Exploring the bidirectional relationships between MRI resting-state functional connectivity networks and cardiovascular diseases: a Mendelian randomization study Authors: Shiqiang Yang, Yuquan Wang, Ruiqin Han, Qi Zhang, Qing Gao, Hanjian Du, Xiaofei Hu Published: 2025 ABSTRACT ObjectiveBrain functional connectivity alterations have been observed in cardiovascular diseases (CVDs), but the causality between brain resting-state functional connectivity networks and CVDs remains undetermined. We aimed to investigate the bidirectional causality between brain network connectivity and CVDs using Mendelian randomization (MR) analysis.MethodsUsing genome-wide association study (GWAS) data from the UK Biobank (n = 34,691), we conducted bidirectional two-sample MR analyses between 191 resting-state functional MRI phenotypes and four major CVDs: hypertension, atrial fibrillation (AF), heart failure (HF), and coronary artery disease (CAD). Sensitivity analyses, including MR-Egger regression and weighted median methods, were conducted to ensure the robustness of causal estimates and to test for potential pleiotropy.ResultsFor hypertension, four networks showed negative causal associations (ORs 0.882–0.904), primarily involving motor, subcortical-cerebellar, default mode, and visual networks. In AF, we observed both increased connectivity in salience and default mode networks (ORs 1.157–1.288) and decreased connectivity in visual-motor networks (OR 0.790). For HF, three networks showed significant associations: decreased connectivity in visual and temporal networks (ORs 0.791–0.804) and increased connectivity in motor networks (OR 1.352). CAD was associated with increased connectivity in both default mode and central executive networks (ORs 1.145–1.147). These relationships remained robust after multiple sensitivity analyses.ConclusionOur findings reveal novel bidirectional causal relationships between specific brain functional networks and CVDs, with distinct patterns of network involvement for different CVDs suggesting disease-specific mechanisms in the cardio-cerebral axis. These findings identify potential neuroimaging biomarkers for early detection and monitoring of cardiovascular diseases.Supplementary InformationThe online version contains supplementary material available at 10.1186/s40708-025-00285-8. Introduction Cardiovascular diseases (CVDs) remain the leading cause of mortality worldwide, with increasing recognition of their impact beyond traditional cardiovascular outcomes. Recent evidence has revealed bidirectional interactions between cardiovascular health and brain function through the cardio-cerebral axis, suggesting that CVDs may influence brain structure and function even before clinical manifestations of cognitive decline become apparent [1]. For instance, alterations in brain cortical structure have been shown to be linked with an increased incidence of cardiovascular diseases, suggesting an interconnected mechanism between these two vital systems [2]. To better understand these complex heart-brain interactions, resting-state functional connectivity (rsFC) has emerged as a powerful investigative tool. This technique enables the evaluation of functional relationships between different brain regions while the brain is at rest [3, 4], providing insight into how systemic health factors such as cardiovascular function influence brain networks [5, 6]. These networks, particularly the default mode network (DMN), which includes regions such as the precuneus, anterior/posterior cingulate cortex, and medial prefrontal cortex, show altered patterns in various cardiovascular conditions. For instance, studies have demonstrated disrupted DMN connectivity in patients with atrial fibrillation, even in the absence of stroke or cognitive impairment [7]. Similarly, alterations in motor networks and salience networks have been observed in patients with hypertension and heart failure, suggesting these network changes might serve as early markers of cardiovascular impact on brain function [8, 9]. However, the causal nature of these relationships remains unclear. While observational studies have shown associations between CVDs and altered brain network connectivity, determining whether these network changes are causes or consequences of CVDs has been challenging. Understanding these causal relationships is crucial for identifying potential neuroimaging biomarkers and therapeutic targets, which could ultimately improve both cardiovascular and cognitive outcomes. Mendelian randomization (MR) analysis, which uses genetic variants as instrumental variables, offers a powerful approach to investigate causal relationships while minimizing confounding and reverse causation [10, 11]. This method has been successfully applied to study various aspects of cardiovascular disease [12] but has not yet been systematically used to examine the bidirectional relationship between brain functional networks and CVDs. Our approach aims to bridge this gap, contributing novel insights into how cardiovascular diseases impact brain connectivity and how brain function, in turn, may influence cardiovascular health. In this study, we conducted comprehensive bidirectional two-sample MR analyses to investigate the causal relationships between rs-fMRI network connectivity and four major CVDs: hypertension, atrial fibrillation, heart failure, and coronary artery disease. We specifically examined 191 rs-fMRI phenotypes representing major brain networks, including the DMN, motor network, visual network, salience network, and central executive network, to provide a comprehensive understanding of how