Multi-trait GWAS reveals divergent local and polygenic architecture of cardiovascular traits across African and European ancestries in the UK Biobank

Abstract

The genetic architecture of cardiovascular traits is poorly characterised in non-European populations, limiting our understanding of disease aetiology and contributing to health disparities. Here, we analyse the genetic architecture of four cardiovascular traits (systolic and diastolic blood pressure, pulse rate, and maximum heart rate) using multi-trait analysis of genome-wide association studies and local genetic correlation analysis in 459,327 European (EUR) and 6654 African (AFR) ancestry individuals from the UK Biobank. Our analysis identifies 957 and 45 novel variants in the EUR and AFR cohorts, respectively, but reveals a profound divergence in the pleiotropic architecture of blood pressure. We identify 181 genomic loci with significant local genetic correlation between systolic and diastolic blood pressure (SBP-DBP) in the European sample, whereas such signals are completely absent in the African ancestry cohort. This marked disparity in local genetic correlation structure highlights that pleiotropic mechanisms can be highly ancestry-specific, underscoring the limitations of transferring genetic risk models across populations and the critical need for inclusive genomic research.

Sinkala et al. use multi-trait GWAS to reveal divergent pleiotropic architecture of cardiovascular traits across ancestries, finding local genetic correlations between blood pressure traits in Europeans that are entirely absent in Africans.

Introduction

Cardiovascular traits encompass a complex spectrum of phenotypes dictated by a confluence of genetic and environmental determinants^1,2. The distribution of these cardiovascular indices exhibits striking disparities across various ancestral backgrounds^3,4. In recent years, Genome-wide association studies (GWAS) have illuminated numerous genomic loci associated with these cardiovascular traits^5–7, highlighting the intricate genetic fabric that underpins cardiovascular health. Traits such as systolic blood pressure (SBP), diastolic blood pressure (DBP), and heart rate share loci with pathologies like stroke and myocardial infarction, and these associations often serve as harbingers of early mortality and general health^8–11. For instance, seminal investigations identified 535 novel variants linked with blood pressure traits in over a million European individuals⁷, while another study unearthed hundreds of systolic and diastolic BP-associated variants from the UK Biobank cohort¹². Crucially, most of these findings stem from populations of European descent¹³.

Despite bearing the highest hypertension burden, the African population remains comparatively under-explored in GWAS^13,14. Emerging large-scale GWAS and meta-analyses are beginning to shed light on the genetic underpinnings of this complex condition within this and other populations^15–18. However, few extensive studies have concurrently examined the genetic and social health determinants, despite compelling evidence substantiating their contribution to hypertension¹².

Drawing on evidence from a host of GWAS investigations into other traits^19–25, we posit that genetic variants influencing cardiovascular traits may manifest divergently among European and African individuals. Elucidating such variances may shed light on the differential genetic underpinnings and pathophysiological discrepancies between these demographic groups observed in cardiovascular traits and disease risk^14,15,26. However, despite the significant advances in multi-trait GWAS, there is a distinct paucity of research efforts focused on identifying and contrasting genomic loci associated with cardiovascular traits in African and European populations (note, here we use these terms to refer to populations of European and African descent, respectively, rather than those necessarily living in these geographical regions) using multiple cardiovascular measurement parameters.

This study explores the heterogeneities in cardiovascular traits, encompassing SBP, DBP, pulse rate, and ‘maximum heart rate during fitness Test’ (MHR), between African and European individuals represented in the UK Biobank²⁷. We performed multi-trait genome-wide association analyses (MTAG) to discern sets of variants influencing these cardiovascular traits (SBP, DBP, pulse rate, and MHR) within each population group. Finally, leveraging integrative enrichment analyses, we sought to identify the associated genes, pathways, and phenotypes. Collectively, our findings reveal distinct genetic determinants influencing cardiovascular traits in African and European populations, thereby advancing our understanding of the genetic intricacies of cardiovascular health across diverse populations.

Results

Distinct cardiovascular profiles characterise African and European populations

In our exploration of the UK Biobank datasets, encompassing cardiovascular metrics from 460,096 Europeans and 6,551 Africans, we noted a significant distinction in the mean diastolic blood pressure (DBP). Africans exhibited a markedly higher mean DBP of 84.7 mmHg, compared to 82.0 mmHg in Europeans (Welch test: t = 13.9, p = 1.2 × 10⁻⁴²; Fig. 1a). In contrast, there was no significant difference observed in the mean systolic blood pressure (SBP) between Africans (mean = 139.8 mmHg) and Europeans (139.4 mmHg), p = 0.187; Fig. 1b).

Fig. 1. Cardiovascular parameters and anthropometric correlations in Africans and Europeans.

Comparison of the cardiovascular function parameter in Africans and Europeans for (a) diastolic blood pressure, (b) systolic blood pressure, (c) pulse rate, and (d) maximum heart rate during fitness test. The boxplots indicate the distribution of each measurement among Africans and Europeans. The p-values shown for each comparison were calculated from Welch’s t-test. On each box, the central mark indicates the median, and the left and right edges of the box indicate the 25th and 75th percentiles, respectively. The whiskers extend to the most extreme data points not considered outliers, and the outliers are plotted individually using the '+' symbol. To make the visualisation clearer, the filled circle mark showing the distribution only include 1000 randomly sampled point from the entire samples size of each group. (e) Correlation between SBP, DBP, pulse rate, and MHR with other anthropometric measurements in Africans and Europeans. f Error bars show the DBP variation across BMI percentiles among Africans and Europeans. The middle point indicates the mean DBP, and the error bars indicate the standard error of the mean at the BMI percentile. Binned scatter plot showing Pearson’s linear correlation between (g) DBP and waist circumference and (h) DBP and SBP. The data points are spaced into rectangular bins, and each point is coloured based on logarithm bin size, with redder colours indicating a higher number of plots. The colour bar shows the colour scale. In panels (a–d), n = 460,096 biologically independent samples for Europeans and n = 6,551 biologically independent samples for Africans. In panels (e–h), sample sizes vary by trait availability; exact n values per group and trait are provided in Supplementary Data 1. In (f), error bars represent the standard error of the mean (SEM). In panels (g, h), data are presented as binned scatter plots; no error bars are shown.

Furthermore, our investigation revealed significant differences in pulse rate and MHR across the two populations. Africans consistently exhibited higher values (mean pulse rate = 70.3 bps, mean MHR = 116.0 bps) than their European counterparts (pulse rate = 68.9 bps, MHR = 110.3 bps). The statistical analysis highlighted these differences with p-values of 8.2 × 10⁻¹⁴ for pulse rate (Fig. 1c) and 1.9 × 10⁻¹³ for MHR (Fig. 1d), further highlighting the distinctive cardiovascular profiles between these two populations.

Existing studies indicate that SBP, DBP and pulse rate vary with both age and anthropometric measures such as BMI and waist circumference^28–31. To place these ancestry-specific differences in a broader phenotypic context, we first computed pair-wise Pearson correlations between the four cardiovascular traits and 32 anthropometric measures; the resulting heat map is shown in Fig. 1e (also see Supplementary Fig. 1 and Supplementary Data 1). DBP emerged with some of the strongest anthropometric correlations, motivating a focused examination of its BMI dependence. Figure 1f plots mean DBP ± SEM across BMI-percentile bins for each ancestry and shows that DBP rises significantly with BMI in both groups, but with a notably steeper slope in Africans. Formal trend tests (Ordinary-Least-Squares slope and Spearman rank-correlation coefficient ρ) confirm this pattern: DBP increases by ≈ 0.36 mm Hg per 5-percentile BMI step in Africans versus ≈ 0.54 mm Hg in Europeans (all p < 10⁻⁶; detailed statistics are provided in Supplementary Data 1).

We also examined two key bivariate relations across both populations. Figure 1g shows a modest yet highly significant association between DBP and waist circumference across the full cohort (Pearson r = 0.29, p < 1 × 10⁻³⁵⁰). Figure 1h highlights the intrinsic coupling of systolic and diastolic pressures (Pearson r = 0.67, p < 1 × 10⁻³⁵⁰).

To extend these ancestry-stratified observations, Supplementary Fig. 2 presents BMI-binned trajectories for SBP, MHR, and resting pulse-rate. These additional panels show similarly significant—but trait-specific—BMI trends, again with steeper slopes in Europeans for the blood-pressure traits and comparable slopes for pulse-related measures. Collectively, Fig. 1 and Supplementary Fig. 2 underscore the need to model both ancestry and BMI when analysing cardiovascular phenotypes and support our emphasis on directly measured traits rather than derived clinical diagnoses such as hypertension. A full set of descriptive statistics, mean-comparison tests, and trend-test results is available in Supplementary Data 1.

