). Furthermore, Gene Ontology biological pathway analysis highlighted the enrichment of gene sets related to mitochondrial respiratory chain complex assembly, regulation of reactive oxygen species metabolic processes, and NADH dehydrogenase complex assembly (Supplementary Fig. S3D). SMR analysis was additionally performed to find new genes and loci associated with AF and ovarian cancer (Supplementary Fig. S3E). We selected mutual significant genes in both SMR and FUMA results. In our SMR analysis, we identified seven probes tagging five unique genes that exhibited pleiotropic or potential genetic associations with both AF and ovarian cancer. These five genes were consistent with our FUMA analysis (Supplementary Table S12). The top three probes were ILMN_2115154 (tagging nucleoporin-like 2, P SMR = 0.00028), ILMN_1789616 (tagging nucleoporin-like 2, P SMR = 0.000356), and ILMN_1725612 (tagging NUP50 , P SMR = 0.000552). One probe tagged NUP50 , and two probes tagged SYTL2 . NUP50 and SYTL2 seemed significant in both SMR results and FUMA results. We further performed survival analysis using the TCGA database and found that the expression of NUP50 ( Fig. 5A ) and SYTL2 ( Fig. 5B ) genes was significantly correlated with the prognosis of patients with ovarian cancer (both P < 0.05). Prognosis analysis of NUP50 and SYTL2 . Kaplan–Meier survival analysis of patients with ovarian cancer based on gene expression levels of ( A ) SYTL2 and ( B ) NUP50 , as derived from TCGA data. C, RT-qPCR analysis and ( D ) Western blot of SKOV3 after the knockdown of NUP50 and SYTL2 genes by siRNA. E, CCK-8 assays of SKOV3 transfected with siRNA targeting human NUP50 (si- NUP50 ) and SYTL2 (si- SYTL2 ) or control. Representative images are shown. Data are presented as means ± SD. **, P < 0.01; ****, P < 0.0001; n = 5. To further investigate the potential function of NUP50 and SYTL2 , the ovarian cancer cell line SKOV3 was selected for the RNA interference experiment and transfected with siRNA targeting human NUP50 and SYTL2 (si- SYTL2 ) siRNA. The silencing efficiency of the siRNA on NUP50 and SYTL2 expression was detected by RT-qPCR ( Fig. 5C , P < 0.05) and Western blot ( Fig. 5D ; Supplementary Fig. S4). CCK-8 was performed to examine the effect of NUP50 and SYTL2 on the proliferation of ovarian cancer cells. As determined by the CCK-8 assay, downregulating NUP50 and SYTL2 ( Fig. 5E ) slowed cell proliferation in a time-dependent manner in SKOV3 cells compared with controls ( P < 0.05).

To the best of our knowledge, this is the first study to use multiple methods to examine the potential connection between AF and ovarian cancer. Our study has three main findings. First, our analysis using the UKB cohort highlights a bidirectional association between new-onset AF and ovarian cancer. Second, we demonstrate a genetic association between AF and an elevated risk of ovarian cancer, as well as high- and low-grade serous ovarian cancer. Lastly, our investigation has also identified shared significant genes in both AF and ovarian cancer pathogenesis, such as NUP50 and SYTL2 , which have been found to be correlated with the prognosis of patients with ovarian cancer. The bidirectional association between AF and ovarian cancer may be explained by several biological pathways. In the AF to ovarian cancer direction, the prothrombotic state seen in AF may also alter the tumor microenvironment, promoting angiogenesis and facilitating tumor growth ( 30 ). In addition, AF and ovarian cancer share common risk factors such as metabolic syndrome, obesity, and hormonal changes, which may contribute to their co-occurrence ( 31 ). Conversely, in the ovarian cancer to AF direction, the stronger association observed may reflect both disease- and treatment-related effects. Large-volume ascites or pelvic masses in ovarian cancer can increase intra-abdominal pressure and cardiac load, potentially triggering AF onset ( 32 ). Moreover, common chemotherapy agents for ovarian cancer, such as cisplatin, gemcitabine, and 5-fluorouracil, have been associated with atrial remodeling and heightened AF risk ( 30 , 33 ). Therefore, current evidence supports our findings and highlights the necessity for a more comprehensive clinical approach to the management and prevention of both AF and ovarian cancer. Previous MR studies have reported heterogeneous findings about cardiovascular diagnoses and cancer risk. For example, AF liability has been linked to lung, breast, cervical, endometrial, and melanoma cancers (medRxiv 2021.01.10.21249534), whereas other large-scale MR investigations found little evidence for a broad causal effect of common cardiovascular conditions (heart failure, hypertension, stroke) on cancer incidence ( 34 ). In