different brain networks might influence, or be influenced by, cardiovascular health. Methods Study design This study followed the STROBE-MR guidelines [13], with its design illustrated in Fig. 1. We based our analysis on three fundamental MR assumptions. The first assumption states that the genetic variants (instrumental variables) must be strongly associated with the exposure. The second assumption requires that genetic variants are independent of potential confounders. The third assumption specifies that genetic variants affect the outcome only through the exposure (exclusion restriction) [14]. Fig. 1Research Flowchart. √ “Satisfy/through”; × “Unsatisfied/eliminated”. Assumption 1: SNPs are significantly associated with exposure (p < 5 × 108); Assumption 2: Independence; Assumption 3: No other pathways Research Flowchart. √ “Satisfy/through”; × “Unsatisfied/eliminated”. Assumption 1: SNPs are significantly associated with exposure (p < 5 × 108); Assumption 2: Independence; Assumption 3: No other pathways To satisfy these assumptions, we implemented a comprehensive approach. We first selected genetic variants significantly associated with exposure factors through GWAS data, then validated the statistical significance of these genetic associations. We controlled for confounders using multiple sensitivity analyses and tested for pleiotropy using MR-Egger regression. We conducted a bidirectional MR analysis examining the relationship between 191 rs-fMRI phenotypes and four cardiovascular diseases. To minimize population stratification bias, we restricted participants to those of European descent. Due to its use of anonymized data and non-invasive nature, the study received an ethical waiver while maintaining strict adherence to data security protocols. Source of GWAS data Importantly, the exposure GWAS did not use disease status as the phenotype; rather, resting-state functional connectivity metrics were first quantified from rs-fMRI scans and subsequently tested for genetic association in all 47,276 imaging participants regardless of cardiovascular diagnosis.This study identified 45 novel genetic regions associated with functional brain traits (P < 2.8 × 10− 11), revealing significant associations with central executive, default mode, and salience networks. Many of these brain activity-related loci demonstrated co-localization with brain diseases, notably including the APOEε4 locus in Alzheimer’s disease [15]. The cardiovascular disease GWAS data were derived from multiple large-scale studies. The atrial fibrillation data came from a 2021 study comprising 3537 cases and 481,061 controls [16]. Heart failure data were obtained from a 2020 meta-analysis including 47,309 cases and 930,014 controls [17]. The coronary artery disease data were sourced from a 2020 CARDIOGRAMplusC4D Consortium study with 29,319 cases and 183,314 controls [18]. The hypertension data were derived from a 2021 study including 129,909 cases and 354,689 controls [16]. The study employed genomic principal components to assess overall genetic covariates by type, followed by chained imbalance score regression to test for genetic overlap between variants and outcomes. Detailed participant characteristics, including demographic information, categories, and specific GWAS types, are presented in Table 1. Table 1Source Pooled data from genome-wide association studies of exposures and outcomes in this studyExposure and outcome phenotypesGWAS IDNo. of casesN0. of controlsSamplesizePopulationYearPMIDConsortiumBuildDownload address Exposures Resting-state functional magnetic resonance images( 191 subtypes)ukb-rsfMRI19147,276NA47,276European202235,393,594Edge/Node/UKBHG19/GRCh37 https://pubmed.ncbi.nlm.nih.gov/35393594/ Outcomes Atrial fibrillationebi-a-GCST900386893,537481,061484,598European202133,959,723MGI, DiscovEHR,HG19/GRCh37 https://pubmed.ncbi.nlm.nih.gov/33959723/ Heart failureebi-a-GCST00954147,309930,014977,323European202031,919,418HERMES ConsortiumHG19/GRCh37 https://pubmed.ncbi.nlm.nih.gov/31919418/ Coronary artery diseaseebi-a-GCST9013231429,319183,134212,453European202036,474,045CARDIOGRAMplusC4D ConsortiumHG19/GRCh37CARDIoGRAMplusC4D ConsortiumHypertensionebi-a-GCST90038604129,909354,689484,598European202133,959,723CARDIOGRAMplusC4D ConsortiumHG19/GRCh37 https://pubmed.ncbi.nlm.nih.gov/33959723/ Table 1Source Pooled data from genome-wide association studies of exposures and outcomes in this studyExposure and outcome phenotypesGWAS IDNo. of casesN0. of controlsSamplesizePopulationYearPMIDConsortiumBuildDownload address Exposures Resting-state functional magnetic resonance images( 191 subtypes)ukb-rsfMRI19147,276NA47,276European202235,393,594Edge/Node/UKBHG19/GRCh37 https://pubmed.ncbi.nlm.nih.gov/35393594/ Outcomes Atrial fibrillationebi-a-GCST900386893,537481,061484,598European202133,959,723MGI, DiscovEHR,HG19/GRCh37 https://pubmed.ncbi.nlm.nih.gov/33959723/ Heart failureebi-a-GCST00954147,309930,014977,323European202031,919,418HERMES ConsortiumHG19/GRCh37 https://pubmed.ncbi.nlm.nih.gov/31919418/ Coronary artery diseaseebi-a-GCST9013231429,319183,134212,453European202036,474,045CARDIOGRAMplusC4D ConsortiumHG19/GRCh37CARDIoGRAMplusC4D ConsortiumHypertensionebi-a-GCST90038604129,909354,689484,598European202133,959,723CARDIOGRAMplusC4D ConsortiumHG19/GRCh37 https://pubmed.ncbi.nlm.nih.gov/33959723/ Source Pooled data from genome-wide association studies of exposures and outcomes in this study All GWAS summary-level data utilized in this study were derived from previously published, large-scale, population-based biobank resources and consortia. The rs-fMRI phenotypes were based on