Single-trait GWAS analyses pinpoint ancestry-specific cardiovascular loci

We first analysed UK Biobank summary statistics for four cardiovascular traits—SBP, DBP, pulse rate and MHR—in the African and European ancestral groups.

In the AFR cohort, using the conventional genome-wide threshold (p < 5 × 10⁻⁸), we detected 72 candidate variants for pulse rate. Linkage-disequilibrium (LD) clumping in Functional Mapping and Annotation (FUMA)³² reduced these to two independent signals—rs9388010 (p = 6.4 × 10⁻¹⁰) and rs9375066 (p = 1.4 × 10⁻¹⁰) (Supplementary Table 1). No variants reached significance for SBP, DBP or MHR, most likely because of limited sample size in this ancestry group.

In the EUR cohort, contrastingly, the much larger European sample yielded 3365 independent signals for pulse rate, 2763 for SBP, 2988 for DBP, and 36 for MHR (Supplementary Table 1). These results underline the dramatic power difference between the two cohorts and motivated us to adopt a multi-trait approach to boost discovery in AFR.

Multi-trait GWAS (MTAG) integration amplifies discovery power and uncovers additional shared and ancestry-specific loci

To increase power, we next leveraged the genetic correlation among SBP, DBP, pulse rate and MHR by running MTAG within each ancestry, using the Joint Analysis of Summary Statistics (JASS)³³ framework for joint-summary-statistic analysis.

In the EUR cohort, the MTAG boosted discovery to 482 genomic risk loci comprising 956 lead SNPs and 4818 additional SNP–trait associations with p < 1 × 10⁻⁵ (Fig. 2a). These signals arose from 103,472 candidate variants; 11,144 of them are already listed in the GWAS Catalogue, providing external validation (Supplementary Table 2; Supplementary Data 2). We mapped the associated variants to 2134 genes, sketching a broad molecular landscape for cardiovascular regulation.

Fig. 2. Manhattan plots of genetic variants associated with cardiovascular traits in European and African populations.

The plots represent the results of MTAG analysis using JASS, depicting genome-wide significant associations with cardiovascular traits, including diastolic blood pressure (DBP), systolic blood pressure (SBP), pulse rate, and max heart rate (MHR) across each chromosome for (a) European (EUR) and (b) African (AFR) populations.

Whereas in the AFR cohort, despite the smaller sample, MTAG uncovered five risk loci—each represented by one lead SNP—and seven more significant associations (Fig. 2a). Among the 105 candidate variants tested, 16 have prior GWAS support (Supplementary Table 2; Supplementary Data 2), lending credibility to the signals. These variants map to eight genes that are likely to mediate cardiovascular traits in African populations.

Together, the single-trait and multi-trait analyses demonstrate that exploiting shared genetic architecture can markedly enhance locus discovery, especially in ancestries with modest sample sizes.

A total of 403,221 samples for GWA from the EUR and AFR populations were analysed in this study. The EUR population had a significantly larger sample size (396,670) than the AFR population (6551). Thus, the AFR sample size was roughly 60 times smaller than the EUR sample size. The data were evaluated for four traits using MTAG across both populations, which amounted to 26,204 observations in AFR and 1,586,680 in the EUR population. The larger sample size in the EUR population led to the detection of more (4818 individually significant) SNPs^34,35, while the smaller AFR sample size most likely accounts for the fewer SNP-trait (7 individually significant SNPs) associations detected.

Distinct population-specific variants underlie cardiovascular trait differences between Africans and Europeans

Comparative analysis of cardiovascular traits across AFR and EUR ancestries revealed both divergent and shared genetic underpinnings. To systematically map these effects, independent genomic risk loci (genomic regions) and identifies the top-associated, independent lead SNP within each were defined by applying FUMA to multi-trait GWAS summary statistics (P < 1 × 10⁻⁵), an approach that accounts for ancestry-specific linkage disequilibrium structures (see Methods section).

Our cross-ancestry comparison at the locus level revealed substantial sharing for regions first identified in the AFR cohort. Of the five genomic risk loci identified in the AFR analysis, four were also significant in the EUR analysis (Supplementary Data 2). In contrast, a comparison at the single-variant level pointed to considerable allelic heterogeneity. Only one of the five lead SNPs from the AFR cohort replicated in the EUR cohort: rs9388010 and GJA1 gene variant (AFR p = 1.93 × 10⁻¹¹, EUR p = 2.07 × 10⁻¹¹⁷; Table 1).

Table. 1.

Top variants associated with cardiovascular traits in AFR and EUR

Ethnicity	uniqueID	LeadSNP	LeadSNP_P	LinkedSNP	R2	Zscore_EUR	Zscore_AFR	LinkedSNP_P
EUR	6:122114451:A:G	rs9320841	1.5 × 10−117	rs9388004	0.52	28.23	12.14	9.9 × 10−11
EUR	6:122312336:A:G	rs2679675	3.4 × 10−11	rs7767179	0.11	36.04	12.68	7.8 × 10−11
EUR	12:64979818:A:T	rs2448532	3.3 × 10−19	rs1245187	0.20	14.34	6.08	1.7 × 10−6
EUR	5:157806250:A:G	rs7712354	1.6 × 10−9	rs10069494	0.70	−6.96	−10.92	3.1 × 10−5
EUR	19:2161443:A:G	rs8102624	1.7 × 10−19	rs4807210	0.27	2.18	4.26	4.4 × 10−5
AFR	6:122153737:A:G	rs9388010	1.9 × 10−11	rs2091624	0.97	36.04	12.62	2.1 × 10−177

A similar pattern was observed in the reciprocal analysis. While nine of the 482 risk loci identified in the EUR analysis replicated in the AFR cohort, this was driven by only ten specific lead SNPs, including rs9320841 and rs2679675 (Table 1). This disparity—high replication of risk loci but low replication of specific lead SNPs—suggests that while the same genomic regions often contribute to cardiovascular risk across ancestries, the specific causal variants within those regions may differ. Frequencies, chromosomal positions, and mapped genes for all replicated variants are provided in Supplementary Data 2.

Furthermore, our analysis revealed numerous SNPs uniquely associated with either EUR or AFR populations. Specifically, 946 SNPs were identified exclusively within EUR, while 4 SNPs were exclusive to AFR. These population-specific variants, detailed comprehensively in Supplementary Data 2, likely contribute significantly to observed phenotypic diversity and disease risk differences between these populations.

We identified a total of 11 common variants shared between AFR and EUR, which include SNPs in linkage disequilibrium with significantly associated variants within each population (Supplementary Fig. 3).

These findings highlight both population-specific and shared genetic contributions to cardiovascular risk, providing important insights into the genetic architecture underlying population differences in cardiovascular phenotypes.

Comparative analysis of effect sizes reveals population-specific and shared variant impacts

We conducted a reciprocal comparison of the 200 most significant MTAG-identified SNPs in each ancestry—plotting for each variant both the AFR and EUR beta estimates and standard errors (Fig. 3a shows AFR-top SNPs; Fig. 3b shows EUR-top SNPs)—to further elucidate differences and similarities in their contribution to cardiovascular traits. Our analysis revealed a substantial subset of 36,213 SNPs exhibiting consistent effect directions across both populations (Supplementary Data 2).

Fig. 3. Comparison of Beta Estimates for Top 200 Cardiovascular Trait-Associated Variants.

The 200 most significant variants associated with cardiovascular traits in (a) African and (b) European populations. The y-axis illustrates beta estimates and associated errors derived from MTAG analyses, estimated from the z-score output of JASS. The central point of each error bar represents the beta estimate, while the whiskers demonstrate the associated standard error. The markers are coloured differently by ethnicity. Panels (b, d) show the corresponding association strengths (−log₁₀(P)) for the same sets of variants. In all panels, green symbols represent estimates from the European-ancestry cohort and orange/red symbols represent estimates from the African-ancestry cohort. Variants are ranked by their MTAG P-value within the focal ancestry group. Effect sizes (β) and standard errors were derived from MTAG summary statistics (z-score output of JASS). In panels (a, b), error bars represent ± 1 standard error of the beta estimate derived from MTAG analysis. n = 200 variants shown per panel, selected as the most significant in each focal ancestry. European cohort: n = 396,670 biologically independent samples; African cohort: n = 6551 biologically independent samples. The individual variant-level source data for (a, c) (beta estimates and standard errors for the top 200 variants) are provided in Supplementary Data 2 (Beta Comparison sheet).