our MR analyses, genetically proxied AF liability was associated with a modest but statistically significant increase in ovarian cancer risk. This supports a potential genetic contribution of AF to ovarian cancer development, consistent with observational estimates. In our cohort, this corresponds to an absolute risk increase from ∼0.47% to ∼0.49%, or about 24 additional cases per 100,000 women (Supplementary Methods). Although the absolute effect size is small and the public health implications are limited, these findings still point to potential shared biological pathways linking AF and ovarian cancer, which merit further mechanistic investigation. Importantly, the reverse direction (ovarian cancer to AF) was not supported by MR, suggesting that the strong observational association is unlikely to be genetically driven but rather reflects long-term disease burden (e.g., ascites, tumor mass; ref. 35 ) and treatment-related effects (e.g., cardiotoxic chemotherapy; ref. 36 ). Our time-stratified analyses further indicated that the ovarian cancer to AF association was most pronounced after extended follow-up, consistent with a chronic rather than short-term process. Improved survival in patients with ovarian cancer may also contribute ( 37 ), as longer survival increases the likelihood of AF diagnosis many years after the initial cancer diagnosis. Together, these findings underscore the importance of monitoring cardiovascular complications in ovarian cancer survivors and highlight the need to elucidate biological pathways linking cancer progression with atrial remodeling. The observed association between AF and ovarian cancer seemed to be driven primarily by serous histotypes. This pattern may be explained by differences in pathogenesis and prevalence. Serous tumors, especially high-grade serous carcinoma, are more common, often diagnosed at advanced stages ( 38 ), and are associated with systemic inflammation and widespread genomic instability, including p53 alterations ( 39 ), which may intersect with AF-related inflammatory and metabolic pathways ( 40 ). In contrast, mucinous ovarian cancers may have origins in the gastrointestinal tract, and clear cell ovarian cancers are often linked to endometriosis ( 41 ); both subtypes are less prevalent and may follow distinct biological pathways with limited overlap with AF-related mechanisms ( 41 ). The smaller sample sizes for these histotypes may also have reduced the statistical power to detect associations. NUP50 and SYTL2 were identified as shared genetic signals between AF and ovarian cancer through integrative genomic analyses combining GWAS, SMR, and FUMA. NUP50 and SYTL2 could enhance ovarian cancer cell proliferation, and patients with elevated NUP50 and SYTL2 expression may have a decreased likelihood of survival compared with those with low expression levels of these genes. NUP50 , a nucleoporin protein, has been implicated in nuclear transport and gene regulation, and recent studies suggest its involvement in cardiac electrical activity and myocardial ischemic injury ( 42 ), which could influence AF susceptibility. Additionally, in ovarian cancer, NUP50 has been linked to p53 ubiquitination, contributing to platinum resistance ( 43 ), indicating its role in tumor progression. In addition to NUP50 , SYTL2 has also emerged as a gene of interest. Consistent with previous reports ( 44 ), SYTL2 knockdown reduced proliferation in ovarian cancer cells. Our findings extend this observation by identifying SYTL2 as a potential shared pathogenic gene linking AF and ovarian cancer in genomic analyses. SYTL2 , a SYTL protein, may also act as a prometastatic factor, indicating poorer patient survival ( 44 ). GO biological pathways revealed metabolic pathways related to these two genes, which were the assembly of mitochondrial respiratory chain complexes, regulation of reactive oxygen species metabolic processes, and assembly of NADH dehydrogenase complexes. Taken together, these findings indicate that NUP50 and SYTL2 may be involved not only in ovarian cancer progression but also in early biological processes linked to AF-related pathways (e.g., nuclear transport for NUP50 , vesicle trafficking for SYTL2 ), providing a plausible mechanistic link between AF and ovarian cancer. Several limitations should be acknowledged. First, as the UKB cohort is predominantly White, the generalizability of our findings to more diverse populations may be limited, highlighting the need for further validation in multiethnic cohorts ( 45 ). Second, potential biased random effects in the estimated influence of MR on cancer risk may arise due to AF being a binary exposure ( 46 ). Third, it is important to note that the coverage of the MR database may