Heart-brain connections: phenotypic and genetic insights from 40,000 cardiac and brain magnetic resonance images

Phenotypic heart-brain connections

To verify that the 82 CMR traits are well-defined and biologically meaningful, we first examined their reproducibility using the repeat scans obtained from the UKB repeat imaging visit (n = 2,903, average time between visits = 2 years). For each trait, we calculated the correlation between two observations from all revisited individuals. The average reproducibility was 0.652 (range = (0.369, 0.970), Table S1). A few volumetric traits had very high reproducibility (> 0.9), including the LV end-diastolic volume (LVEDV), LV myocardial mass (LVM), RV end-diastolic volume (RVEDV), RV end-systolic volume (RVESV), AAo maximum area, AAo minimum area, DAo maximum area, and DAo minimum area. The ejection fraction (such as the LV ejection fraction (LVEF)) and distensibility traits (such as the descending aorta distensibility (DAo aorta distensibility)) had the lowest reproducibility among all volumetric traits (mean = 0.573 and 0.537, respectively). In addition, the average reproducibility was 0.758 for the 17 wall thickness traits, 0.530 for the 7 longitudinal peak strains, 0.568 for the 17 circumferential strains, and 0.514 for the 17 radial strains. Overall, these results suggest that the extracted CMR traits have moderate to high within subject reliability and can consistently annotate the cardiac and aortic structure and function.

With the control of a large number of covariates, we examined the associations between the CMR traits and brain MRI traits in UKB individuals of white British ancestry (n = 31,875, Methods). Particularly, we adjusted for body size (height and weight) and body mass index in all our analyses. At the Bonferroni significance level (P < 1.33 × 10^-6), CMR traits were associated with a wide variety of brain MRI traits, including regional brain volumes, cortical thickness, DTI parameters, and resting and task fMRI traits (Fig. 2A and Fig. S2). For example, the total brain volume had strong positive associations with volumetric measures of right and left heart as well as aortic sections, with the top ranked traits being the AAo maximum area, RA stroke volume (RASV), DAo minimum area, RVESV, RVEDV, and right atrium maximum volume (RAV max) (β > 0.112, P < 1.60 × 10^-61, Fig. S3). These volumetric traits were also widely associated with regional brain volumes, after adjusting for the total brain volume. A large proportion of these associations were positive, and the negative ones were specifically observed in the cerebrospinal fluid (CSF) and ventricle volumes (e.g., the lateral ventricle and third ventricle). The wall thickness traits were positively associated with subcortical structures (e.g., the putamen, caudate, and hippocampus) and negatively associated with the lateral occipital and rostral middle frontal. In addition, cortical thickness measures were associated with multiple CMR traits, and the strongest associations were observed in the left ventricular cardiac output (LVCO) and DAo aorta distensibility (Fig. S4). These results uncover that heart structure and function have close relationships with basic brain morphometry.

• Download figure
• Open in new tab

Fig. 2 Phenotypic heart-brain associations.

(A) The -log10(pvalue) of phenotypic correlations between 82 CMR traits and 5 groups of brain MRI traits, including 101 regional brain volumes, 63 cortical thickness traits, 110 DTI parameters, 92 resting fMRI traits, and 92 task fMRI traits. The dashed line indicates the Bonferroni-significance level (P < 1.33 × 10^-6). Each CMR trait category is labeled with a different color. (B) The -log10(pvalue) of phenotype correlations between 82 CMR traits and 8,531 resting fMRI functional connectivity traits across 12 functional networks. The dashed line indicates the Bonferroni-significance level (P < 7.15 × 10^-8). Functional networks are labeled with different colors. (C) Significant correlations between resting functional connectivity traits and left ventricular myocardial mass (LVM, left) and right ventricular ejection fraction (RVEF, right). AAo, ascending aorta; DAo, descending aorta; LA, left atrium; LV, left ventricle; RA, right atrium; RV, right ventricle; Visual1, the primary visual network; and Visual2, the secondary visual network.

CMR traits were also strongly correlated with brain structural and functional connectivity. For example, fractional anisotropy (FA) was a robust measure of brain structural connectivity and white matter microstructure. The average FA was positively associated with the LV ejection fraction (LVEF), DAo aorta distensibility, and AAo aorta distensibility (β > 0.035, P < 1.69 × 10^-8, Fig. S5), and had negative correlations with many other CMR traits, including the LVM, LVEDV, left ventricular end-systolic volume (LVESV), global wall thickness, AAo maximum area, and DAo maximum area. For resting fMRI, both mean functional connectivity and mean amplitude (functional activity) were negatively associated with volumetric measures of 4 cardiac chambers, such as the LVCO, RV ejection fraction (RVEF), LA stroke volume (LASV), and RA ejection fraction (RAEF) (β < -0.062, P < 4.87 × 10^-24, Fig. S6). Positive correlations were mainly located in the AAo aorta distensibility, DAo aorta distensibility, longitudinal strain, and peak circumferential strain. Similar patterns were observed for task fMRI traits (Fig. S7).