Effect sizes showed broad directional consistency but marked differences in magnitude (Fig. 3a, c). In contrast, the same variants exhibited profound attenuation of statistical significance when evaluated in the opposite ancestry (Fig. 3b, d): top AFR hits became only modestly significant in Europeans, whereas top European hits were almost uniformly non-significant in Africans, with most p-values falling orders of magnitude above genome-wide thresholds despite preserved effect directions.

Among these, the most significantly associated variant was rs10069648, displaying a notably stronger effect in the AFR population (β = 5.50; MTAG p = 3.77 × 10⁻⁸) compared to the EUR population (β = 3.93; MTAG p = 8.34 × 10⁻⁵). Additional prominent variants included rs9886832 a SLC24A2 variant (AFR β = −5.75; MTAG p = 9.03 × 10⁻⁹; EUR β = −3.47; MTAG p = 5.31 × 10⁻⁴), further demonstrating significant cross-population associations with variability in the magnitude of genetic effects.

Population-specific differences in effect sizes may be influenced by the distinct linkage disequilibrium structures and allele frequencies unique to each population, as well as the substantial differences in sample sizes, which typically enhance the statistical power of GWAS in larger cohorts^36–38. For example, rs112255786 (a FTH1P5 and TIAL1P1 variant) shows a strong association in EUR (β = 8.24; MTAG P = 1.91 × 10⁻¹⁶) but does not meet our discovery threshold in AFR (β = –3.00; MTAG P = 2.72 × 10⁻³), reflecting both reduced power and potential ancestry-specific effect heterogeneity. Conversely, in the EUR population, the variant rs9388004 exhibited the largest effect (EUR β = 28.23; MTAG p = 1.50 × 10⁻¹¹⁷; AFR β = 12.14; MTAG p = 9.92 × 10⁻¹¹), highlighting the enhanced statistical detection capabilities afforded by larger EUR cohort sizes.

Our findings underscore both shared genetic influences and notable population-specific variants, contributing to the phenotypic diversity and differential genetic susceptibility to cardiovascular diseases observed between AFR and EUR populations. Our findings emphasize the necessity of accounting for population variation in genetic studies to enhance precision in risk prediction and therapeutic strategies^39,40.

Widespread local genetic correlations are detected across cardiovascular traits

To investigate the shared genetic architecture of four key cardiovascular traits, we performed a comprehensive local genetic correlation analysis in both EUR and AFR ancestry cohorts. The analysis revealed a dense landscape of pleiotropy, with substantial differences observed between the two populations.

In the AFR ancestry sample, we identified 4090 loci with significant local heritability and 1148 significant bivariate correlations (Fig. 4a, Supplementary Data 4). In the European ancestry sample, we identified 7624 distinct genomic loci with significant local heritability for at least one trait (FDR q < 0.05; Fig. 4b and Supplementary Data 4). Subsequent bivariate analysis across these regions uncovered 2934 significant local genetic correlations between trait pairs (q < 0.05; Fig. 4c and Supplementary Data 4). These findings highlight a substantially larger number of detectable pleiotropic loci in the European sample compared to the African sample.

Fig. 4. Ancestry-specific landscape of local genetic correlations for cardiovascular traits.

a Equivalent plot for the EUR ancestry sample. Each point represents a genomic locus, plotted by its chromosomal position (x-axis). The y-axis shows the significance of the local genetic correlation for the most significant trait pair within that locus, expressed as −log10(P−value). The dashed horizontal line indicates the threshold for significance after correcting for multiple testing (FDR q < 0.05). Loci containing a genome-wide significant lead SNP from our MTAG analysis (identified via FUMA) are highlighted to distinguish them from other regions. The heatmaps quantifying the number of genomic loci with significant pleiotropic effects shared between pairs of four cardiovascular traits. b Results from the AFR ancestry sample. c Results from the EUR ancestry sample. Each cell in the matrix shows the total count of loci where a significant local genetic correlation (FDR q < 0.05) was detected between the corresponding pair of traits. The colour intensity corresponds to the number of shared significant loci, as indicated by the scale bar.

Patterns of pleiotropy are dominated by blood pressure and heart rate associations

In the European sample, the strongest signal for local genetic correlation was observed between SBP and DBP at locus 14 on chromosome 1, a region exhibiting near-perfect local correlation (ρ = 1.0, p = 1.77 × 10⁻⁸⁶). In the African sample, the most significant signal was detected between DBP and pulse rate at locus 2241 on chromosome 18 (ρ = 0.98, p = 2.19 × 10⁻¹²).

To characterise the broader patterns of shared genetics, we quantified the number of significant loci shared between each pair of traits (Fig. 4b, c). In both ancestries, the relationship between DBP and pulse rate accounted for the largest number of shared loci. However, the magnitude of this co-occurrence was considerably greater in the European sample (1515 loci) compared to the African sample (667 loci). Most of the detected local correlations were positive (Supplementary Fig. 4a–d), indicating that the same genetic variants tend to increase or decrease both traits in tandem. For example, of the 1516 significant DBP-Pulse Rate loci in Europeans, 1515 showed a positive correlation.

Genetic architecture of blood pressure differs markedly between ancestries

Local genetic correlation analysis revealed extensive ancestry-specific differences in the genomic distribution of heritability and pleiotropy. Univariate local SNP-heritability analysis identified 1248 loci in European ancestry and 412 loci in African ancestry individuals with significant local heritability for at least one of the four cardiovascular traits (FDR < 0.05). Only a small fraction ( < 5%) of these loci showed significant local h² for the same trait in the other ancestry. Bivariate analysis revealed an even more striking divergence: 181 genomic loci exhibited significant local genetic correlation between systolic and diastolic blood pressure in the European sample (maximum ρ = 0.99), whereas no such signals were detected in the African ancestry cohort.

Bivariate analysis revealed an even more striking divergence in the shared genetic architecture of blood pressure between the two populations. We identified 181 distinct loci with significant local genetic correlation between SBP and DBP in the European sample. Of these, 156 (86.2%) contained a genome-wide significant lead SNP for blood pressure traits as identified by FUMA annotation, confirming these regions as robustly associated with blood pressure regulation.

In stark contrast, our analysis detected no significant SBP-DBP loci in the African ancestry sample (q > 0.05 for all loci). This divergence in the local genetic relationship between SBP and DBP is visually apparent in the Manhattan plots (Supplementary Fig. 5 and Supplementary Data 4), which show numerous high-significance peaks for the EUR analysis that are absent in the AFR analysis, highlighting a profound difference in the detectable shared genetic control of blood pressure between ancestries.

Ancestry-specific local pleiotropy in cardiovascular traits

Conditional local genetic correlation analysis using the full four-trait covariance matrix revealed marked differences in pleiotropic architecture between African and European ancestry individuals. The unconditional local genetic correlation between systolic and diastolic blood pressure failed to converge in any of the 2495 tested loci in either ancestry, in contrast to the hundreds of near-perfect correlations observed in Europeans under pairwise modelling (see Supplementary Data 3). Nevertheless, a joint test of heritability across all four traits was highly significant in 1033 loci in both ancestries after Benjamini–Hochberg FDR correction (41.4% of loci). Partial genetic correlations for systolic–diastolic blood pressure were estimable after conditioning on pulse rate (681 loci FDR < 0.05), maximum heart rate (279 loci), or both (15 loci), with identical counts in African and European samples.

The non-convergence of pairwise and unconditional systolic–diastolic correlations in the full model, likely due to complex opposing effects across traits within loci, precludes direct quantification of heart-rate-independent blood-pressure pleiotropy in Europeans^41,42.

Identification of novel and known genomic risk loci associated with cardiovascular traits in AFR and EUR populations

Following the multi-trait analysis, we employed FUMA to define independent genomic risk loci and distinguish between novel and previously reported associations. The p-values for all identified loci reflect their joint genetic association across the four cardiovascular traits.

In the African ancestry cohort, this revealed five significant genomic risk loci, of which four were novel (Supplementary Data 2). These novel loci included associations mapped to RBM15-AS1 on chromosome 1 (lead SNP rs12568839, p = 6.63 × 10⁻⁸), GLT8D2 on chromosome 12 (rs537104253, p = 3.63 × 10⁻⁷), and CA5A on chromosome 16 (rs56195233, p = 7.33 × 10⁻⁸). The single previously reported locus in this cohort, tagged by rs9388010 near LOC105377979 on chromosome 6 (p = 1.93 × 10⁻¹¹), has established associations with several cardiovascular phenotypes.