not be comprehensive enough and could overlook other pleiotropic phenotypes. Fourth, we were unable to exclude participants with elevated genetic or molecular biomarkers (e.g., BRCA mutation carriers), as these data were not available for the entire cohort. Additionally, negative results in MVMR analysis might stem from low conditional power after considering putative gene-mediator effects. It is crucial to understand that negative outcomes in MVMR should not be interpreted as definitive confirmation or absence of mediating pathways ( 47 ). Finally, potential misclassification of AF should be acknowledged. As a highly prevalent condition, AF may be underdiagnosed in routine clinical practice, and even when identified through ICD-10 coding (I48), the positive predictive value has been reported to be 67.5% ( 48 ), which could have led to some degree of exposure misclassification in our study. Therefore, future studies are imperative to confirm and validate our findings. Our study provides observational and genetic evidence supporting a bidirectional association between AF and ovarian cancer, with the association from ovarian cancer to AF being stronger. NUP50 and SYTL2 emerged as potential genes of interest linking the two diseases, warranting further investigation in future studies.

Cardiovascular diseases and cancer are the two predominant causes of morbidity and mortality globally, exhibiting a significant interrelation that may potentially worsen each other ( 1 ). Atrial fibrillation (AF), the most common type of arrhythmia characterized by irregular heart rhythms, affected 46.3 million individuals worldwide in 2016 ( 2 ) and is projected to triple its prevalence over the next two decades ( 3 ). Several population-based studies have reported a higher prevalence of AF among patients with cancer compared with individuals without cancer, even prior to any AF-related treatment, suggesting that AF and cancer may share common underlying mechanisms ( 4 ). Additionally, patients with cancer face an increased risk of developing AF, particularly within the first 90 days after diagnosis, suggesting a shared pathophysiologic process ( 5 ). A possible link between AF and ovarian cancer has been proposed in recent studies. A study encompassing 269,742 Danish individuals from the Danish National Registry found an association between new AF and subsequent cancer diagnoses within a 3-month period, including ovarian cancer ( 6 ). However, given that the majority of ovarian cancers are diagnosed at an advanced stage ( 7 ), it is plausible that many patients diagnosed with AF in that time frame might already have undiagnosed ovarian cancer. Conversely, an analysis of 816,811 Korean patients with cancer from the National Health Insurance Service database pointed out that patients with ovarian cancer were associated with a higher risk of AF 5 years after diagnosis ( 8 ). Although traditional epidemiologic studies might be affected by residual confounders and reverse causality, it is important to note that AF and ovarian cancer share mutual risk factors such as age ( 9 , 10 ) and obesity ( 10 , 11 ). This highlights the challenge of establishing temporality between AF and ovarian cancer in observational data and underscores the need for analyses that can better address reverse causality, such as long-term follow-up and Mendelian randomization (MR). Clarifying the relationship between AF and ovarian cancer is clinically and scientifically important. Prior literature has suggested possible bidirectional associations between AF and cancer ( 12 ), but most studies were limited by short follow-up and could not determine whether AF precedes cancer or vice versa . To overcome these limitations, we leveraged the large UK Biobank (UKB) with long-term follow-up to conduct bidirectional time-to-event analyses focused on ovarian cancer and, in parallel, performed MR to test whether genetic liability to AF is associated with ovarian cancer risk while reducing confounding and minimizing bias from reverse temporality. We further integrated cis-expression quantitative trait loci (cis-eQTL)–based gene prioritization, including Functional Mapping and Annotation (FUMA), summary data–based MR (SMR), and brief functional experiments, nominating nucleoporin 50 ( NUP50 ) and synaptotagmin-like 2 ( SYTL2 ) as genes of interest. We subsequently conducted loss-of-function experiments in the SKOV3 ovarian cancer cell line as a means to provide functional support for the genomic findings. Together, these complementary approaches provide novel insights beyond prior observational reports by clarifying temporality, evaluating genetically proxied risk, and offering biological context for the AF–ovarian cancer association.