Brain functional traits could be more directly related to human behavioral and cognitive differences than brain structural traits⁴⁵. To further discover fine details of CMR-connections with brain functions, we examined pairwise associations between 82 CMR traits and 64,620 high-resolution functional connectivity traits⁴² in resting fMRI. Significant associations (P < 7.15 × 10^-8) were observed across the functional connectivity of the whole brain, with enrichment in specific brain functional areas and networks (Fig. 2B and Fig. S8). For example, the somatomotor network and its connectivity with the secondary visual network had strong associations with multiple CMR traits. Specifically, positive somatomotor-associations were observed in the LVM, LVESV, RVEDV, RVESV, RAV min, AAo aorta distensibility, DAo aorta distensibility, global peak circumferential strain, and global longitudinal peak strain (Fig. 2C and Figs. S9-S16), and negative correlations were observed in all the 4 ejection fraction traits (RVEF, LAEF, RAEF, and LVEF), LVCO, and global radial strain (Fig. 2D and Figs. S17-S22). Furthermore, the auditory network had strong negative associations with volumetric measures of heart, including all the 4 stroke volumes (LVSV, RVSV, LASV, and RASV), LVCO, LVEDV, RVEF, left atrium maximum volume (LAV max), and RAEF (Figs. S17, S19, S21, and S23-S28). On the other hand, the AAo aorta distensibility and DAo aorta distensibility were positivity correlated with the auditory network (Figs. S13-S14). There were also significant correlations with cognitive networks, including the default mode, cingulo-opercular, dorsal attention, and frontoparietal networks. For example, the default mode network was associated with the 4 ejection fraction traits, most of the associations were positive (Fig. S29). The default mode network was also negatively associated with other volumetric heart measures (such as the LVESV, LVSV, LVM, LASV, RVEDV, RVESV, RVSV, and RASV), whose associations were enriched in the hippocampal and visual areas of the default model network (Figs. S30-S31). For aortic measures, the associations were mainly located in the hippocampal areas of the default model network (Fig. S32). In summary, MRI-based endophenotype analysis reveals substantial neuro-cardiac interactions. Our results show the specific pattern of heart-brain connections and highlight important brain areas and networks that might be strongly related to heart health.

Mediation analysis with cardiovascular risk factors and biomarkers

Environmental factors and biomarkers may play an important role in the underlying mechanisms of heart-brain interactions. Clear evidence from clinical and epidemiological studies exists that cardiovascular risk factors (such as hypertension, high cholesterol, high fasting glucose, metabolic syndrome, and chronic kidney disease) negatively influence brain health and neurocognitive performance^1,3,46,47. However, it remains challenging to understand how these cardiac risk factors cause brain damages. To bridge this gap, we performed mediation analysis by using the 82 CMR traits as intermediate variables⁴⁸. Specifically, we investigated whether cardiovascular risk factors could influence brain structure and function indirectly through heart conditions captured by the 82 CMR traits (Methods). We examined cardiovascular risk factors, including diastolic blood pressure, systolic blood pressure, diabetes mellitus, smoking, drinking, and basal metabolic rate, as well as 34 biomarkers collected in the UKB study.

Significant CMR-mediation effects were observed for 1,148 pairs between 34 cardiovascular risk factors/ biomarkers and 235 brain MRI traits (P < 2.21 × 10^-6, Table S3). These results suggest that associations between cardiovascular risk factors and brain health were partially mediated by cardiac structure and function. For example, the diastolic blood pressure was widely associated with cortical thickness, regional brain volumes, DTI parameters, and functional connectivity (Fig. 3A). On average, 40.3% of diastolic blood pressure’s effects on DTI parameters were indirect and mediated through CMR traits (Fig. 3B). The other diastolic blood pressure’ effects (59.7%) could be direct effects on DTI parameters or indirect effects through non-cardiac mechanisms. The average proportion of mediated effects was 56.2% for cortical thickness, 38.6% for regional brain volumes, 37.5% for resting fMRI traits, and 33.6% for task fMRI traits. Similar meditation patterns were also observed on systolic blood pressure and hypertension (Fig. S33). Hypertension has been reported to be associated with cognitive dysfunction⁴⁹ and brain white matter hyperintensities, which might be a consequence of chronic ischaemia caused by cerebral small vessel disease⁵⁰. Mediation effects were also observed on multiple cardiovascular biomarkers, such as low-density lipoprotein (LDL) cholesterol, high-density lipoprotein (HDL) cholesterol, and triglyceride (Fig. S34).

• Download figure
• Open in new tab

Fig. 3 Effects of cardiovascular risk factors on brain MRI mediated through CMR traits.

(A) We illustrate marginal effects (after adjusting for covariates) of diastolic blood pressure on brain MRI traits and the corresponding conditional effects after further adjusting for CMR traits. The difference between marginal and condition effects indicates the indirect effect mediated through CMR traits. As illustrated in (B), the average proportion of CMR-mediated diastolic blood pressure’s effects was 40.3% for DTI parameters, 37.5% for resting fMRI, 33.6% for task fMRI, 56.2% for cortical thickness, and 38.6% for regional brain volumes. (C) Marginal effects (after adjusting for covariates) of type 2 diabetes (ICD 10 code E11.9) on brain MRI traits and the corresponding conditional effects after further adjusting for CMR traits. (D) Marginal effects (after adjusting for covariates) of glycated hemoglobin A (HbA1c) on brain MRI traits and the corresponding conditional effects after further adjusting for CMR traits.