A substantially larger set of 482 loci were identified in the European ancestry analysis, including 108 novel associations, highlighting a considerable expansion of the genetic map for these traits. This analysis reaffirmed key cardiovascular risk regions, including those containing loci CDK11B on chromosome 1 (lead SNP rs28401288, P = 3.14 × 10⁻¹³) and CLCN6 on chromosome 1 (rs55857306, P = 1.65 × 10⁻²⁸), associated with multiple blood pressure-related traits⁴³, and which have been extensively linked to blood pressure regulation in prior GWAS^44–46. Detailed statistics for all lead SNPs, including per-trait effect sizes (beta), confidence intervals, and allele frequencies, are provided in Supplementary Data 2.

Known and novel genetic variants associated with cardiovascular and related traits in AFR and EUR populations

In the AFR cohort, among the five individually significant SNPs identified, considering the distinct linkage disequilibrium structures in AFR populations, we discovered four novel variants (defined as SNPs not previously reported in the GWAS Catalog or LDLink). These novel variants included rs537104253, located within the GLT8D2 gene on chromosome 12; rs56195233 in CA5A on chromosome 16 (Fig. 5a); and rs75484900 in SYNDIG1 on chromosome 20 (Fig. 5b; Table 2; Supplementary Data 5). Notably, rs56195233 (p = 7.33 × 10⁻⁸) stands out due to its association with multiple immune-related traits, such as basophil count and asthma, suggesting potential pleiotropic effects^36,47–49.

Fig. 5. Regional Association Plots Highlighting Novel and Previously Identified MTAG Loci Associated with Cardiovascular Traits in African Population.

The lead SNPs (a) rs56195233 and (b) rs75484900 represent novel findings, while the lead SNPs (c) rs9388010 and (d) rs12568839 have been previously linked to cardiovascular traits. Genes within the chromosomal loci are depicted in the lower panel. The filled circles indicate the position of the SNPs across the region on the x-axis and the negative logarithm of the association p-value on the y-axis. The lead SNP is highlighted in purple, with other SNPs within the locus coloured based on their linkage disequilibrium correlation value (r²) with the lead SNP, utilizing the African HapMap haplotype from the 1000 Genomes Project.

Table. 2.

Known and novel variants significantly associated with cardiovascular and related traits

Population	Cardiovascular Traits	Cardiovascular Associated Traits	Novel SNPs
Africans	1	1	3
Europeans	506	64	386

In addition, one SNP (rs9388010; p = 1.93 × 10⁻¹¹) identified in the AFR cohort has been robustly linked to cardiovascular traits in previous studies, including resting heart rate, pulse pressure, diastolic blood pressure, and atrial fibrillation, underscoring its role in cardiovascular regulation^5,50–52 (Fig. 5c and Supplementary Data 5).

We also identified one SNP (rs12568839 located in RBM15-AS1 on chromosome 1; p = 6.63 × 10⁻⁸) associated with traits indirectly influencing cardiovascular function (termed “Cardiovascular Associated Traits”, Fig. 5d). This variant was notably linked to serum nickel levels and the J-wave and T-wave interval⁵³, implicating potential pathways connecting electrolyte homeostasis and ventricular repolarisation dynamics with cardiovascular risk.

Conversely, the analysis in the EUR cohort identified 956 individually significant SNPs, including 386 novel variants previously unreported in GWAS Catalog or LDLink databases. Among the most significant novel findings were rs4811602 (p = 3.69 × 10⁻¹⁰⁷; Fig. 6a), rs35308076 (p = 1.09 × 10⁻⁵⁹), and rs11153730 (p = 8.66 × 10⁻⁵¹; Fig. 6b) (Supplementary Data 5 and Table 2). These variants represent compelling targets for further biological investigation to elucidate novel cardiovascular mechanisms.

Fig. 6. Regional Association Plots Highlighting Novel and Previously Identified MTAG Loci Associated with Cardiovascular Traits in European Population.

Regional association plots for MTAG loci significantly associated with cardiovascular traits in the African population. The lead SNPs (a) rs4811602 and (b) rs11153730 represent novel findings, while the lead SNPs (c) rs9320841 and (d) rs34732995 have been previously linked to cardiovascular traits. Genes within the chromosomal loci are depicted in the lower panel. The filled circles indicate the position of the SNPs across the region on the x-axis and the negative logarithm of the association p-value on the y-axis. The lead SNP is highlighted in purple, with other SNPs within the locus coloured based on their linkage disequilibrium correlation value (r²) with the lead SNP, utilizing the European HapMap haplotype from the 1000 Genomes Project.

Moreover, our EUR cohort analysis confirmed associations for 506 previously identified SNPs with cardiovascular traits, including highly significant variants such as rs62481856 (p = 7.54 × 10⁻¹¹⁹), rs9320841 (p = 1.50 × 10⁻¹¹⁷; Fig. 6c), and rs452036 (p = 1.88 × 10^-100). For instance, rs9320841 has known associations with heart rate and blood pressure phenotypes^46,54, further validating its critical role in cardiovascular health.

In the EUR cohort, we also identified 64 variants associated with traits known to indirectly influence cardiovascular function (designated ‘Cardiovascular-Associated Traits’). Prominent variants in this category include rs6040076 (P = 3.13 × 10⁻⁴²), rs55855614 (P = 6.47 × 10⁻²⁶), and rs34732995 (P = 1.82 × 10⁻²²; Fig. 6d). By annotating these SNPs against the GWAS Catalog, we confirmed they have strong, previously established associations with metabolic and anthropometric factors, including body mass index and weight.

Overall, our findings underscore substantial genetic diversity underlying cardiovascular risk in both African and European populations, revealing population-specific variants as well as shared loci. These data provide valuable new insights and potential avenues for personalised medicine approaches in cardiovascular disease prevention and management.

Population-specific differences in SNP frequencies associated with cardiovascular traits

Our analysis revealed notable allele frequency differences in SNPs associated with cardiovascular traits between AFR and EUR populations, highlighting the influence of genetic ancestry (Supplementary Fig. 6, Supplementary Data 2, and Supplementary Data 5). The SNPs with the highest frequency differences favouring the EUR population were rs2289125 (EUR: 0.79, AFR: 0.25), rs11632112 (EUR: 0.76, AFR: 0.26), and rs223116 (EUR: 0.74, AFR: 0.17). Conversely, SNPs that exhibited markedly higher frequencies in the AFR population included rs28623312 (AFR: 0.70, EUR: 0.07), rs35225531 (AFR: 0.71, EUR: 0.14), and rs11928580 (AFR: 0.69, EUR: 0.16). These pronounced disparities in allele frequencies highlight population-specific genetic variants. Furthermore, these genetic variants could have divergent effects on cardiovascular traits, potentially contributing to population-specific disease risk profiles and outcomes^55,56.

Functional annotation reveals population-specific genomic enrichment of cardiovascular trait-associated SNPs

We assessed the functional implications of SNPs associated with cardiovascular traits in AFR and EUR populations through functional annotation using FUMA. Our analysis uncovered distinct patterns of genomic enrichment between the two populations (Fig. 7).

Fig. 7. Functional Annotation of SNPs in AFR and EUR Populations.

a African (AFR) population SNP annotations with corresponding proportions shown in the bar graph and pie chart. The bar graph indicates the proportion of SNPs in various genomic regions, with asterisks denoting significance levels: ***p < 0.001, **p < 0.01, *p < 0.05. The pie chart shows the percentage distribution of SNPs across intronic, intergenic, and ncRNA-intronic regions. b European (EUR) population SNP annotations, presented in the same format as the AFR population. The significance levels are marked similarly on the bar graph, and the pie chart reflects the distribution within intronic, intergenic, and ncRNA-intronic regions. The heat map legend on the right side represents the -log2(Enrichment) value of SNPs in each annotation category, with 1 indicating the highest level of enrichment.

Consistent with expectations, most SNPs across both populations were predominantly located in non-coding regions, notably intergenic and intronic regions. However, significant population-specific differences emerged in the enrichment and distribution of these variants. The AFR population showed a significant enrichment of SNPs in intergenic regions (70.2%; enrichment = 1.51, p = 1.55 × 10⁻⁶), markedly higher than in the EUR population (33.62%; enrichment = 0.72, p < 0.0001). Conversely, SNPs in intronic regions were significantly enriched in the EUR population (49.36%; enrichment = 1.35, p < 0.0001) but notably underrepresented in the AFR population (9.62%; enrichment = 0.26, p = 5.54 × 10⁻¹⁰).