We included female participants in the UKB with complete baseline data. We excluded (i) those with a history of AF prior to baseline (to ensure all AF cases were incident cases during follow-up), (ii) those with missing covariate data, and (iii) those with a diagnosis of ovarian cancer prior to baseline (to ensure cancer incidence was captured prospectively). For covariates such as hypertension, diabetes, heart failure, and myocardial infarction, we defined “previous history” based on International Classification of Diseases, 10th Revision (ICD-10) codes recorded before baseline; these participants were retained in the study cohort, and their comorbidity status was included as an adjustment variable in the regression models. For this study, we included subjects from the UKB, a population-based cohort study that has been tracking more than 500,000 United Kingdom residents since 2006. The data collection process involved questionnaires, interviews, regular visits to assessment centers, and linkage to health records in order to gather a comprehensive range of psychosocial, sociodemographic, physical, and genetic information. Subjects who did not provide informed consent for follow-up data collection at baseline or during the follow-up period were excluded from the UKB. For MR analysis, we utilized publicly available cancer genome-wide association study (GWAS) summary data from the Medical Research Council Integrative Epidemiology Unit ( https://gwas.mrcieu.ac.uk/ ) and the Ovarian Cancer Association Consortium (OCAC; https://ocac.ccge.medschl.cam.ac.uk/ ). Details of the GWAS data used in this study are shown in Supplementary Table S1. For SMR analysis, we employed summarized eQTL data from the Consortium for the Architecture of Gene Expression, which included 2,765 participants from whole blood samples. This eQTL dataset is available for download at the following link: https://cnsgenomics.com/data/SMR/#eQTLsummarydata . Subjects were continuously monitored for disease occurrences through linkages with health-related medical records, including primary care data, hospital inpatient data, death register records, and self-reported medical conditions. All diseases were recorded based on ICD-10 codes. Cases were defined as subjects with a report of any AF or ovarian cancer after the first assessment date. AF was defined as I48 of the ICD-10 code. In UKB, AF (I48) and ovarian cancer (C56/C57) definitions have remained stable across all linked datasets, including after the adoption of the ICD-10 fifth edition in 2016; ICD-10-CM codes are not used in the UK ( 13 ). For this study, “ovarian cancer” was defined as ICD-10 codes C56 (ovarian cancer) and C57 (fallopian tube cancer), given that fallopian tube cancers are often high-grade serous ovarian cancers and are clinically managed as part of the same disease entity. These two codes were combined into a single outcome category in all analyses, with follow-up until December 31, 2023. The follow-up time was defined as the period from baseline enrollment until the diagnosis of incident ovarian cancer or AF, death, or the end of the study, whichever occurred first. Histotype information for ovarian cancer was obtained from the UK Cancer Registry, specifically the serous, endometrioid, clear cell, and mucinous subtypes. Data on age (continuous), ethnicity (White or non-White), smoking status (current, never, or previous), alcohol usage status (current, never, or previous), use of cholesterol-lowering medication (yes or no), use of blood pressure–lowering medication (yes or no), hormone replacement therapy (yes or no), and menarche status (yes or no) were collected through questionnaires and interviews at the assessment center. Furthermore, body mass index (BMI), defined as body weight divided by the square of height (kg/m 2 ), and total body fat mass (kg) were measured at the date of the first assessment center visit. If ICD-10 codes for hypertension (I10 or I15), heart failure (I50), myocardial infarction (I21), or diabetes (E10) were recorded before the first assessment date, participants were classified as having a baseline history of these conditions. These comorbidities were not considered part of the exclusion criteria but were included as covariates in the multivariable Cox regression models. The study population’s characteristics were presented as counts and percentages. We fitted two prespecified Cox proportional hazards models to assess the association in both directions: 1. AF to ovarian cancer model: Ovarian cancer was the outcome; AF status was modeled as a time-updated exposure (unexposed until incident AF; exposed thereafter). 