CMR-mediated effects of type 2 diabetes (T2D) on the brain were observed (Fig. 3C). Insulin signaling plays a critical role in brain health and patients with diabetes are at higher risk of cognitive impairment and dementia⁵¹. The average proportion of CMR-mediated T2D effects was 16.8% for DTI parameters, 10.3% for cortical thickness, and 13.1% for regional brain volumes. For example, T2D was negatively related to total brain volume (β = -0.025, P = 9.64 × 10^-9). After adjusting for the CMR traits, the effect became smaller (β = -0.017, P = 5.80 × 10^-5), suggesting that 32% (0.008/0.025) of the T2D effects were mediated through CMR traits. In addition, mediation effects were observed on glycated hemoglobin A (HbA1c) and glucose, which were biomarkers of diabetes. High HbA1c level (hyperglycaemia) is associated with worse brain health and cognitive functions⁵², such as memory loss, poorer processing speed, attention, concentration, and executive functions^53,54. As the primary source of energy for the brain, glucose metabolism plays an important role in the physiological and pathological functioning of the brain⁵⁵. About 15.5% of HbA1c’s effects on DTI parameters and 14.1% of glucose’s effects on cortical thickness were CMR-mediated (Fig. 3D and Fig. S35). Other biomarkers whose effects on brain MRI were mediated by CMR included creatinine, urate, total protein, and gamma-glutamyl transpeptidase (GGT) (Figs. S36-S37). Creatinine, total protein, and urate are kidney biomarkers of renal function, which have been consistently associated with white matter brain deficits⁵⁶ and neurodegenerative diseases⁵⁷. GGT is a liver biomarker of potential hepatic or biliary diseases and has been reported to be associated with cognitive decline and increased risk of dementia⁵⁸. In summary, multiple categories of factors and biomarkers strongly influence brain health, and their influences are partly mediated by cardiovascular conditions. Understanding these mediation mechanisms underlying heart-brain connections might assist in preventing and detecting brain diseases.

Heritability and the associated genetic loci of 82 CMR traits

We estimated the single-nucleotide polymorphism (SNP) heritability for the 82 CMR traits using UKB individuals of white British ancestry⁵⁹ (n = 31,875, Methods). The mean heritability (h²) was 23.0% for the 82 traits (range = (7.06%, 70.3%), Fig. 4A), all of which remained significant adjusting for multiple testing using the Benjamini-Hochberg procedure to control the false discovery rate (FDR) at 0.05 level (Table S4). The h² of the AAo maximum area, AAo minimum area, DAo maximum area, and DAo minimum area was larger than 50%. Among cardiac traits, the global wall thickness, RVESV, RVEDV, LVESV, LVEDV, and LVM had the highest heritability (h² > 37.7%). As expected, more reproducible CMR traits show higher heritability (correlation = 0.88, P < 2.2 × 10^-16, Fig. S38). To identify reliable genetic signals for these CMR traits, large-scale samples from a homogeneous cohort are needed, especially for CMR traits with relatively low heritability.

• Download figure
• Open in new tab

Fig. 4 Genetics of CMR traits in the UK Biobank.

(A) SNP heritability of 82 CMR traits across the 6 categories. The x axis displays the ID of CMR traits, see Table S1 for full names of these traits. AAo, ascending aorta; DAo, descending aorta; LA, left atrium; LV, left ventricle; RA, right atrium; and RV, right ventricle. The average heritability of each category is labeled. (B) Ideogram of 80 genomic regions associated with CMR traits (P < 6.09 × 10^-10). (C) Left ventricular end-systolic volume (LVESV) was associated with the 22q11.23 region in both UK Biobank and Biobank Japan studies (shared index variant rs5760061). (D) Left ventricular end-systolic volume (LVESV) was associated with the 8q24.13 region in both UK Biobank and Biobank Japan studies (shared index variant rs34866937). (E) Left ventricular ejection fraction (LVEF) was associated with the 8q24.13 region in the Biobank Japan study (index variant rs11025521), but not in the UK Biobank study.

We next performed GWAS for the 82 CMR traits using this white British cohort (n = 31,875, Methods). All Manhattan and QQ plots can be browsed through the server on Heart-KP (http://67.205.180.40:443/). The intercepts of linkage disequilibrium (LD) score regression (LDSC)⁶⁰ were all close to one, suggesting no genomic inflation of test statistics due to confounding factors (mean intercept = 1.0002, range = (0.987, 1.019)). At the significance level 6.09 × 10^-10 (5 × 10^-8/82, that is, the standard GWAS significance threshold, additionally Bonferroni-adjusted for the 82 traits), we identified independent (LD r² < 0.1) significant associations in 80 genomic regions (cytogenetic bands) for 49 CMR traits, including 35 for left ventricular, 35 for ascending aorta, 14 for descending aorta, 10 for right ventricular, and 1 for left atrium (Fig. 4B and Table S5, Methods). Detailed interpretations of these identified regions can be found in next Section. When relaxing the significance level to 5 × 10^-8, there were 156 significant genomic regions (76 more) for 72 CMR traits.

On average, the 80 (covering 12.3% of genetic variants) and 76 (10.9% variants) CMR-associated genetic regions separately explained 20.5% (that is, 4.7%/23%) and 12.6% (2.9%/23%) of the heritability of CMR traits, indicating that about 66.9% (15.4%/23%) of the heritability remained unexplained by these significant regions (Fig. S39). These results illustrate a highly polygenic genetic architecture for CMR traits⁶¹. In addition, the 80 CMR-associated regions explained 14.3% (6.8%/47.6%) of the SNP heritability of DTI parameters, suggesting that the genetic effects on white matter microstructure were enriched in these CMR-associated regions. We next used the previously identified DTI-associated genes as an annotation⁴⁰ to perform partitioned heritability enrichment analysis⁶² for the 82 CMR traits. At FDR 5% level, heritability of 26 CMR traits was significantly (P range = (6.89 × 10^-5, 1.15 × 10^-2)) enriched in genomic regions influencing DTI parameters (Fig. S40). In addition to DTI parameters, other brain MRI traits showed overlapped genetic influences with CMR traits, but with less evidence for enrichment. Specifically, the 80 CMR-associated regions can explain 13.7% (1.7%/12.8%) of the heritability of task fMRI traits, 13.4% (2.8%/20.9%) of cortical thickness, 12.5% (4.6%/36.8%) of regional brain volumes, and 12.2% (1.8%/14.8%) of resting fMRI traits. Overall, these results suggest that the CMR traits had shared heritability with brain MRI traits, especially with DTI parameters measuring white matter microstructure.