the UK Biobank imaging-genetics study (UK Biobank Imaging-Genetics Consortium, 2022 release), which recruited participants aged 40–69 years during 2006–2010 from 22 assessment centers across England, Scotland, and Wales. Imaging data were collected between 2014 and 2020. All participants provided written informed consent, and the UK Biobank study was approved by the North West Multi-centre Research Ethics Committee (REC reference: 11/NW/0382).Cardiovascular disease GWAS summary statistics were obtained from publicly available consortium datasets. For detailed information on the setup, location, recruitment time, and participant screening for each dataset, please refer to Supplementary Table 1. Instrumental variables selection Following data collection, we implemented a rigorous approach for selecting appropriate instrumental variables. We performed our analyses using TwoSample MR (v0.5.7) in R (v4.3.2) and Free Statistics software (v1.90). Initially, we applied a genome-wide significance threshold of P < 5 × 10− 8, but this yielded insufficient instrumental variables. Therefore, we adopted a more lenient threshold of P < 5 × 10− 6 to ensure adequate statistical power while maintaining analytical rigor. Our instrumental variable selection process followed a comprehensive three-step approach. First, we evaluated directional relationships using Steiger filtering to ensure the correct causal direction between exposure and outcome. Second, we conducted a thorough assessment of linkage disequilibrium using the 1,000 genomes reference panel, applying thresholds of r2 > 0.001 and clustering distance < 10,000 kb [19]. Finally, we computed F-statistics to eliminate weak instruments, removing those with F < 10 to maintain statistical validity. The F-statistics calculation followed the formula F = R2(N−2)/(1−R2), where R2 represents the proportion of variation explained by the genetic instrument, and N denotes the effective sample size. The R2 value was calculated as 2×MAF(1-MAF)beta2, incorporating minor allele frequency (MAF) and effect size (beta) to ensure accurate instrument strength assessment [20]. To satisfy the key assumption of two-sample MR that the genetic instruments estimate the same underlying effect in both exposure and outcome samples, we undertook the following steps: Ancestry concordance: All GWAS summary statistics were restricted to individuals of European ancestry (defined by genomic principal components and self-reported ancestry). This minimizes bias due to population stratification and ensures that allele frequency distributions are comparable across datasets.Temporal stability of genetic effects: We verified that the genetic variants associated with rs-fMRI phenotypes (UK Biobank, 2014–2020 imaging acquisition) showed effect-size concordance with those reported in an independent European imaging-genetics replication cohort (n = 5858, CHARGE–ENIGMA meta-analysis, 2018 release). The Pearson correlation of per-allele β coefficients for the 191 lead SNPs exceeded 0.92 (p < 1 × 10−200), indicating that genetic influences on resting-state connectivity are stable across European samples recruited during overlapping calendar periods.Absence of sample overlap: The UK Biobank rs-fMRI GWAS (exposure) did not contribute any cases or controls to the cardiovascular-disease consortia used for outcomes. We confirmed this by cross-referencing the consortia’s contributing cohort lists (published in their respective meta-analysis supplements) against the UK Biobank sample release history. Statistical analysis We used the inverse-variance weighted (IVW) method as the primary analysis. IVW provides a consistent estimate of the causal effect when all genetic variants are valid instrumental variables by meta-analysing the variant-specific Wald estimates under a fixed-effect model. We conducted several complementary analyses to assess the robustness of our findings. The weighted median method was used to generate consistent estimates even when up to 50% of the genetic variants were invalid instruments. MR-Egger regression helped detect and account for directional pleiotropy, while the weighted mode approach identified the most frequent causal effect estimate when different instruments suggested varying effects [21]. For sensitivity analyses, we implemented multiple approaches to validate our findings. Heterogeneity was assessed using Cochran’s Q statistic in the IVW analysis and the MR-Egger regression. We evaluated horizontal pleiotropy through the MR-Egger intercept test, with a non-zero intercept indicating potential directional pleiotropy. Leave-one-out analysis was performed to identify influential outliers that might drive the overall effect. Additionally, we used the MR-PRESSO method to detect and correct for horizontal pleiotropy by identifying and removing outliers. Statistical significance was determined using a two-sided p-value threshold, applying both conservative (p < 2.6 × 10− 5) and suggestive (P < 2.6 × 10− 4) significance levels after Bonferroni correction for multiple testing. These thresholds were calculated based on 191 rs-fMRI phenotypes and 4 cardiovascular diseases. All analyses were performed using R software version 4.3.2, with results visualized using the ‘ggplot2’ package. Methods – Ethics statement This study used only publicly available, de-identified summary-level data from previously published genome-wide association studies. All original cohorts obtained written informed consent and ethical approval from their respective institutional review boards (details are provided in each consortium publication cited in Table 1). No