2. varian cancer to AF model: AF was the outcome; ovarian cancer status was modeled as a time-updated exposure (unexposed until incident ovarian cancer; exposed thereafter). 1. AF to ovarian cancer model: Ovarian cancer was the outcome; AF status was modeled as a time-updated exposure (unexposed until incident AF; exposed thereafter). 2. varian cancer to AF model: AF was the outcome; ovarian cancer status was modeled as a time-updated exposure (unexposed until incident ovarian cancer; exposed thereafter). HRs and 95% confidence intervals (CI) were calculated. The analysis was adjusted for age (continuous), ethnicity, BMI, body fat mass, use of cholesterol-lowering medication, use of blood pressure–lowering medication, hormone replacement therapy, smoking status, alcohol usage status, history of hypertension, history of diabetes, history of heart failure, and history of myocardial infarction. The cumulative incidence of ovarian cancer and AF was calculated in both directions (AF to ovarian cancer and ovarian cancer to AF) using the Kaplan–Meier method, with the log-rank test applied to compare individuals with and without the exposure condition. To further evaluate the genetic susceptibility in the association between AF and ovarian cancer, we performed two-sample MR using traditional methods (inverse variance weighted, MR-Egger, weighted median). MR analysis has specific constraints, and detailed information can be found elsewhere ( 14 ). In our two-sample MR, sample 1 refers to the GWAS of the exposure (genetic liability to AF), and sample 2 refers to the GWAS of the outcomes (overall ovarian cancer and histotypes). Specifically, sample 1 used the AF GWAS ebi-a-GCST006414 (European ancestry), and sample 2 used the OCAC summary GWAS for overall ovarian cancer and subtypes (see Supplementary Table S1). Genome-wide significant SNPs for AF ( P < 5 × 10 −8 ) were clumped for independence (r 2 < 0.001, 10 Mb window) and harmonized with the ovarian cancer GWASs. During harmonization, we employed a conservative approach, correcting the strand for nonpalindromic SNPs and excluding all palindromic SNPs from the MR analysis ( 15 ). Lastly, the final genetic instrument variables (IV) were determined through this comprehensive screening process, with an F-statistic exceeding 10 indicating that weak instruments had a relatively low risk of bias ( 16 ) and R 2 representing the degree to which instrumental variables explain exposure ( 17 ). To evaluate the role of genetic susceptibility in the association between AF and ovarian cancer, we conducted several additional analyses: (i) utilizing PhenoScanner version 2 to identify SNPs that are not associated with potential confounding factors such as BMI and trunk fat-free mass, and conducting MR analysis with these 51 SNPs; (ii) determining the causal direction from AF to ovarian cancer using the MR-Steiger test; (iii) using an alternative AF GWAS dataset to reassure the causal relationship between AF and ovarian cancer; and (iv) using the MR pleiotropy residual sum and outlier (MR-PRESSO) method to detect two pleiotropic SNPs for ovarian cancer. We used MR-PRESSO to detect horizontal pleiotropic outliers and reported inverse-variance weighted (IVW) estimates before and after outlier removal ( 18 ). As a prespecified sensitivity analysis within the same two-sample MR pipeline, we filtered AF instruments using PhenoScanner version 2 tools ( 19 ) to remove variants associated ( P < 5 × 10 −8 ) with potential confounders (e.g., BMI and body composition traits), retaining 51 SNPs for reanalysis. We performed leave-one-out analyses, Cochran’s Q for heterogeneity, and MR-Egger intercept tests for directional