To replicate the identified loci, we performed separate GWAS using hold-out datasets in the UKB study that were independent from our discovery dataset. First, we repeated GWAS on a European dataset with 8,252 subjects (Methods). For the 248 independent (LD r² < 0.1) CMR-variant associations in the 80 genomic regions, 57 (23%, in 25 regions) passed the Bonferroni significance level (2.02 × 10^-4, 0.05/248) in this European validation GWAS, and 184 (74.19%, in 61 regions) were significant at nominal significance level (0.05). All the 184 associations had concordant directions in the two independent GWAS and the correlation of their genetic effects was 0.93 (Fig. S41 and Table S6). These results show a high degree of generalizability of our GWAS findings among European cohorts. We also performed GWAS on two non-European UKB validation datasets: the UKB Asian (UKBA, n = 500) and UKB Black (UKBBL, n = 271). One association between 8q24.3 and the RVEF passed the Bonferroni significance level (P = 6.20 × 10^-5) in UKBA, and 13 more regions passed nominal significance level. For UKBB, 11 regions passed the nominal significance level and none of them survived the Bonferroni significance level, which may partially be due to the small sample size of this non-European GWAS.

Additionally, we evaluated the ancestry specific effects using Asian GWAS summary statistics of three CMR traits (analogous to the LVEDV, LVESV, and LVEF³⁴), which were generated from 19,000 subjects in the BioBank Japan (BBJ) study⁶³. At the 5 × 10^-8 threshold, the BBJ identified independent (LD r² < 0.1) significant associations in 6 regions (2p14, 11p15.1, 22q11.23, 8q24.13, 10q22.2, and 18q12.1), 4 of which (22q11.23, 8q24.13, 10q22.2, and 18q12.1) were among the 156 regions we discovered based on data from UKB white British. Specifically, the 22q11.23 and 8q24.13 regions were significantly associated with the LVSEV in both UKB and BBJ studies (Figs. 4C-4D). The 10q22.2 and 18q12.1 regions were separately associated with the LVEDV and LVESV in BBJ but not in UKB (Figs. S42-S43). Instead, the two regions were significantly associated with the wall thickness traits in UKB (P < 6.77 × 10^-10). Additionally, only CMR traits in BBJ were associated with the 11p15.1 and 2q14 regions (Figs. 2E and S44), likely representing population specific genetic influences in Asian population.

Finally, we constructed polygenic risk scores (PRS) via lassosum⁶⁴ to evaluate the out of sample prediction power of the discovery GWAS results. We used 2,000 subjects from our European validation GWAS as validation data and evaluated the performance on the left independent subjects (n = 5,551). Among the 82 CMR traits, 73 had significant PRS at FDR 5% level (P range = (7.38 × 10^-136, 3.49 × 10^-2), Table S7). The highest incremental R-squared (after adjusting the effects of age, sex, and ten genetic principal components (PCs)) was observed on the AAo minimum area and AAo maximum area (8.52% and 8.36%, respectively). Other traits whose incremental R-squared larger than 1% included the DAo minimum area, DAo maximum area, RVEDV, RVESV, LAV max, and wall thickness traits (R-squared range = (1.01%, 3.32%), P < 4.33 × 10^-15). To evaluate the cross-population performance, PRS was also constructed on UKB white British discovery GWAS data using BBJ GWAS summary statistics of the LVEDV, LVESV, and LVEF. We found that the PRS of these three traits were all significant in the UKB (P range = (1.58 × 10^-11, 8.13 × 10^-7), R-squared range= (3.90 × 10^-4, 1.35 × 10^-3)). The prediction accuracy was lower than that in the above within European prediction analysis (R-squared range= (7.72 × 10^-3, 9.67 × 10^-3)), which may be explained by the smaller training GWAS sample size in BBJ⁶⁵ and the differences of genetic effects and LD among European and Asian populations⁶⁶. More efforts are needed to identify causal variants associated with CMR traits in global diverse populations and quantify population-specific heterogeneity of genetic effects.

Pleiotropy of genetic variants across body systems

To identify the shared genetic effects between CMR traits and complex traits, we carried out association lookups for independent (LD r² < 0.1) significant variants (and variants in their LD, r² ≥ 0.6, P < 6.09 × 10^-10) detected in our UKB white British GWAS. In the NHGRI-EBI GWAS catalog⁶⁷, our results tagged variants that have been linked to a wide range of traits and diseases, including heart diseases, heart structure and function, blood pressure, lipid traits, blood traits, diabetes, stroke, neurological and neuropsychiatric disorders, psychological traits, cognitive traits, lung function, parental longevity, smoking, drinking, and sleep. To evaluate whether two associated genetic signals were consistent with the shared causal variant, we applied the Bayesian colocalization analysis⁶⁸ for CMR traits and selected complex traits whose GWAS summary statistics were publicly available. Evidence of pairwise colocalization was defined as having a posterior probability of the shared causal variant hypothesis (PPH4) > 0.8^68,69.