Further analysis revealed significant enrichment of SNPs in untranslated regions (UTRs) within the EUR population, specifically UTR3 (1.43%; enrichment = 1.53, p = 3.66 × 10⁻⁵³) and UTR5 (0.45%; enrichment = 1.58, p = 5.47 × 10⁻²⁰). In contrast, no SNPs were detected in these regions within the AFR population. Additionally, SNPs located in upstream (EUR: 1.25%, enrichment = 1.17, p = 2.39 × 10⁻⁸; AFR: 0.96%, enrichment = 0.92, p = 1) and downstream regions (EUR: 1.34%, enrichment = 1.18, p = 1.04 × 10⁻⁹; AFR: 2.88%, enrichment = 2.57, p = 0.11) showed differential patterns of enrichment, highlighting their potential functional significance, particularly in Europeans.

Overall, these findings highlight distinct genomic architectures underpinning cardiovascular traits in African and European populations, emphasizing the importance of population-specific genomic contexts in interpreting genetic associations with cardiovascular disease.

Deleterious variants associated with cardiovascular traits in African and European populations

We next investigated the proportion and distribution of deleterious SNPs associated with cardiovascular traits in the AFR and EUR populations, employing the Combined Annotation Dependent Depletion (CADD)⁵⁷ scores to predict variant pathogenicity. A higher fraction of deleterious variants was observed in the AFR population (57%) compared to the EUR population (51%) (Supplementary Fig. 7a). Examination of the distribution of CADD scores showed that both datasets exhibited a right-skewed pattern, characterised by a high frequency of variants with lower scores, decreasing gradually at higher scores (Supplementary Fig. 7b and 7c). However, the AFR dataset showed a steeper decline after the deleteriousness threshold, indicating fewer variants with moderate-to-high pathogenicity scores relative to the EUR dataset. These findings underscore subtle yet biologically relevant differences in the frequency and severity distribution of potentially deleterious genetic variants between these populations.

Population-specific genetic associations and functional enrichment

We employed FUMA³² to identify population-specific genetic associations by mapping MTAG-significant SNPs to genes. Our analysis highlighted substantial disparities between AFR and EUR populations, revealing 8 candidate genes uniquely associated with AFR and 2217 genes uniquely associated with EUR (Supplementary Fig. 8). Remarkably, there were no overlapping genes identified between the two populations, highlighting strongly divergent genetic contributions.

Further functional enrichment analysis using Enrichr⁵⁸ identified distinct enriched biological pathways for each population (Fig. 8a). Using the Gene Ontology Biological Process database⁵⁹, the AFR group was significantly enriched in pathways related to epigenetic modifications such as ‘Histone H3-K4 methylation’ (4.5 × 10⁻⁵) and chromatin remodelling (p = 0.003). Conversely, the EUR population showed significant enrichment for cardiovascular-specific processes including ‘heart development’ (p = 8.6 × 10⁻⁶) and ‘negative regulation of complement activation’ (7.6 × 10⁻⁵) (see Supplementary Data 6).

Fig. 8. Functional enrichment analyses for African and European populations.

Bar plots illustrating gene enrichment results based on MTAG analysis-derived SNPs mapped using FUMA. Functional enrichment analyses were conducted with Enrichr and included (a) Gene Ontology (GO) Biological Processes, (b) MGI Mammalian Phenotypes, (c) PhenGenI disease associations, and (d) GWAS Catalog terms derived from the UK Biobank. Comprehensive details are available in Supplementary Data 5.

Pathway analysis using the Mouse Genome Informatics (MGI) Mammalian Phenotype database⁶⁰ highlighted AFR-specific enrichments predominantly in immune-related phenotypes such as ‘myocardium hypoplasia’ (p = 0.002) and ‘abnormal innate immunity’ (p = 0.003) (Fig. 8b). In contrast, EUR-specific enrichments were strongly linked to cardiac morphology and function, including ‘dilated cardiomyopathy,’ (p = 8.5 × 10⁻⁸) ‘cardiac hypertrophy’ (p = 1.1 × 10⁻⁶), and ‘atrial fibrillation’, (p = 6.4 × 10⁻⁶).

We also investigated disease associations through the Phenotype-Genotype Integrator (PhenGenI) database⁶¹ (Fig. 8c). The AFR population presented diverse associations, with the strongest connections to hepatic and metabolic conditions such as ‘hepatitis C’ (p = 0.01), ‘lactate dehydrogenases’ (p = 0.02), and ‘liver cirrhosis’ (p = 0.02). In contrast, the EUR group was notably enriched for cardiovascular traits (p = 4.3 × 10⁻²⁰), including ‘heart rate’(p = 2.2 × 10⁻¹³), ‘arterial pressure’ (p = 2.5 × 10⁻⁷), and ‘hypertension’ (p = 7.9 × 10⁻⁷).

Analysis using the GWAS Catalog database reinforced these observations (Fig. 8d). AFR-specific associations were varied, reflecting non-cardiovascular traits such as ‘cystatin-F levels’ (p = 0.005), and ‘glucagon levels in response tests’ (p = 0.006). Conversely, the EUR group exhibited strong and numerous associations exclusively with cardiovascular traits, including ‘pulse pressure’ (p = 4.5 × 10⁻¹²²), ‘systolic blood pressure’ (p = 7.8 × 10⁻¹¹⁶), and ‘diastolic blood pressure’ (p = 9.0 × 10⁻¹¹⁵).

Finally, enrichment analysis using the UK Biobank GWAS dataset further supported these differences (Supplementary Fig. 9 and see Supplementary Data 6). The AFR group demonstrated enrichment in haematological and behavioural traits, notably ‘left mean signal-to-noise ratio’ (p = 0.02) and ‘reticulocyte percentage’ (p = 0.04). Meanwhile, the EUR group showed compelling enrichments specifically for cardiovascular traits, including ‘systolic blood pressure reading’ (p = 2.8 × 10⁻¹³⁰) and ‘automated pulse rate reading’ (p = 5.7 × 10⁻¹²⁰), and ‘diastolic blood pressure’ (p = 1.4 × 10⁻¹¹⁹).

Collectively, these findings highlight pronounced genetic and functional differences between AFR and EUR populations, with a strikingly stronger cardiovascular-related genetic predisposition in Europeans. These observations emphasise the critical importance of considering genetic ancestry in understanding the complex genetic architecture underlying cardiovascular health and related traits.

Discussion

Our study presents a comprehensive analysis of the genetic variants associated with cardiovascular traits across AFR and EUR populations in the UKB. Our findings highlight the substantial genetic diversity underlying these traits in different ethnic groups, revealing both shared and unique genetic determinants that can contribute to disparities in cardiovascular health outcomes.

The identification of novel SNPs associated with cardiovascular traits in both the AFR and EUR populations substantiates the power of MTAG to reveal novel genetic associations. Previous studies have employed MTAG to elucidate the genetic architecture of complex traits⁶², and our work extends these findings by identifying 45 novel variants in the AFR population and an impressive 957 novel variants in the EUR population.

These discoveries highlight the genetic diversity underlying traits and emphasize the importance of more diverse and inclusive genomic research. Previous literature has noted that a disproportionate focus on European-ancestry populations in genetic studies has limited our understanding of disease genetics across diverse populations¹⁸. Importantly, our findings caution against the uncritical application of genetic risk prediction models across populations, emphasizing the need for population-specific genetic research in precision medicine⁶³.

In addition, our study reveals remarkable differences in SNP frequencies between the AFR and EUR populations. These disparities, in line with previous findings^64–67, highlight the complex interplay between genetic background and cardiovascular health. However, the discovery of shared genetic factors among AFR and EUR populations hints at some shared genetic aetiology for specific traits, a finding that could potentially be leveraged for broad therapeutic interventions⁶⁸. The limited overlap of significant loci between the African and European ancestry cohorts aligns with the established principle that genetic architecture is not uniform across global populations, largely due to differences in linkage disequilibrium and allele frequencies. However, the specific loci that do replicate are of particular interest, as they may represent deeply conserved biological mechanisms essential for cardiovascular regulation across all ancestries. These shared regions provide high-confidence targets for functional investigation and potential therapeutic development that could be broadly applicable. The observation that these loci are often tagged by different lead SNPs in each population points towards allelic heterogeneity, where multiple functional variants within the same gene or regulatory region contribute to a trait.