pleiotropy. We applied the MR-Steiger test to verify that genetic instruments explain more variance in AF than in ovarian cancer, supporting the AF to cancer direction. MR treasure your exceptions (MR-TRYX) was applied as a sensitivity analysis to detect pleiotropic outliers using radial MR ( 20 ). Outlier-associated traits were retrieved from the MR-Base catalogue, prioritized with least absolute shrinkage and selection operator (LASSO), and then incorporated into reestimation under three settings: (i) remove all outliers, (ii) remove only candidate-trait outliers, and (iii) multivariable adjustment for candidate traits. To assess robustness to instrument choice, we repeated the primary pipeline using an independent AF GWAS (ebi-a-GCST006061) as sample 1 exposure instruments and reestimated effects on the same OCAC outcomes. Multivariable MR (MVMR) was employed to address shared risk factors. Specifically, BMI, childhood BMI, and class II obesity ( 21 , 22 ) were included in our MVMR analysis to calculate estimates separately from the effects of potential confounding variables. To efficiently identify genetic variants with subtle effect sizes ( 23 ), we applied meta-analysis to AF and four subgroups of ovarian cancer using a sample size–based analytic model with METAL ( 24 ). METAL combines P values from multiple studies, considering study-specific weights based on sample size and the direction of genetic effects. To identify potential SNPs, we employed two methods by FUMA (version 1.3.6; ref. 25 ): positional mapping and eQTL mapping. Positional mapping utilized a default gene window of 10 kb on both sides, based on ANNOVAR ( 26 ) annotations. For cis-eQTL mapping, we utilized two datasets: Genotype-Tissue Expression version 8 data covering various tissues ( 27 ) and eQTLGen cis-eQTLs. Default settings in FUMA were employed for both mapping approaches. Gene biotypes were retrieved from Ensembl BioMart (Ensembl build 92, https://mart.ensembl.org/info/data/biomart/index.html ). Functional enrichment analyses were conducted using hypergeometric tests with pathway and functional gene set information from MSigDB version 7.0. We employed MAGMA (version 1.08; ref. 28 ) within the FUMA pipeline for gene-based analyses. A gene annotation window of 10 kb was used. SNPs were mapped to 19,383 genes obtained from Ensembl build 92 GRCh37. Tissue expression (gene property) analysis was conducted using Genotype-Tissue Expression version 8 ( 27 ) RNA sequencing (RNA-seq) data to assess the tissue specificity of gene expression. We conducted SMR analysis using cis-eQTL as IVs, gene expression as exposure, and meta-analysis results of AF and ovarian cancers as outcomes. This approach simultaneously investigated potential pleiotropic associations between gene expression and outcome GWAS data by analyzing summarized GWAS data and eQTL data from separate datasets. Further details on the SMR method can be found in Zhu and colleagues’ work in 2016 ( 29 ). To assess the presence of linkage in the observed associations, we conducted the Heterogeneity in Dependent Instruments test. A P HEIDI (where HEIDI indicates Heterogeneity in Dependent Instruments) value of <0.05 would reject the null hypothesis, suggesting that the observed association might be due to two distinct genetic variants in high linkage disequilibrium. We used default settings in SMR, including criteria such as P eQTL 0.01, and employed FDR adjustment to address multiple testing. RNA-seq expression, comprising level 3 data from Illumina Genome Analyzer and HiSeq platforms, along with clinical data for patients with ovarian cancer, was obtained from The Cancer Genome Atlas (TCGA) data portal ( http://cancergenome.nih.gov/ ). RNA-seq by Expectation-Maximization expression values were employed for statistical analysis. Differences in overall survival between groups with “high” and “low” expression (categorized by the median value of gene expression) were evaluated using Kaplan–Meier curves. P values were calculated using the log-rank test implemented in the Survival package in R. The human ovarian cancer cell line SKOV3 was obtained from the ATCC and cultured in DMEM