First, our results replicated 25 genomic regions that have been linked to cardiac and aortic traits in previous GWAS, including 5 regions with left ventricular traits (such as 22q11.23 with fractional shortening⁷⁰, ejection fraction⁶³, and left ventricular internal dimension⁶³, Fig. S45); 17 regions with heart rate and electrocardiographic traits (such as 1p36.32, 5q33.2, 10q25.2, 12q24.21, and 14q24.2 with PR interval⁷¹ and QRS duration⁷²) (Figs. S46-S50); and 6 regions with aortic measures (such as in 15q21.1 with thoracic aortic aneurysms and dissections⁷³ and in 15q24.1 with aortic root size⁷⁰, Figs. S51-S52). In addition, 24 regions had shared associations (LD r² ≥ 0.6) with cardiovascular diseases, including 9 regions with coronary artery disease^74,75 (such as 17p13.3 and 12q24.12, Figs. S53-S54); 9 regions with atrial fibrillation^76,77 (such as 6p21.2, 1p13.1, 2q31.2, 8q24.13, 15q26.3, and 22q12.1, Figs. 5A and S55-S59); 6q22.33, 10q23.33, and 11p15.5 with hypertension^78,79 (Figs. S60-S62); 15q21.1 and 13q12.11 with abdominal aortic aneurysm^80,81 (Figs. S51 and S63); 14q24.2 with mitral valve prolapse⁸² (Fig. S50); and 10q26.11 and 1p36.13 with idiopathic dilated cardiomyopathy⁸³ (Figs. S64-S65). There was widespread evidence of colocalization on many on many loci (PPH4 > 0.849). Additionally, 41 of the 80 genomic regions were associated with blood pressure traits, such as systolic blood pressure, diastolic blood pressure, pulse pressure⁸⁴, and mean arterial pressure⁸⁵ (Figs. 5B and S66-S85). CMR traits were in LD (r² ≥ 0.6) with various cardiovascular biomarkers in 32 genomic regions, such as 12q24.12, 6q22.33, 5q31.1, and 17q12 with blood metabolite levels⁸⁶ and lipid traits^87,88 (including HDL, LDL, and triglycerides, Figs. S54, S60, S71, and S86); and 3p13, 3p25.1, 5q15, 12q14.1, and 22q13.1 with red blood cell count⁸⁹, blood protein levels⁹⁰, red cell distribution width⁹¹, and plateletcrit⁸⁹ (Figs. S87-S91).

• Download figure
• Open in new tab

Fig. 5 Selected genetic loci that were associated with both CMR trait and other complex traits and diseases.

(A) In 6p21.2, we observed colocalization between the global myocardial-wall thickness at end-diastole (WT global, index variant rs4151702) and atrial fibrillation (index variant rs3176326). The posterior probability of Bayesian colocalization analysis for the shared causal variant hypothesis (PPH4) is 0.904. In this region, the WT global was also in LD (r² ≥ 0.6) with ischemic stroke. (B) In 7p21.1, we observed colocalization between the ascending aorta minimum area (DAo min area) and systolic blood pressure (shared index variant rs2107595, PPH4 = 0.960). In this region, the DAo min area was also in LD (r² ≥ 0.6) with stroke, intracranial aneurysm, coronary artery disease, and moyamoya disease. (C) In 15q25.2, we observed colocalization between the regional myocardial-wall thickness at end-diastole (WT AHA 7, index variant rs11635505) and schizophrenia (index variant rs12902973, PPH4 = 0.890). In this region, the WT AHA 7 was also in LD (r² ≥ 0.6) with bipolar disorder. (D) In 16q24.3, we observed colocalization between the ascending aorta minimum area (AAo min area, index variant rs488327) and neuroticism (index variant rs1673931, PPH4 = 0.991). (E) We illustrated the colocalization between the ascending aorta maximum area (AAo max area) and cerebrospinal fluid (CSF) volume (shared index variant rs7936928) in 11q24.3 (PPH4 = 0.902). (F) We illustrated the colocalization between the AAo max area and functional connectivity between the default mode and orbito-affective networks (Default <-> Orbito-Affective, shared index variant rs1678983) in 15q21.1 (PPH4 = 0.964).