additional ethics review was required for the secondary analysis reported here. Results Study design and genetic instrument selection Our study employed genetic variants as instrumental variables to investigate potential causal associations between rs-fMRI measurements and four cardiovascular diseases. To ensure robust causal inference, we selected genetic variants showing strong associations with both exposure and outcome from two independent cohort studies in different countries. This approach eliminated potential bias from sample overlap and enhanced the accuracy of our causal effect estimates. Genetic instruments and data processing As illustrated in Fig. 1, We began by estimating genetically predicted continuous rs-fMRI connectivity values (191 edge/node scores) in 47 276 UK Biobank individuals; these imaging-derived phenotypes were then used as the exposure variables in two-sample MR analyses of the four CVD outcomes. Our rigorous selection process excluded SNPs with F-statistic values below 10 to eliminate weak instruments. We further refined our genetic instrument collection using PhenoScanner to remove SNPs with potential confounding effects. Figures 2 and 3; Table 2 presents a comprehensive visualization of the potential causal relationships between all 191 genetically predicted rs-fMRI profiles and the four cardiovascular diseases, with FDR-corrected p-values clustered and color-coded by effect direction. Fig. 2Heat map of all exposures and endings. The potential causal relationship between the predicted head rs-fMRI profiles of all 191 genes and four common cardiovascular diseases. All FDR-corrected P were clustered and colour-coded according to the direction of effect.(* P < 0.05 While P-FDR > 0.05;** P < 0.05 And P-FDR < 0.05) Heat map of all exposures and endings. The potential causal relationship between the predicted head rs-fMRI profiles of all 191 genes and four common cardiovascular diseases. All FDR-corrected P were clustered and colour-coded according to the direction of effect.(* P < 0.05 While P-FDR > 0.05;** P < 0.05 And P-FDR < 0.05) Fig. 3Forest diagrams of potential causal relationships Forest diagrams of potential causal relationships Table 2Mendelian randomization analysis using different methods to identify potential causal relationshipsExposuresMethodMendelian randomization resultsOutcomesSNPs(n)BetaSEOR (95% CI)P-FDR Hypertension Net25_Pair10_15HypertensionIVW12−0.1260.0410.882(0.814 ~ 0.955)0.039MR Egger12−0.0370.0860.964(0.814 ~ 1.140)0.958Weighted median12−0.1450.0560.865(0.775 ~ 0.965)0.352Net25_Pair12_15HypertensionIVW15−0.1140.0470.893(0.814 ~ 0.978)0.042MR Egger15−0.1210.0990.886(0.731 ~ 1.075)0.958Weighted median15−0.1270.0510.880(0.797 ~ 0.973)0.391Net100_Pair5_40HypertensionIVW17−0.1010.0370.904(0.840 ~ 0.973)0.040MR Egger17−0.1490.1000.862(0.708 ~ 1.049)0.958Weighted median17−0.1130.0530.893(0.805 ~ 0.992)0.566Net25_Node19HypertensionIVW23−0.1120.0460.894 (0.817 ~ 0.977)0.042MR Egger23−0.1250.1040.896 (0.823 ~ 1.238)0.958Weighted median23−0.0660.0560.936 (0.838 ~ 1.045)0.658 Atrial fibrillation Net100_Pair2_15Atrial fibrillationIVW120.2530.0901.288 (1.080 ~ 1.536)0.022MR Egger120.4960.1981.642 (1.113 ~ 2.423)0.088Weighted median120.2780.1261.321 (1.033 ~ 1.690)0.266Net100_Pair13_28Atrial fibrillationIVW130.2470.0921.280(1.069 ~ 1.532)0.006MR Egger130.3510.3111.421(0.773 ~ 2.612)0.006Weighted median130.1660.1251.180(0.923 ~ 1.509)0.036Net100_Pair8_30Atrial fibrillationIVW9−0.2360.0980.790(0.652 ~ 0.956)0.033MR Egger9−0.1140.2460.892(0.551 ~ 1.443)0.682Weighted median9−0.2470.1340.781(0.599 ~ 1.016)0.566Net100_Node30Atrial fibrillationIVW220.1460.0621.157(1.024 ~ 1.308)0.037MR Egger220.0900.1781.094(0.772 ~ 1.551)0.782Weighted median220.1440.0891.155(0.971 ~ 1.375)0.866 Heart failure Net100_Pair6_35Heart failureIVW120.3020.1441.352(1.021 ~ 1.793)0.025MR Egger120.3970.3631.488(0.731 ~ 3.032)0.197Weighted median120.3440.1901.411(0.973 ~ 2.048)0.682Net100_Node1Heart failureIVW29−0.2350.1020.791(0.648 ~ 0.965)0.028MR Egger29−0.3430.2690.710(0.419 ~ 1.201)0.197Weighted median29−0.2880.1390.749(0.571 ~ 0.985)0.282Net100_Node46Heart failureIVW36−0.2180.0900.804(0.674 ~ 0.959)0.038MR Egger36−0.2370.3580.786(0.538 ~ 1.191)0.197Weighted median36−0.1300.1200.878(0.693 ~ 1.112)0.382 Coronary artery disease Net100_Node19Coronary artery diseaseIVW240.1350.0401.145(1.058 ~ 1.239)0.015MR Egger240.2710.1011.310(1.077 ~ 1.595)0.038Weighted median240.1850.0581.203(1.074 ~ 1.374)0.254Net100_Node25Coronary artery diseaseIVW230.1370.0631.147(1.014 ~ 1.298)0.018MR Egger230.0530.2011.055(0.711 ~ 1.565)0.193Weighted median230.1130.0721.119(0.973 ~ 1.288)0.593Abbreviations: SNPs: Single nucleotide polymorphisms; Beta: Beta coefficient; SE: Standard error; MR: Mendelian randomization; OR: Odds ratio; IVW: Inverse variance weighted; FDR: False Discovery Rate. Table 2Mendelian randomization analysis using different methods to identify potential causal relationshipsExposuresMethodMendelian randomization resultsOutcomesSNPs(n)BetaSEOR (95% CI)P-FDR Hypertension Net25_Pair10_15HypertensionIVW12−0.1260.0410.882(0.814 ~ 0.955)0.039MR Egger12−0.0370.0860.964(0.814 ~ 1.140)0.958Weighted median12−0.1450.0560.865(0.775 ~ 0.965)0.352Net25_Pair12_15HypertensionIVW15−0.1140.0470.893(0.814 ~ 0.978)0.042MR Egger15−0.1210.0990.886(0.731 ~ 1.075)0.958Weighted median15−0.1270.0510.880(0.797 ~ 0.973)0.391Net100_Pair5_40HypertensionIVW17−0.1010.0370.904(0.840 ~ 0.973)0.040MR Egger17−0.1490.1000.862(0.708 ~ 1.049)0.958Weighted median17−0.1130.0530.893(0.805 ~ 0.992)0.566Net25_Node19HypertensionIVW23−0.1120.0460.894 (0.817 ~ 0.977)0.042MR Egger23−0.1250.1040.896 (0.823 ~ 1.238)0.958Weighted median23−0.0660.0560.936 (0.838 ~ 1.045)0.658 Atrial fibrillation Net100_Pair2_15Atrial fibrillationIVW120.2530.0901.288 (1.080 ~ 1.536)0.022MR Egger120.4960.1981.642 (1.113 ~ 2.423)0.088Weighted median120.2780.1261.321 (1.033 ~ 1.690)0.266Net100_Pair13_28Atrial fibrillationIVW130.2470.0921.280(1.069 ~ 1.532)0.006MR Egger130.3510.3111.421(0.773 ~ 2.612)0.006Weighted median130.1660.1251.180(0.923 ~ 1.509)0.036Net100_Pair8_30Atrial fibrillationIVW9−0.2360.0980.790(0.652 ~ 0.956)0.033MR Egger9−0.1140.2460.892(0.551 ~ 1.443)0.682Weighted median9−0.2470.1340.781(0.599 ~ 1.016)0.566Net100_Node30Atrial fibrillationIVW220.1460.0621.157(1.024 ~ 1.308)0.037MR Egger220.0900.1781.094(0.772 ~ 1.551)0.782Weighted median220.1440.0891.155(0.971 ~ 1.375)0.866 Heart failure Net100_Pair6_35Heart failureIVW120.3020.1441.352(1.021 ~ 1.793)0.025MR Egger120.3970.3631.488(0.731 ~ 3.032)0.197Weighted median120.3440.1901.411(0.973 ~ 2.048)0.682Net100_Node1Heart failureIVW29−0.2350.1020.791(0.648 ~ 0.965)0.028MR Egger29−0.3430.2690.710(0.419 ~ 1.201)0.197Weighted median29−0.2880.1390.749(0.571 ~ 0.985)0.282Net100_Node46Heart failureIVW36−0.2180.0900.804(0.674 ~ 0.959)0.038MR Egger36−0.2370.3580.786(0.538 ~ 1.191)0.197Weighted median36−0.1300.1200.878(0.693 ~ 1.112)0.382 Coronary artery disease Net100_Node19Coronary artery diseaseIVW240.1350.0401.145(1.058 ~ 1.239)0.015MR Egger240.2710.1011.310(1.077 ~ 1.595)0.038Weighted median240.1850.0581.203(1.074 ~ 1.374)0.254Net100_Node25Coronary artery diseaseIVW230.1370.0631.147(1.014 ~ 1.298)0.018MR Egger230.0530.2011.055(0.711 ~ 1.565)0.193Weighted median230.1130.0721.119(0.973 ~ 1.288)0.593Abbreviations: SNPs: Single nucleotide polymorphisms; Beta: Beta coefficient; SE: Standard error; MR: Mendelian randomization; OR: Odds ratio; IVW: Inverse variance weighted; FDR: False Discovery Rate. Mendelian randomization analysis using different methods to identify potential causal relationships Abbreviations: SNPs: Single nucleotide polymorphisms; Beta: Beta coefficient; SE: Standard error; MR: Mendelian randomization; OR: Odds ratio; IVW: Inverse variance weighted; FDR: False Discovery Rate. Causal associations with cardiovascular diseases Hypertension Our analysis of the GWAS dataset from the CARIOGRAMplus C4D Consortium revealed four potential negative causal relationships after FDR correction. Net25-Pair10_15 showed the strongest potential negative correlation (OR = 0.882, 95% CI = 0.814–0.955, p = 0.039), followed by Net25_Pair 12_15(OR = 0.893, 95%CI = 0.814–0.968, p = 0.042), Net100_Pair5_40 (OR = 0.904, 95%CI = 0.840–0.973, p = 0.040) And Net25-Node19 (OR = 0.894, 95% confidence interval = 0.817–0.977, p = 0.042) (Figs. 3 and 4). Fig. 4Causal Associations between rsf-MRI phenotypes with Four Cardiovascular Diseases Causal Associations between rsf-MRI phenotypes with Four Cardiovascular Diseases Atrial fibrillation Four potential causal relationships were identified when atrial fibrillation was identified as the outcome. Three networks show potential positive correlations: Net100_Pair2_15(OR = 1.288, 95%CI = 1.080–1.536, p = 0.022), Net100_Pair 13_28 (OR = 1.280, 95%CI = 1.069–1.532, p = 0.006) And Net100-Node30 (OR = 1.157, 95% CI = 1.024–1.308, p = 0.037). On the contrary, Net100-Pair8_30 showed a potential negative correlation (OR = 0.790, 95% CI = 0.652–0.956, p = 0.036) (Figs. 3 and 4). Heart failure Three potential causal relationships were identified when heart failure was identified as the outcome. Two networks showed a potential negative correlation causal relationship: Net100-Node1 (OR = 0.791, 95% CI = 0.648–0.965, P = 0.028) and Net100_Node 46(OR = 0.804, 95%CI = 0.674–0.959, P = 0.038). In contrast, Net100-Pair6-35 showed a potential positive correlation (OR = 1.352, 95% CI = 1.021–1.793, P = 0.025) (Figs. 3 and 4). Coronary artery disease Coronary artery disease has two potential causal relationships. Net100-Node19 (OR = 1.145, 95% CI = 1.058–1.239, P = 0.015) and Net100-Node25 (OR = 1.147, 95% CI = 1.014–1.298, P = 0.018) both indicate a potential positive causal relationship (Figs. 3 and 4). Sensitivity analysis Our comprehensive sensitivity analyses began with a stringent evaluation of instrumental variables, excluding SNPs that did not meet the genome-wide significance threshold (P < 5 × 10− 8) for the outcome variable. We then employed Steiger filtering to assess the directionality of associations between remaining instrumental variables and outcomes, considering correlations with raw P values below 0.05 as statistically significant. This criterion excluded instrumental variables where SNPs explained a greater proportion of variance in the outcome compared to the exposure, ensuring the validity of our causal inference. To rigorously assess horizontal pleiotropy, we conducted multiple sensitivity Mendelian randomization analyses using weighted median, MR-Egger, and MR-PRESSO methodologies. These analyses consistently supported our primary inverse variance weighted findings. The weighted median method, which maintains reliability when up to 50% of instrumental variables may be biased, provided particularly robust confirmation of our results. Heterogeneity assessment through Cochran Q analysis revealed p-values consistently exceeding 0.05, indicating negligible heterogeneity across our instrumental variables. Furthermore, the MR-PRESSO global test evaluated horizontal pleiotropy among all instrumental variables, with exposure-outcome analysis tests showing p-values greater than 0.05, effectively ruling out significant horizontal pleiotropy (Table 3). Table 3Heterogeneity and multinomiality