Conversely, the large number of loci unique to each ancestry, particularly the 108 novel loci identified in the European cohort, highlights the profound challenge in transferring genetic discoveries across populations. The significant number of unique genes identified in each population highlights the influence of population-specific genetic factors, likely due to varying population history, environmental interactions, or statistical factors, such as sample size^63,69–71. These ancestry-specific signals underscore that a substantial portion of the genetic basis for cardiovascular traits is modified by population-specific evolutionary history and gene-environment interactions. This finding carries critical implications for precision medicine, reinforcing that polygenic risk scores and therapeutic strategies developed based on European-centric data are unlikely to be effective or equitable without specific research and calibration in diverse populations like those of African descent.

A key finding from our local genetic correlation analysis was the profound divergence in the shared architecture of SBP and DBP between ancestries. While we identified 181 distinct genomic loci exhibiting significant, predominantly positive, genetic correlation between SBP and DBP in the EUR sample, a striking absence of such signals was observed in the AFR sample. This disparity in local architecture is particularly notable given that genome-wide genetic correlations for blood pressure traits are generally high and consistent across populations⁷². This suggests that while the overall polygenic control of SBP and DBP may be shared, the specific regional genetic variants driving this relationship, and their correlational structure, may differ substantially or be more complex in AFR ancestry individuals.

Several factors could contribute to this observed difference. AFR ancestry populations are characterised by greater genetic diversity and lower average levels of LD compared to AFR populations⁷³. Consequently, a pleiotropic signal from a single causal variant may be captured within a well-defined LD block in EURs, while in AFRs, the same signal may be distributed across different, smaller blocks or be in linkage with different nearby variants, reducing the power to detect a significant local correlation. Furthermore, the disparity in statistical power, driven by the smaller sample size of the AFR cohort, undoubtedly limits the ability to detect more subtle local effects. These findings underscore the challenges in transferring genetic discoveries and prediction models, such as polygenic risk scores, across populations and highlight that an absence of evidence for local pleiotropy is not evidence of its absence⁶³.

Functional mapping and annotation of the SNPs reveal unique patterns of SNP distribution in non-coding regions across the AFR and EUR populations. The increasing recognition of the importance of non-coding regions in the regulation of gene expression and disease susceptibility^74,75 is supported by our findings. Notably, the associations spanned multiple traits, suggesting pleiotropic effects where a single gene or SNP can influence multiple traits⁴⁰. An example of such a phenomenon is the SNP rs10069648, which we found linked to diverse traits, including blood pressure regulation, medication use, and longevity^36,43. However, it is important to note that further studies are necessary to understand the biological mechanisms driving these associations³⁹.

The enrichment of distinct functional pathways in each population highlights potential divergences in the pathophysiology of cardiovascular diseases, indicative of a complex interplay between genetic and environmental factors that differ across populations. For instance, in the EUR population, we identified pathways associated with cardiomyopathy and limb-girdle muscular dystrophy, both of which have previously been associated with cardiovascular traits^76,77. Conversely, in the AFR population, BMI, body fat, enrichment of rasopathy and oxidative phosphorylation deficiency pathways were observed, which might suggest unique molecular mechanisms underlying cardiovascular disease in this population^78,79. The significant enrichment of the identified variants within gene sets known to influence cardiovascular parameters, such as blood pressure, pulse rate, and heart rate, in the EUR population provides further evidence of a nuanced genetic landscape. Notably, this enrichment aligns well with previously conducted GWAS, thus supporting our findings. However, this also highlights the extensive research focus on the EUR population in cardiovascular studies, emphasizing the need for greater efforts in investigating underrepresented populations, such as the AFR group.

Overall, our results, in which we found different variants associated with cardiovascular traits at the common genomic risk loci in AFR and EUR populations, are supported by the fact that a SNP can have different effects on a trait in different populations due to various factors. For example, a SNP that is associated with a trait in one population may not be associated with the same trait in another population if the SNP has a low frequency or is absent⁸⁰, is in LD with a causal variant in one population but not the other⁸¹, interacts with different environmental factors⁸², or has a different genetic background or epistatic effects with other SNPs that modulate the trait⁸³.

Our study is not without limitations. The functional implications and mechanistic roles of the identified genetic variants need further validation. This caveat is common to many GWAS studies and underlines the importance of combining GWAS with functional studies for a more comprehensive understanding of the genotype-phenotype relationship³⁹. Moreover, the smaller sample size of the AFR population may have limited our ability to identify all associated variants, echoing calls for larger and more diverse genetic studies⁸⁴.

In conclusion, our findings add to the growing body of literature describing the complexity and diversity of the genetic underpinnings of cardiovascular traits across different populations^85–87. This work lays a foundation for more precise and personalised approaches to preventing and managing cardiovascular diseases, ultimately aiming to reduce health disparities across different ethnic groups.

Limitations

A primary limitation of this study is the substantial sample size disparity between the European (n = 460,096) and African (n = 6551) ancestry cohorts, a known challenge reflecting historical biases in genomic databases. This results in considerably lower statistical power for discovery in the African ancestry group. Therefore, while our within-ancestry analyses are robust, any direct comparison—either in the number of significant loci or the magnitude of effect sizes—must be interpreted with caution, as a lack of a significant finding in the smaller cohort cannot be interpreted as evidence of a weaker biological effect.

Furthermore, our analysis was restricted to these two groups as other ancestries represented in the UK Biobank had insufficient sample sizes for adequately powered multi-trait analyses. We anticipate that forthcoming resources, such as the Pan-UK Biobank, will be critical for extending this work to create a more inclusive and globally representative understanding of cardiovascular genetic architecture.

Several additional considerations should be noted. First, although MTAG increases power by leveraging cross-trait genetic correlation, it assumes that the input traits share a degree of genetic architecture within each ancestry group; any ancestry-specific heterogeneity in genetic correlation structure may influence the degree of gain from multi-trait modelling and the comparability of signals across groups.

Second, our analysis is based on UK Biobank participants, which may not be representative of broader population distributions of cardiovascular traits due to volunteer selection and cohort-specific socio-environmental profiles. Accordingly, the transferability of effect estimates and locus prioritisation to non-UK settings, particularly to diverse African populations with distinct environmental exposures and demographic histories, should be evaluated in independent cohorts.

Third, phenotype heterogeneity remains a potential source of noise. Blood pressure and heart-rate measurements can be influenced by medication use, acute physiological state, measurement protocol differences, and residual confounding despite standard quality control procedures.

Fourth, locus interpretation relies on statistical fine-mapping surrogates and post-GWAS annotation (for example, nearest-gene mapping and functional annotation frameworks). These procedures are useful for prioritisation but do not establish causal variants, causal genes, or mechanistic pathways.

Finally, our conclusions regarding ancestry-specific versus shared loci are constrained by imputation quality, allele-frequency distribution, and linkage disequilibrium differences; therefore, apparent non-replication should be interpreted as “not detected under current power and design” rather than as definitive evidence of ancestry-specific biology.

Methods

Data Collection and Demographic Selection

We analysed datasets derived from the UK Biobank²⁷, including genotyping array data, SBP, DBP, pulse rate, MHR, and additional anthropometric measurements. Our investigations were constrained to a subset of these data, comprising 460,096 individuals of European descent (classified as White, British, Irish, or “any other White background”) and 6551 individuals of recent African ancestry (classified as Black). The specifics of UK Biobank participant selection, recruitment methodologies, and sample collection and analysis protocols have been comprehensively documented in prior publications²⁷. The ancestry groups were initially defined by self-identification. Then, as recommended by the UKB, a principal component analysis was performed, followed by a random forest on the projected principal component analysis data to reassign the initial self-defined ancestries of individuals with a membership posterior probability > 0.5. Other individuals with a posterior probability less than 0.5 for any given ancestry group were dropped from further analysis.

Comparison of cardiovascular traits in Europeans and Africans

We compared mean SBP, DBP, pulse rate and MHR between European (n = 460,096) and African (n = 6551) participants using Welch’s t-test (two-sample t-test with unequal variances). To further probe the relationships between these cardiovascular measures and other anthropometric traits, we calculated Pearson’s correlation coefficients, providing a measure of the linear dependencies between these variables.

Furthermore, to explore how cardiovascular traits vary across the body mass index (BMI) distribution, we plotted the population-level trends of SBP, DBP, pulse rate, and MHR across BMI percentile bins, separately for African and European ancestry groups. These plots reveal ancestry-specific patterns, particularly for DBP and pulse rate, and are provided in Supplementary Fig. 1 with 95% confidence intervals to illustrate uncertainty around the mean values.