We found genetic pleiotropy between CMR traits and multiple brain-related complex traits and disorders. In 6p21.2, 7p21.1, and 12q24.12 regions, CMR traits were in LD (r² ≥ 0.6) with stroke⁹² (such as ischemic stroke, large artery stroke, and small-vessel ischemic stroke), intracranial aneurysm⁹³, and Moyamoya disease⁹⁴ (Figs. 5A-5B and S54). The index variants of 7p21.1 (rs2107595) and 12q24.12 (rs597808) were expression quantitative trait loci (eQTLs) of TWIST1, ALDH2, and NAA25 in human brain tissues⁹⁵, suggesting that these CMR-associated variants were known to affect gene expression levels in human brain. When the GWAS significance threshold was relaxed to 5 × 10^-8, the shared associations (LD r² ≥ 0.6) with stroke were found in four more CMR-significant loci (12q13.3, 16q23.1, 4q25, 7p15.1). CMR traits were also in LD (r² ≥ 0.6) with neurodegenerative and neuropsychiatric disorders, such as in 17q21.31 (rs62062271, brain eQTL of multiple genes such as CRHR1 and WNT3) with Parkinson’s disease⁹⁶, corticobasal degeneration⁹⁷, multiple system atrophy⁹⁸, Alzheimer’s disease⁹⁹ and associated cognitive impairment¹⁰⁰, autism spectrum disorder¹⁰¹, and progressive supranuclear palsy¹⁰² (Fig. S92); in 12p12.1 (rs4148674, brain eQTL of ABCC9) with hippocampal sclerosis of aging¹⁰³ (Fig. S78); in 15q25.2 (rs11635505, brain eQTL of UBE2Q2L, WDR73, and GOLGA6L4) and 17p13.3 (rs2281727, brain eQTL of SRR) with schizophrenia¹⁰⁴ (Figs. 5C and S53); in 17q12 (rs903503, brain eQTL of multiple genes such as PGAP3 and MED24) and 6q22.31 (rs1334489, brain eQTL of CEP85L) with bipolar disorder^105,106 (Figs. S86 and S93); and in 3p14.3 (rs2686630, brain eQTL of ABHD6) with eating disorder¹⁰⁷ (Fig. S94). In addition, CMR traits were in LD (r² ≥ 0.6) with mental health traits, such as in 16q24.3, 17q21.31, 8p23.1 (rs903503, brain eQTL of multiple genes such as FAM167A and DEFB134), and 11p11.2 (rs11039348, brain eQTL of multiple genes such as FAM180B and SLC39A13) with neuroticism¹⁰⁸, depressive symptoms¹⁰⁹, subjective well-being¹¹⁰, and risk-taking tendency¹¹¹ (Figs. 5D, S92 and S95-S96). For cognitive traits and education, we tagged 17q21.31, 11p11.2, and 11q13.3 with cognitive function¹¹² and educational attainment¹¹³ (Figs. S92, S96, and S97); 7q32.1 (rs2307036, brain eQTL of ATP6V1FNB) with reading disability¹¹⁴ (Fig. S98); and in 12q24.12 (rs597808, brain eQTL of ALDH2 and NAA25) with reaction time¹¹² (Fig. S54). We also found shared associations (LD r² ≥ 0.6) in 11q24.3 (rs11222084, brain eQTL of ADAMTS8), 12q24.12 (rs7310615, brain eQTL of ALDH2 and NAA25), 17p13.3 (rs10852923, brain eQTL of SRR and SMG6), 17q12 (rs10852923, brain eQTL of SRR and SMG6), and 17q21.31 with DTI parameters⁴⁰ (Figs. S99-S103); in 11q24.3 (rs7936928, brain eQTL of ADAMTS8), 3p13, 11p11.2 (rs7107356, brain eQTL of multiple genes such as FAM180B and SLC39A13), 17q21.31 (rs118087478, brain eQTL of multiple genes such as ARL17A and ARL17B) with regional brain volumes³⁸ (Fig. 5E and Figs. S104-S106); and in 15q21.1, 8p23.1 (rs10093774, brain eQTL of multiple genes such as RP1L1 and FAM167A), 10q23.33 (rs10093774, brain eQTL of HELLS), 11q13.3 (rs10093774, brain eQTL of HELLS), and 17q21.31 (rs62062271, brain eQTL of multiple genes such as ARL17A and LRRC37A2) with fMRI traits⁴² (Fig. 5F and Figs. S107-S110). Furthermore, CMR traits were colocalized with these brain-related complex traits in many regions, such as in 15q25.2 with schizophrenia, in 16q24.3 with neuroticism, in 11q24.3 with cerebrospinal fluid volume, and in 15q21.1 with functional connectivity (Figs. 5C-5F, PPH4 > 0.890). There is substantial evidence to support the interaction between cardiovascular health status and brain health. For example, people with better heart healthy have better cognitive abilities¹¹⁵ and lower risk for brain disorders, such as stroke and Alzheimer’s disease¹¹⁶. In addition, mental health disorders may result in biological processes and behaviors that related to cardiovascular risk factors, such as smoking initiation and physical inactivity^11,117. Our results reveal that cardiovascular conditions have substantial genetic overlaps with both mental health and cognitive functions, which may partially explain the heart-brain connections. The majority of shared genetic variants were found to be eQTLs in brain tissues. Future studies are needed to explore the potential genomic pathways and mechanisms underlying the shared genetic influences between heart and brain.

Genetic overlaps with other diseases and complex traits were also observed. For example, RVEDV is in LD (r² ≥ 0.6) with type 1 diabetes¹¹⁸, T2D¹¹⁹, coronary artery disease⁷⁵, and parental longevity¹²⁰ in the 12q24.12 region (Fig. S54). CMR traits were in LD (r² ≥ 0.6) with lung function and diseases in 7 regions, such as in 17q12 with asthma¹²¹ (Fig. S86); in 17q21.31 with idiopathic pulmonary fibrosis¹²² and interstitial lung disease¹²³ (Fig. S92); and in 10q26.11, 3p22.1, 5q31.1, and 15q23 with lung function (FEV1/FVC)¹²⁴ (Figs. S64, S68, S71, and S81). We also found shared genetic associations (LD r² ≥ 0.6) in 11p11.2, 12q24.12, and 17q12 with smoking¹¹¹ (Figs. S96, S54, and S86); and in 17p13.3, 11p11.2, and 17q21.31 with alcohol consumption and alcohol use disorder¹²⁵ (Figs. S53, S96, and S92). All the above results can be found in Table S8.