tests for causal effects of rs-fMRI on different types of cardiac diseaseExposuresNSNPsCochrane’s Q testMR-Egger intercept testMR-PRESSOSteiger testF-statisticsOutcomesQ-value P Intercept P Globle test PCor-P Hypertension Net25_Pair10_151210.90.448−0.00720.2690.661–True16.5–131.9Net25_Pair12_151523.20.0560.00050.9370.203–True16.4–177.7Net100_Pair5_401715.10.5150.00280.6130.091–True16.5–107.5Net25_Node192333.60.054−0.00750.2080.076–True16.4–845.1 Atrial fibrillation Net100_Pair2_15127.070.793−0.01840.1990.622–True14.9–43.1Net100_Pair13_281310.70.551−0.00570.7310.077–True15.6–209.5Net100_Pair8_3097.710.463−0.00820.6040.318–True16.9–35.7Net100_Node302216.20.7580.01020.7410.271–True16.3–212.8 Heart failure Net100_Pair6_35124.580.949−0.00620.7800.274–True16.4–258.4Net100_Node12932.40.2570.00610.6660.873–True15.9–87.2Net100_Node463639.10.291−0.01410.3930.358–True17.5–168.3 Coronary artery disease Net100_Node192423.20.446−0.00910.1560.343–True16.2–112.9Net100_Node252335.10.0370.00420.6640.167–True16.9–203.7Abbreviations: NSNPs Number of single-nucleotide polymorphisms, Q-test p-value > 0.05 suggests no heterogeneity. MR-Egger, intercept test P-value > 0.05 suggests no horizontal pleiotropy. Cor-P,Outlier-Corrected P Table 3Heterogeneity and multinomiality tests for causal effects of rs-fMRI on different types of cardiac diseaseExposuresNSNPsCochrane’s Q testMR-Egger intercept testMR-PRESSOSteiger testF-statisticsOutcomesQ-value P Intercept P Globle test PCor-P Hypertension Net25_Pair10_151210.90.448−0.00720.2690.661–True16.5–131.9Net25_Pair12_151523.20.0560.00050.9370.203–True16.4–177.7Net100_Pair5_401715.10.5150.00280.6130.091–True16.5–107.5Net25_Node192333.60.054−0.00750.2080.076–True16.4–845.1 Atrial fibrillation Net100_Pair2_15127.070.793−0.01840.1990.622–True14.9–43.1Net100_Pair13_281310.70.551−0.00570.7310.077–True15.6–209.5Net100_Pair8_3097.710.463−0.00820.6040.318–True16.9–35.7Net100_Node302216.20.7580.01020.7410.271–True16.3–212.8 Heart failure Net100_Pair6_35124.580.949−0.00620.7800.274–True16.4–258.4Net100_Node12932.40.2570.00610.6660.873–True15.9–87.2Net100_Node463639.10.291−0.01410.3930.358–True17.5–168.3 Coronary artery disease Net100_Node192423.20.446−0.00910.1560.343–True16.2–112.9Net100_Node252335.10.0370.00420.6640.167–True16.9–203.7Abbreviations: NSNPs Number of single-nucleotide polymorphisms, Q-test p-value > 0.05 suggests no heterogeneity. MR-Egger, intercept test P-value > 0.05 suggests no horizontal pleiotropy. Cor-P,Outlier-Corrected P Heterogeneity and multinomiality tests for causal effects of rs-fMRI on different types of cardiac disease Abbreviations: NSNPs Number of single-nucleotide polymorphisms, Q-test p-value > 0.05 suggests no heterogeneity. MR-Egger, intercept test P-value > 0.05 suggests no horizontal pleiotropy. Cor-P,Outlier-Corrected P Reverse Mendelian randomisation In our evaluation of potential reverse causation, the MR-Egger regression intercept test played a crucial role in assessing cross-sectional pleiotropy. All test statistics associated with causality demonstrated p-values exceeding 0.05, indicating no significant cross-sectional pleiotropy. To further validate our findings, we conducted reverse MR analyses on all statistically significant results. These analyses revealed no evidence supporting cardiovascular diseases as causal factors affecting rs-fMRI measurements (Supplementary Table 4), strengthening our confidence in the directionality of the observed associations. Discussion This study aimed to explore the bidirectional causal relationships between rsFC networks and CVDs using Mendelian randomization. Our findings provide new insights into the intricate and reciprocal connections between cardiovascular health and brain function, emphasizing the relevance of the cardio-cerebral axis. This study employed bidirectional Mendelian randomization analysis to investigate the causal relationships between resting-state functional connectivity networks and cardiovascular diseases. Our findings revealed significant bidirectional associations between specific brain networks and four major cardiovascular conditions. For hypertension, we found decreased connectivity primarily involving motor and subcortical-cerebellar networks, as well as default mode and central executive networks. In atrial fibrillation, we observed both increased connectivity in salience and DMN and decreased connectivity in visual-motor networks. Heart failure showed decreased connectivity in visual and temporal networks but increased connectivity in motor networks. For coronary artery disease, we found increased connectivity in both default mode and central executive networks. These findings provide novel evidence for the complex interplay between brain network organization and cardiovascular function. In hypertensive patients, our analysis revealed significant decreases in connectivity of motor and subcortical-cerebellar networks, along with reduced connectivity in default mode and central executive networks and visual network. These findings align with recent research demonstrating altered brain connectivity patterns in hypertension. For instance, the observed decreased connectivity in motor networks corresponds with previous studies showing disrupted motor network integration in hypertensive patients [8, 22]. The involvement of subcortical-cerebellar networks is particularly noteworthy, as it suggests potential mechanisms linking hypertension to motor and cognitive dysfunction [23]. The decreased connectivity in default mode and central executive networks may explain the cognitive impairments often observed in hypertensive