BMI-stratified trend analysis of cardiovascular traits

To evaluate how BMI modulates cardiovascular physiology, we analysed resting pulse rate, SBP, DBP and MHR in African-ancestry and European-ancestry participants, after excluding any record with missing values. BMI values were converted to ordered 5‑percentile bins (5%, 10%, 95%) using empirical sample quantiles; each participant was assigned to the lowest bin whose upper bound they did not exceed. This ordinal variable was treated as the exposure. Within each ancestry we quantified linear trends with ordinary least-squares (OLS) regression,

$$Y={\beta }_o+{\beta }_1{BMI}_{pct}+\epsilon ,$$

where Y denotes the cardiovascular trait and BMI_pct percentile bin. The slope β̂₁ (change in trait per 5‑percentile step) and its two‑sided P value are reported in Supplementary Data 1. To verify that the associations were not driven by distributional assumptions, we also computed Spearman’s rank‑correlation coefficient (ρ) between BMI_pct and each trait. A relationship was deemed significant at α = 0.05 if either test rejected the null hypothesis of no monotonic trend.

Differences in slopes between ancestries were assessed in a combined model that included an interaction term,

$$Y={\beta }_o+{\beta }_1{BMI}_{pct}+{\beta }_2\mathrm{A}centry+{\beta }_3\left({BMI}_{pct}\times Ancentry\right)+\epsilon ,$$

where a significant β₃ indicates divergent BMI‑to‑trait gradients.

Genome-wide identification of genetic loci associated with cardiovascular traits

We collected GWAS summary statistics of cardiovascular traits, namely SBP, DBP, pulse rate and MHR, of 396,670 Europeans and 6551 Africans profiled by the UK Biobank. The approach towards participant genotyping within the UK Biobank has been thoroughly detailed in previous reports^27,88. Further information regarding the quality control measures employed in the genotyping analyses can be accessed at https://pan.ukbb.broadinstitute.org/docs/qc. GWAS summary statistics, as calculated by the UK Biobank project for each cardiovascular trait, were used in our investigation.

The GWAS protocol applied to the cardiovascular and ancestry groups, and the specifics of the analytical techniques are extensively documented in earlier publications⁸⁹. The procedure, in brief, implemented the Scalable and Accurate Implementation of the Generalized Mixed Model approach³⁷. A linear or mixed logistic model was adopted, incorporating a kinship matrix as a random effect and certain covariates as fixed factors. The covariates incorporated in the model include the participant’s age and sex, the interaction term of age and sex, squared values of age, the interaction of squared age and sex, and the initial ten principal components derived from the genotype datasets.

We processed these summary statistics through a two-stage approach. In the first stage, we conducted separate single GWA for African and European populations, examining associations with SBP, DBP, pulse rate, and MHR.

In the second stage, we applied MTAG across all four traits to boost the discovery power for SNPs and investigate the genetic underpinnings of crucial cardiovascular traits. Utilising summary statistics from the UK Biobank, we analysed AFR and EUR populations separately, thereby capturing population-specific genetic variation. The MTAG analyses were implemented through the JASS (Joint Analysis of Summary Statistics) software³³, adhering to default settings to ensure the consistency and replicability of the findings. The creation of Manhattan plots was facilitated using MATLAB, utilising the software delineated in prior literature⁹⁰.

Fine-Mapping and Functional Annotation of Genetic Associations

Intending to identify independent SNP associations and genomic risk loci, along with mapping the consequential genes, we employed Functional Mapping and Annotation (FUMA)³² version v1.3.7, an integrative platform for functional annotation, for our subsequent analyses. This was executed utilising the MTAG summary statistics, distinctly for the AFR and EUR populations.

We applied FUMA to facilitate a comprehensive interpretation of the GWA results. The analysis pipeline started with the independent SNP identification based on the single trait and multi-trait GWA results. We then applied the default FUMA settings to identify lead SNPs, selecting variants with a p-value < 1 × 10⁻⁵ and an LD r² > 0.1 within ±500 kb in the AFR and EUR populations separately. SNPs with a minor allele frequency (MAF) < 0.01 were excluded. SNPs meeting these criteria were considered independent signals, consistent with the LD clumping approach described by Watanabe et al. (2017)³², which balances signal resolution with control for multiple testing. Post-identification, these independent SNPs were aligned into genomic risk loci.

After isolating the lead SNPs, we used FUMA to project each variant onto the reference genome, assign its nearest or credible effector gene, and retrieve functional annotations (coding consequence, eQTL evidence and chromatin-state overlap). This procedure linked the AFR and EUR lead SNPs to candidate genes that plausibly modulate vascular tone or cardiac excitability. A complete listing of every SNP, its mapped gene, predicted functional class, and enriched pathway term is provided in Supplementary Table 3.

In summary, the exhaustive analyses in FUMA, employing the MTAG summary statistics distinctly for AFR and EUR populations, encompassed the identification of lead and independent SNPs, delineating genomic risk loci, assessing the functional impact of SNPs, and mapping related genes.

Local SNP-heritability and genetic correlation using LAVA GWAS summary statistics for SBP, DBP, pulse rate, and MHR were harmonised using the munge_sumstats.py pipeline from LD Score Regression (LDSC, v1.0.1)⁹¹. Prior to harmonisation, variants were filtered to include only bi-allelic sites with a high imputation quality score (INFO ≥ 0.9) and a minor allele frequency > 1% in the 1000 Genomes Phase 3 reference panel (GRCh37).

Local genetic correlation analysis was performed using LAVA (Local Analysis of [co]Variant Association, v0.1.5)⁴² independently for EUR and AFR ancestry samples. The European analysis utilised the 1700 established linkage disequilibrium (LD) blocks derived from the 1000 Genomes Phase 3 panel⁹². The genome was divided into various LD-independent blocks using the default LAVA partitioning. For the African ancestry analysis, a corresponding set of LD blocks was generated by applying the same partitioning algorithm to the 1000 Genomes Phase 3 African reference. To account for potential sample overlap between traits, genetic covariance intercepts were calculated via cross-trait LDSC and supplied to the LAVA model, along with trait-specific effective sample sizes and the appropriate PLINK⁹³ reference panel for each ancestry.

For each genomic block, LAVA was used to estimate the univariate local SNP-heritability (h²_loc for each trait and the bivariate local genetic correlation (ρ_loc)for each trait pair. A two-stage analytical approach was employed to assess statistical significance while controlling for multiple testing.

First, a permissive filter was applied at the univariate stage: loci were only considered for bivariate analysis if both traits in a given pair exhibited at least nominal evidence of local heritability (unadjusted P < 0.05). Second, for all trait pairs passing this initial filter, the p-values from the subsequent bivariate correlation tests were adjusted using the Benjamini-Hochberg procedure across all tests within each ancestry. A false discovery rate (FDR) q < 0.05 was used as the threshold for significant local genetic correlation.

Finally, local pleiotropy between cardiovascular traits was assessed exclusively using bivariate local genetic correlation analysis implemented in the LAVA framework. This approach quantifies shared genetic effects within predefined LD-based genomic blocks by estimating local SNP-heritability and local bivariate genetic correlation between trait pairs. Importantly, this analysis evaluates regional genetic coupling at the level of polygenic signal, rather than testing individual variant effects.

Conditional local genetic correlation analysis. To evaluate whether the observed local pleiotropy between systolic and diastolic blood pressure (SBP–DBP) is independent of pulse rate and maximum heart rate, we estimated the full 4×4 genetic covariance matrix in each of the 2495 LD-independent genomic blocks using the estimate. locus function in LAVA. From this matrix, we computed unconditional local genetic correlations (ρ_SBP − DBP), partial genetic correlations conditioning on pulse rate, maximum heart rate, or both, and a joint heritability test across all four traits (χ² on 10 df). Analyses were performed separately for African and European ancestry individuals. P-values were corrected for multiple testing using the Benjamini–Hochberg false discovery rate (FDR < 0.05) within each ancestry.

Visualisation and Annotation of Local Genetic Architecture To visualise the landscape of local genetic correlations, results were presented as Manhattan plots generated using custom MATLAB scripts. To integrate these findings with known genome-wide significant signals, each linkage disequilibrium block was annotated for the presence of an independent lead SNP (P < 1 × 10⁻⁵) from FUMA³² v1.3.8 analyses applied to the MTAG analyses results obtained through the JASS. Loci harbouring a genome-wide significant variant were highlighted on the plots to distinguish them from other regions. FDR < 0.05, applied to all tested loci and trait pairs within each ancestry, was used to define significant local genetic correlations.