Genetic correlations with brain disorders and complex traits

First, we examined the genetic correlations (GC)¹²⁶ among the 82 CMR traits. Strong genetic correlations were observed within and between categories of CMR traits (Fig. S111). For example, the RVEDV was genetically correlated with other RV traits, including RVSV and RVESV (GC > 0.85, P < 1.01 × 10^-137), as well as RVEF (GC = -0.47, P = 4.93 × 10^-10). The RVEDV was also significantly correlated with CMR traits from other categories, such as AAo maximum area and DAo maximum area (GC > 0.38, P < 1.36 × 10^-13), LASV and RASV (GC > 0.37, P < 3.02 × 10^-7), LVEDV, LVESV, and LVM (GC > 0.594, P < 2.74 × 10^-35), and LVEF (GC = -0.55, P = 5.25 × 10^-20). In addition, we found a strong relationship between phenotypic and genetic correlations among all CMR traits (β = 0.751, P < 2 × 10^-16).

Next, we examined the genetic correlations between 82 CMR traits and 33 complex traits, many of which were colocalized traits in the above section. At the FDR 5% level (82 × 33 tests), the CMR traits were significantly associated with heart diseases, lung function, cardiovascular risk factors, and brain-related complex traits and diseases (Table S9). For example, hypertension had strong genetic correlations with aortic traits and LV traits (Fig. 6A). The strongest correlation between LV traits and hypertension was found in wall thickness traits (GC range = (0.237, 0.406), P < 6.18 × 10^-9), which were also significantly associated with coronary artery disease, T2D, and stroke. In addition, atrial fibrillation was significantly associated with aortic, LA, and RA traits (|GC| range = (0.175, 0.252), P < 7.59 × 10^-4), suggesting that atrial fibrillation might have a higher genetic similarity with LA/RA traits than LV/RV traits.

• Download figure
• Open in new tab

Fig. 6 Genetic correlations and integrative prediction.

(A) We illustrated selected genetic correlations between CMR traits (x axis) and complex traits and diseases (y axis). The asterisks highlight significant genetic correlations after at the FDR 5% level. (B) Predicting heart diseases using genetic variants and CMR traits. Genetic PRS, polygenic risk scores of genetic variants. (C) The accuracy of diabetes prediction analysis using different types of data. Brain volume, region brain volumes; and DTI parameters, diffusion tensor imaging parameters.

In both schizophrenia and bipolar disorder, multiple LV traits showed significant genetic correlations. Specifically, on LVCO, LVEF, radial strains, and wall thickness, positive genetic correlations were observed with schizophrenia and bipolar disorder (GC range = (0.122, 0.236), P < 9.37 × 10^-3). Negative genetic correlations with the two brain disorders were observed on peak circumferential strains (GC range = (−0.168, -0.102), P < 1.26 × 10^-3). Additionally, anorexia nervosa (eating disorder) was significant associated with LAV min and LAEF (|GC| range = (0.189, 0.220), P < 8.11 × 10^-3) and cognitive traits and neuroticism were mainly associated with right heart traits (RA and RV traits). For example, intelligence was genetically correlated with the RAV max, RAV min, and RVEDV (GC range = (0.08, 0.12), P < 7.10 × 10^-3). Lung functions (FEV and FVC) had genetic correlations with multiple CMR traits, with longitudinal strains showing the strongest correlations (GC range = (0.296, 0.357), P < 7.29 × 10^-6). There were more associations with other complex traits analyzed in previous GWAS, such as smoking, PR interval, blood pressure, education, risky behaviors, lipid traits, mean corpuscular hemoglobin, and mean corpuscular volume (Fig. S112A). We also found very high genetic correlations with previously reported four LV traits³⁴ (GC > 0.865, P < 6.28 × 10^-99) (Fig. S112B). In summary, genome-wide genetic similarities have been found between CMR traits and a wide range of complex traits and diseases. Discovering such genetic co-variations may improve our understanding of genetic pathways of clinical outcomes from a multi-organ perspective.

Biological and gene level analyses

We performed gene level association test using GWAS summary statistics of the 82 CMR traits with MAGMA¹²⁷. We identified 160 significant genes for 49 CMR traits (P < 3.24 × 10^-8, Bonferroni-adjusted for 82 traits) (Table S10). Next, we mapped significant variants (P < 6.09 × 10^-10) to genes by combining evidence of physical position, eQTL (expression quantitative trait loci) association, and 3D chromatin (Hi-C) interaction via FUMA¹²⁸. We found 576 mapped genes, 434 of which were not identified in MAGMA (Table S11). Moreover, 90 MAGMA or FUMA-significant genes had a high probability of being loss-of-function (LoF) intolerant¹²⁹ (pLI > 0.98), indicating significant enrichment of intolerant of LoF variation among these CMR-associated genes (P = 1.58 × 10^-4).