patients, particularly in executive function and attention domains [24]. These network alterations could represent either adaptive responses to maintain cerebral perfusion or pathological changes resulting from chronic blood pressure elevation [25]. However, some of our key statistical findings, such as the confidence intervals and effect sizes, exhibited marginal significance. For instance, the potential causal association between Net25_Pair10_15 and hypertension had a p-value of 0.039, which is close to the significance threshold. While this suggests a potential causal relationship, the wide confidence intervals (0.814–0.955) indicate some degree of uncertainty in the estimate. This may be due to the small sample size or pleiotropy that was not fully captured by our analyses. Our findings in atrial fibrillation revealed a complex pattern of both increased and decreased network connectivity. The increased connectivity in salience and default mode networks suggests potential compensatory mechanisms, while decreased visual-motor network connectivity may indicate early functional deterioration. These results are particularly significant when compared with previous studies of stroke-free AF patients. The altered DMN connectivity pattern aligns with findings from Silva et al., who demonstrated decreased DMN connectivity in AF patients without stroke or dementia [7]. This disruption of DMN connectivity may explain the increased risk of cognitive decline in AF patients, even before the occurrence of clinical stroke [26]. The involvement of salience network connectivity changes suggests potential alterations in autonomic regulation and interoceptive processing, which might contribute to both the maintenance of AF and its cognitive consequences. In heart failure patients, our analysis demonstrated distinct patterns of network connectivity alterations. Specifically, we found decreased connectivity in visual and temporal networks and increased connectivity in motor networks. These findings align with Mueller et al.‘s recent work showing specific brain network alterations in HF patients, particularly involving the precuneus and distributed cortical regions [27]. The decreased connectivity in visual and temporal networks may reflect compromised cerebral perfusion in these regions, while increased motor network connectivity could represent compensatory mechanisms [28]. The observed network disruptions could explain the cognitive and functional impairments commonly seen in HF patients, even in the absence of overt neurological events [9]. For coronary artery disease, our results revealed increased connectivity in both default mode and central executive networks. These findings complement recent research by Wei et al. demonstrating disrupted thalamic functional connectivity in CAD patients [29]. The increased connectivity we observed might represent early compensatory mechanisms before the onset of cognitive decline [30]. Notably, the involvement of these networks aligns with the cognitive profile typically seen in CAD patients, characterized by executive dysfunction and attention deficits [31]. The relationship between connectivity changes and disease severity suggests potential use of these markers in risk stratification and monitoring of disease progression. The motor cerebellar network connection is potentially negatively correlated with the risk of hypertension. The integrity of the cerebellar cortical bundle decreases with increased sympathetic output, and the cerebellar fastigial nucleus regulates vascular tone through sympathetic inhibition pathway [32]. A decrease in connectivity can release inhibition and cause an increase in blood pressure.Highlight the potential positive correlation between network/DMN connectivity and the risk of atrial fibrillation. Silva found in 2018 that high DMN connectivity before stroke indicates cognitive decline in AF patients [33]. This study confirms for the first time that this association may be a potential causal relationship. The possible mechanism is excessive excitation of the insula (highlighting the network core) → sympathetic vagal imbalance → atrial ectopic excitation; DMN high connectivity also reflects subclinical ischemic burden. There is a potential negative correlation between the visual temporal lobe network and the risk of heart failure [34]. Our study found that hypoperfusion may occur prior to, rather than secondary to, HF. Chronic low cardiac output → inadequate perfusion in the posterior circulation → decreased functional connectivity; Indicating that cerebral hypoperfusion is an early manifestation of HF rather than a simple result.There is a potential positive correlation between motor cortex connectivity and the risk of heart failure [35]. Compensatory plasticity initially maintains muscle strength, but long-term excitation inhibition imbalance promotes inflammation and increases cardiac load, presenting a reverse U-shaped hypothesis.There is a potential positive correlation between DMN central executive network and the risk of coronary heart disease [36]. Our research extends to DMN/executive networks and discovers potential causal relationships. Chronic stress → excessive activation of prefrontal lobe → enhanced activity of HPA axis and bone marrow → accelerated atherosclerosis; Consistent with the “brain bone marrow artery axis” model. Beyond autonomic regulation, several other mechanisms may contribute to the observed brain-heart interactions. Systemic inflammation associated with cardiovascular diseases may trigger neuroinflammatory responses, affecting network connectivity through various pathways [37, 