supplemented with 10% FBS (Gibco) and 1% penicillin–streptomycin (HyClone) in an incubator at 37°C with 5% CO 2 . SKOV3 cells (10 5 /well) were cultured in a six-well plate before transfection. When the confluence reached 30% to 50%, transfection of the NUP50 and SYTL2 siRNAs was performed using the Rfect siRNA/miRNA Transfection Reagent (Baidai Biotechnology) according to the manufacturer’s recommendations. The amount of siRNAs was 0.5 nmol. The cells were used for future analysis 48 hours after transfection. The NUP50 , SYTL2 , and negative control (NC) siRNAs were synthesized by Accurate Biotechnology (Hunan) Co., Ltd. The sequence of the NUP50 siRNA was 5′-CCA UGU UGA UUC GGG UAA ATT-3′ (sense) and 5′- UUU ACC CGA AUC AAC AUG GTT-3′ (antisense), and the sequence of the SYTL2 siRNA was 5′-ACU UUU ACC UGU UCC AAA GCC-3′ (sense) and 5′-CUU UGG AAC AGG UAA AAG UUA-3′ (antisense). The Cell Counting Kit-8 (CCK-8) assay (DOJINDO Molecular Technologies) was used to assess cell proliferation, following the manufacturer’s instructions. Transfected cells were seeded into 96-well plates at 2 × 10 3 cells/well. The CCK-8 reagent (10 μL) was added to the wells after 24 hours and incubated for 2 hours, followed by the measurement of optical density (OD) at 450 nm using a microplate spectrophotometer (Bio-Tek). According to the manufacturer’s protocol, total cellular RNA was extracted using TRIzol (Invitrogen) and stored at −80°C. For gene detection, cDNA was generated using the RT-PCR kit (Takara Bio, Inc.), following the manufacturer’s instructions. The RT conditions were 37°C for 15 minutes, then 85°C for 5 seconds. The SYBR-Green PCR master mix [Accurate Biotechnology (Hunan) Co., Ltd.] was used for qPCR on a 7900HT Fast RT-PCR instrument (Applied Biosystems; Thermo Fisher Scientific). The amplification protocol was 3 minutes at 95°C, followed by 40 cycles at 95°C for 3 seconds and 60°C for 30 seconds. The NUP50 and SYTL2 mRNA levels were normalized to the GAPDH mRNA levels using the 2 −ΔΔCt method. The primer sequences were as follows: SYTL2 : 5′-TAT GGT GTA TGA TGG GTT CAG GC-3′ (F) and 5′-GTA​GAG​TCC​ATC​CAG​TCC​ACT​TC-3′ (R); NUP50 5′-ACG TTC TTA TCG TCT GTG TTC CA-3′ (F) and 5′-GTG TTC AGG CAT CCT TTT TCT CC-3′ (R); and GAPDH : 5′-GCA CCG TCA AGG CTG AGA AC-3′ (F) and 5′-TGG TGA AGA CGC CAG TGG A-3′ (R). Cells were washed with PBS twice and lysed in radioimmunoprecipitation assay lysis buffer (70 μL/well, Beyotime) after 48 to 72 hours of transfection. Protein concentration was determined using the bicinchoninic acid protein assay (Beyotime). Equal amounts of protein (25–50 μg) were loaded onto an 8% or 10% SDS-PAGE gel for electrophoresis. The proteins were transferred to nitrocellulose membranes (Sangon Biotech). The membranes were blocked at room temperature in 5% skimmed milk diluted with PBS plus Tween 20 (PBST) for 1 hour and hybridized with the primary antibodies overnight at 4°C. The membranes were washed with PBST three times and incubated with the appropriate secondary antibodies for 1 hour. The membranes were washed with PBST three times, and the protein bands were visualized using an Odyssey Scanner (LI-COR Biosciences). The antibodies were as follows: GAPDH (1:10,000, ab181602, Abcam), NUP50 (1:2,000, 20798-1-AP, Proteintech), and Slp2 ( SYTL2 , 1:2000, sc-393847, Santa Cruz Biotechnology). Data were analyzed using SPSS 20.0 (IBM) and GraphPad Prism software 5.0 (GraphPad Prism Software Inc.). All data were from at least three independent experiments and are presented as means ± SD. The Student t test was used for comparison, and P < 0.05 was considered to be statistically significant. This research has been conducted using the UKB Resource under project number 134960. The UKB research protocol and study design were approved by the National Health Service National Research Ethics Service, and all study participants provided written informed consent. Ethical approval was obtained from the Northwest Centre for Research Ethics Committee (11/NW/0382). In Scotland, the UKB has approval from the Community Health Index Advisory Group. Our study was conducted in accordance with the Declarations of Helsinki.