Ten genes (CACNA1I, REN, ATP1A1, CACNB2, KCNJ8, PDE11A, AHR, ESR1, CYP2C9, and ABCC9) were targets for 32 cardiovascular system drugs¹³⁰, such as 15 calcium channel blockers (anatomical therapeutic chemical (ATC) code: C08) to lower blood pressure, 5 cardiac glycosides (C01A) to treat heart failure and irregular heartbeats, and 3 antiarrhythmics (C01B) to treat heart rhythm disorders (Table S12). Three (CACNA1I, ESR1, and CYP2C9) of these 10 genes and 4 more CMR-associated genes (ALDH2, HDAC9, NPSR1, and TRPA1) were targets for 11 nervous system drugs, including 4 antiepileptic drugs (N03A) used in the treatment of epileptic seizures and 2 drugs for addictive disorders (N07B). Some drug target genes have known biological functions both in the heart and brain. For example, ALDH2 plays an important role in clearance of toxic aldehydes, which is an important mechanism related to myocardial and cerebral ischemia/reperfusion injury¹³¹. Therefore, ALDH2 has been proposed to be a protective target for heart and brain diseases/dysfunctions triggered by ischemic injury and related risk factors^132,133. Our results may help identify new drug targets or identify drugs that could be repurposed.

MAGMA gene-set analysis was performed to prioritize the enriched biological pathways, which yielded 36 significantly enriched gene sets at FDR 5% level (P < 3.0 × 10^-6, Table S13). Multiple pathways related to cardiac development and heart disease were detected, including “go cardiac septum morphogenesis” (Gene Ontology [GO]: 0060411), “go cardiac muscle tissue development” (GO: 0048738), “go cardiac muscle tissue regeneration” (GO: 0061026), “go cardiac chamber development” (GO: 0003205), “go cardiac chamber development” (GO: 0003205), “go cardiac muscle contraction” (GO: 0060048), “go cardiac ventricle development” (GO: 0003231), “go vasculogenesis involved in coronary vascular morphogenesis” (GO:0060979), “go adult heart development” (GO: 0007512), “go heart development” (GO: 0007507), and “bruneau septation atrial” (M5223). Finally, we performed partitioned heritability analyses⁶² to identify the tissues and cell types¹³⁴ where genetic variation leaded to differences in CMR traits. At FDR 5% level, the most significant heritability enrichments were found in active gene regulation regions of heart and muscle tissues, supporting the biological validity of the identified GWAS signals for CMR traits (Fig. S113).

Complex traits and diseases prediction using genetic and multi-organ MRI data

In this section, we examined the prediction performance of CMR traits for 95 complex traits and diseases. We evaluated the prediction performance in a training, validation, and testing design and removed the effects of age (at imaging), age-squared, sex, age-sex interaction, age-squared-sex interaction, imaging site, height, weight, body mass index, and the top 40 genetic PCs (Methods). At FDR 5% level, the CMR traits had significant prediction power on 44 traits and diseases, including diseases of the circulatory system (ICD-10 group code: “I”); endocrine, nutritional and metabolic diseases (ICD-10: “E”); endocrine, nutritional and metabolic diseases (ICD-10: “F”); mental health and cognitive traits; cardiovascular diseases (self-reported), biomarkers, and risk factors; and disease family history (prediction correlation range = (0.028, 0.5), P range = (1.60 × 10^-2, 5.05 × 10^-234), Fig. S114). For example, CMR traits significantly predicted 14 circulatory system diseases, including atrial fibrillation, essential (primary) hypertension, angina pectoris, atherosclerotic heart disease, old myocardial infarction, chronic ischaemic heart disease, and acute myocardial infarction (β > 0.097, P < 4.99 × 10^-17). They also had high prediction power for cardiovascular risk factors, such as systolic blood pressure (β = 0.367, P = 3.96 × 10^-219) and diabetes (self-reported) (β = 0.114, P = 9.50 × 10^-23). Moreover, CMR traits can predict brain-related psychological factors and cognitive traits, such as risk-taking (β = 0.068, P = 4.81 × 10^-9), depression (β = 0.046, P = 7.48 × 10^-5), and fluid intelligence (β = 0.091, P = 5.08 × 10^-15).

Next, we performed joint prediction with genetic PRS (Methods). In comparison to genetic PRS, CMR traits may be more accurate and provide additional insights when predicting heart diseases (Fig. 6B). For example, the correlation between genetic PRS and atrial fibrillation was 0.084 (P = 1.14 × 10^-7), suggesting that about 0.7% disease variation can be predicted by genetic profile. The prediction correlation of CMR traits was 0.265 (P = 2.78 × 10^-65), which was comparable to the performance when using both CMR traits and genetic PRS (β = 0.276, P = 1.20 × 10^-70). These results illustrate that CMR traits had much higher prediction accuracy and suggest that most of the genetic prediction power on atrial fibrillation might be medicated through cardiac conditions captured by CMR traits.

We also found that combining genetic PRS, CMR traits, and brain MRI traits can improve the prediction of multi-system diseases, such as diabetes (Fig. 6C). The prediction correlation of diabetes was 0.084 (P = 1.11 × 10^-7) for genetic PRS and 0.077 (P = 1.13 × 10^-6) for CMR traits. Multiple categories of brain MRI traits also had significant prediction power on diabetes, including DTI parameters (β = 0.094, P = 3.01 × 10^-9), regional brain volumes (β = 0.076, P = 1.47 × 10^-6), resting fMRI (β = 0.055, P = 4.85 × 10^-4), and task fMRI (β = 0.045, P = 4.46 × 10^-3). The prediction performance was improved to 0.10 (P = 3.84 × 10^-10) by using all CMR and brain MRI traits and further moved up to 0.125 (P = 2.17 × 10^-15) when adding genetic PRS. In summary, imaging traits could make a unique contribution to the prediction of complex traits and diseases. Multi-organ imaging and genetic PRS can be integrated to improve risk prediction and patient care.