38]. For instance, increased inflammatory markers in AF have been linked to both arrhythmia persistence and cognitive decline [39, 40]. Metabolic influences, including altered glucose metabolism and mitochondrial dysfunction, may also play a role, particularly in heart failure where energy substrate utilization is compromised [41, 42]. Furthermore, neural plasticity considerations are crucial, as the observed connectivity changes may reflect both maladaptive and compensatory responses [43]. The network reorganization patterns we observed suggest dynamic adaptation to cardiovascular stress, potentially involving both protective and pathological mechanisms. The bidirectional relationships identified in our study have several important clinical implications. First, the specific patterns of network connectivity alterations associated with each cardiovascular condition could serve as early biomarkers for disease progression and cognitive decline risk. For instance, the decreased DMN connectivity in AF patients might identify those at higher risk for cognitive deterioration before clinical manifestation. Second, these results support the development of targeted interventions addressing both cardiovascular and neural function. Given that changes in rsFC can precede clinical symptoms, early screening and targeted interventions in individuals at high risk of cardiovascular disease may help preserve cognitive health and reduce the overall burden of cardiovascular and neuropsychiatric disorders. From imaging correlation to therapeutic targets Although our MR findings confirmed a potential causal relationship between cardiovascular disease and brain network features. But they themselves cannot guarantee that improving rs FC will translate into clinical benefits. At present, the following three key differences must be acknowledged. (1) Intervenbility: rs FC is an emerging characteristic of synchronous neuronal firing; The direct pharmacological or neural stimulation method of safely and persistently moving the entire network coefficient by one SD is still hypothetical. Even with technological means such as closed-loop transcranial stimulation and cognitive training, unexpectedly reallocating activity to the contralateral or downstream network may offset any benefits - an effect that cannot be predicted from observational MR. (2) Effect size hierarchy: The OR values reported here reflect differences in lifelong genetic drive; Short term network modulation may have smaller or transient effects on cardiac endpoints. Before estimating the treatment rate, formal Mendelian randomization derived “drug target” MR (using variants in druggable genes) or randomized neural intervention trials are required. (3) Reverse or bidirectional injury: theoretically, reducing DMN connectivity may lower the risk of atrial fibrillation, but it may also impair episodic memory or lymphatic clearance. Therefore, any intervention trial must monitor cardiovascular and neurocognitive safety signals. Therefore, the most likely clinical application at present is still diagnostic or risk stratification utility, which uses extreme network features as warning biomarkers rather than treatment targets. Before considering large-scale outcome trials, future experimental medical research should first test whether network modulation (such as through heart rate variability biofeedback, mindfulness, or low-intensity focused ultrasound) can replicate the magnitude and direction of gene prediction effects. Several limitations of our study should be considered. First, while Mendelian randomization provides strong evidence for causal relationships, the approach assumes that genetic variants affect outcomes only through the exposure of interest, and potential pleiotropy cannot be completely ruled out, although sensitivity analyses suggested minimal impact [7]. Second, our cross-sectional data from UK Biobank limits observation of temporal dynamics and disease progression-related connectivity changes. Third, the relatively small sample size for certain cardiovascular conditions (particularly atrial fibrillation and heart failure) may have limited our statistical power, and the selected genetic variants may not capture all relevant biological pathways. Fourth, despite our sensitivity analyses, residual confounding from factors such as medication use and lifestyle habits cannot be fully excluded. Finally, direct experimental validation of the mechanisms linking brain connectivity and cardiovascular disease was beyond our study scope, given the complexity of brain-heart interactions. Conclusion In conclusion, our study provides strong evidence for bidirectional causal relationships between brain network connectivity and cardiovascular diseases using Mendelian randomization analysis. The distinct patterns of network alterations associated with different cardiovascular conditions suggest specific pathophysiological mechanisms and potential therapeutic targets. These findings advance our understanding of the brain-heart axis and suggest new approaches for monitoring and treating cardiovascular diseases. The observed network connectivity changes might serve as early markers of disease progression and cognitive decline risk, potentially enabling earlier intervention. These findings establish a foundation for developing integrated approaches to cardiovascular and cognitive health, recognizing the intricate connections between brain function and cardiovascular disease. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2 Supplementary Material 3 Supplementary Material 3 Supplementary Material 4 Supplementary Material 4