Supplementary methods, table title, and figure legends. Supplementary Table 1. Summary of the GWAS used in this MR study. Supplementary Table 2. Schoenfeld residual test for proportional hazards assumption in both AF to OCA and OCA to AF models. This table summarizes the results of the proportional hazards assumption tests based on Schoenfeld residuals (cox.zph function in R) for both the AF to OCA and OCA to AF Cox regression models. For each covariate, the chi-square statistic, degrees of freedom (df), and corresponding P-value are presented. A significant P-value (<0.05) indicates potential deviation from the proportional hazards assumption. Supplementary Table 3. The causal relationship between AF and 4 ovarian cancers. This table demonstrated the MR results of AF (ebi-GCST006414) and 4 ovarian cancers using different methods and sensitivity analysis result. a ‘All’ represents analyses without removing pleiotropic SNPs; ‘Removed’ represents analyses after removing pleiotropic SNPs when MR-PRESSO global test showed significant P value. (P<0.05) Supplementary Table 4. The causal relationship between ovarian cancer and AF. This table demonstrated the MR results of ovarian cancer and high grade and low grade serous ovarian cancer and 2 AF dataset (ebi-a-GCST006061, ebi-GCST006414) using different methods and sensitivity analysis result. Supplementary Table 5. Exploratory Comparison of Observational and MR Estimates Supplementary Table 6. The causal relationship between AF and 4 ovarian cancers using validation AF dataset (ebi-GCST006061). This table demonstrated the MR result of AF (ebi-GCST006061, Validation dataset) as exposure and 4 ovarian cancers as outcome using Inverse variance weighted method and sensitivity analysis result. a ‘All’ represents analyses without removing pleiotropic SNPs; ‘Removed’ represents analyses after removing pleiotropic SNPs when MR-PRESSO global test showed significant P value. (P<0.05) This table depicted the MR results of AF and ovarian cancer using inverse variance weighted method and sensitivity analysis result after utilizing PhenoScanner to exclude potential confounder SNPs. Supplementary Table 8. The result of MR-Steiger test. Supplementary Table 9 Supplementary Table 10. Sensitivity analysis of MR-TRYX to validate the causal relationship between AF and ovarian cancer MR-TRYX includes IVW with ‘modified 2nd order weighting’ in three conditions: (1) Outlier removed (all): remove all outliers detected by radial regression; (2) Outlier removed (candidates): remove outliers which are strongly associated with candidate traits; (3) Outliers adjusted: not remove any outliers but perform multivariable MR adjusting for these outliers. Supplementary Table 11. Selection of common risk factors for AF and ovarian cancer prioritized on LASSO regression. Supplementary Table 12. SMR result of 5 high-risk mutual genes of both AF and ovarian cancer (presented in both FUMA analysis and SMR analysis using CAGE eQTL data). PeQTL is the p-value of the top associated cis-eQTL in the eQTL analysis; PGWAS is the p-value for the top associated cis-eQTL in the GWAS analysis; Beta is the estimated effect size in SMR analysis; SE is the corresponding standard error. Bold font means statistical significance after correction for multiple testing using FDR. Supplementary Figure 1. Estimated positive causal effects of genetically increased AF risk on 4 ovarian cancers risk. Forest plot representing the causal estimation of AF risk on (A) ovarian cancer, (B) endometrioid ovarian cancer, (C) low grade serous ovarian cancer, (D) high grade and low grade serous ovarian cancer risk using each or all variants. Red lines are the causal effect of exposure on outcome is estimated using all SNPs using vary methods. Supplementary Figure 2. Forest plots showing Multivariable Mendelian randomization result estimates for the association between AF and ovarian cancer using inverse variance weighted method. Forest plots showing multivariable Mendelian randomization (MVMR) estimates for the association between AF and ovarian cancer using the inverse variance weighted method. Abbreviations: BMI, body mass index. Supplementary Figure 3. Discovery of high-risk mutual genes expression of both AF and OCA. (A) Flowchart of discovering high-risk genes expression of both AF and OCA. (B) Summary of SNPs and mapped genes for AF and 4 OCAs using FUMA. (C) Tissue expression enrichment in GTEx general tissue types. The dashed line indicates the significance threshold at P < 0.001. (D) Bar-dot plots show significantly enriched gene sets annotation genes. (E) SMR analysis identifying genes with shared genetic associations between AF and OCA. Abbreviation: AF, atrial fibrillation; OCA, ovarian cancer; GWAS, genome-wide association studies; GTEx, Genotype-Tissue Expression; MR, mendelian randomization; NUP50, Nucleoporin 50; SMR, summary data-based mendelian randomization; SNP, single nucleotide polymorphisms; SYTL2, Synaptotagmin-like 2. Supplementary Figure 4. Uncropped blot of Figure 5D. Full-length blot corresponding to Figure 5D in the main text.