To characterise shared pleiotropic architecture, co-occurrence matrices of significant local correlations were constructed for each trait pair. These matrices were stratified by the direction of the genetic correlation (positive or negative) and visualised as heatmaps to facilitate comparison of shared architectural patterns across ancestries. Additionally, trait-specific enrichment of local heritability was summarised by quantifying the significant locus and variant counts per trait from the univariate LAVA output.

Categorisation of novel and known variants leveraging LDlink and FUMA

In our study, we identified novel and reported (known) variants from LDLink (v2.0)⁹⁴, an online suite of applications that provides easy access to population-specific LD information. This was accomplished by separately utilising the list of lead SNPs from the Multitrait GWAS with cardiovascular traits in AFR and EUR populations.

The independently significant SNPs were first obtained from FUMA outputs for the AFR and EUR analysis, and then we automated retrieving data from LDlink’s RESTful API using a cURL command for each SNP. In this query, each SNP was tested against the GWAS Catalog⁷⁰ using the LDtrait software⁹⁵, which provides information on known trait associations. Then, in the context of the AFR and EUR populations, we applied the respective AFR and EUR reference panels in LDlink to identify comparable variants. The approach was optimized by employing a window size of 500 kilobase pairs and implementing an r² threshold of 0.1 for LD calculations. We deemed SNPs novel if there were no previously reported significant variants in the GWAS Catalog (version March-2025) or previous studies reported in LDlink (v2.0) that fall within ± 1 Mb from the lead variant.

The significant lead SNPs we identified associated with cardiovascular traits were classified into three categories: ‘Novel’, ‘Has Trait’, and ‘Cardiovascular Risk Associated’. The ‘Novel’ category comprises SNPs not previously identified in the GWAS Catalog or the literature. The ‘Has Trait’ category included SNPs associated with specific traits. Finally, the ‘Cardiovascular Risk Associated’ category, designed per the secondary analysis, included SNPs directly associated with cardiovascular or systemic traits like body weight and BMI.

Cross-Population Replication of Cardiovascular Trait-Associated Variants

To deepen our understanding of population-specific genetic underpinnings, we sought to replicate the significant findings of cardiovascular traits from EUR within AFR populations and vice versa.

Our replication methodology centred on the examination of linked variants. For lead variants that displayed a significant association with cardiovascular traits, we identified all associated variants within an LD block (r² > 0.1) extending ± 500 KB from the lead variant. We then assessed the association of these linked variants within the EUR population with the same trait within the AFR population and vice versa. This was achieved by extracting the corresponding GWAS p-values from the AFR dataset.

To account for multiple testing and control the false discovery rate, we adjusted these p-values using the Benjamini and Hochberg correction procedure⁹⁶. Variants within each LD block were deemed to replicate if they met an FDR-adjusted significance threshold of q < 0.05 (adjusted P < 0.05) and exhibited the same effect direction (either positive or negative) as the original lead SNP. This approach balances error control with power in the smaller AFR sample, providing a transparent and reproducible standard for declaring local replication.

Identification and Functional Annotation of Cardiovascular Trait-Associated Genetic Variants

We individually analysed the AFR and EUR cohorts, extracting genetic variants related to cardiovascular traits from the MTAG summary statistics obtained using JASS. Following this extraction, we applied established variant-to-gene mapping techniques in FUMA to elucidate the genes connected to these variants in both populations. Our next step was to assess the enrichment of the mapped genes, employing the Enrichr⁵⁸ platform. We explored a variety of databases – Gene Ontology Biological Process⁵⁹, Mouse Genome Informatics (MGI) Mammalian Phenotype database⁶⁰, Phenotype-Genotype Integrator (PhenGenI) database⁶¹, and UK Biobank gene set term defined using GWAS p-values, computed by Maya’an Lab⁵⁸, for an array of UKB phenotypes⁵⁸ – for this purpose. This comprehensive strategy enabled us to identify enriched annotation terms, further refining our understanding of the biological implications of our findings.

Quantitative Analysis of Deleterious SNPs Based on CADD Scores

To quantify the prevalence of deleterious mutations within the AFR and EUR populations, we employed the Combined Annotation Dependent Depletion (CADD)⁵⁷ scores, a widely recognized metric for assessing the deleterious potential of SNPs. A threshold CADD score of 20 was adopted to delineate the deleterious variants, as scores above this threshold are typically considered within the top 1% of deleterious variants in the human genome.

We processed the genetic variants datasets for both populations, identifying SNPs and extracting their corresponding CADD scores. The fraction of deleterious SNPs was computed by dividing the number of SNPs exceeding the threshold by the total number of SNPs in the respective dataset. Additionally, we visualised the distribution of CADD scores using histograms, with bin widths set to 1 to capture the granularity of the data.

Statistics and Reproducibility

Data processing and visualisation were conducted in MATLAB (v2024b). Inferential statistics and multiple testing corrections were performed in MATLAB and R (v4.1.2); and enrichment analyses were executed in Python (v3.11.5). Automation of the pipeline was achieved via Bash (v5.1) scripting and MATLAB. Detailed software versions and dependencies are provided in the Supplementary Information. Our established significance threshold was a two-sided p-value of less than 0.05 for individual comparisons. In addition, to account for the potential inflation of the type I error rates inherent in multiple comparisons, we implemented the Benjamini & Hochberg procedure⁹⁶ to control the false discovery rate, thus ensuring the robustness of our findings.

The GWAS summary statistics used in this study were derived from the UK Biobank, which employs standardised genotyping protocols and quality control procedures. Multi-trait analyses were conducted using JASS with default parameters, and results were replicated across independent runs. Functional annotation via FUMA was performed using consistent reference panels and parameter settings. Local genetic correlation analyses via LAVA were run separately for each ancestry using ancestry-matched LD reference panels. Sample sizes for all analyses are as follows: European ancestry, n = 460,096 (GWAS), n = 396,670 (MTAG); African ancestry, n = 6551 (GWAS and MTAG). No data were excluded from the analyses beyond the quality control filters described above.

Ethics Approval

This research was conducted using data from the UK Biobank Resource under Application Number 53163. The UK Biobank has received ethical approval from the Northwest Multi-centre Research Ethics Committee (REC ref. ²¹:/NW/0157). All participants provided written informed consent for the use of their data in health-related research. In addition, the study was reviewed and approved by the University of Cape Town Human Research Ethics Committee (HREC reference: 739/2021; IRB00001938). All procedures and analyses were carried out in accordance with institutional guidelines, national research regulations, and the principles of the Declaration of Helsinki. All ethical regulations relevant to human research participants were followed.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

M.S., S.E. and N.M. conceptualized the study. M.S., N.M., S.E. and J.C. designed the methodology, and M.M. M.S., M.M., J.C. and S.E. formally analysed the data. M.S., N.M. and S.E. drafted the manuscript. Editing and reviewing the manuscript were carried out by M.S., N.M., S.E., J.C. and M.M. Data visualisations were produced by M.S. and S.E.

Peer review

Peer review information

Communications Biology thanks Kruthika Iyer and the other, anonymous, reviewers for their contribution to the peer review of this work. Primary Handling Editors:Laura Rodríguez Pérez.

Data availability

The datasets that support the results presented in this manuscript are available from: the UK Biobank; https://www.ukbiobank.ac.uk, dbSNP; https://www.ncbi.nlm.nih.gov/snp, and the GWAS Catalog; https://www.ebi.ac.uk/gwas. UK Biobank data is accessible exclusively via the UKB platform and only through an approved project plan as outlined in its access policy. In compliance with these restrictions, all UK Biobank data used in this study remain within the UKB infrastructure and have not been deposited in external repositories. The numerical source data underlying all graphs and charts presented in the main figures and Supplementary Figs. are provided in Supplementary Data 1–6. Furthermore, the GWAS summary statistics derived by the Pan-UK Biobank project’s⁸⁹ for the four cardiovascular parameters are available via the Amazon Web Services links: SBP: https://pan-ukb-us-east-1.s3.amazonaws.com/sumstats_flat_files/continuous-4080-both_sexes-irnt.tsv.bgz DBP: https://pan-ukb-us-east-1.s3.amazonaws.com/sumstats_flat_files/continuous-4079-both_sexes-irnt.tsv.bgz Pulse rate: https://pan-ukb-us-east-1.s3.amazonaws.com/sumstats_flat_files/continuous-102-both_sexes-irnt.tsv.bgz MHR: https://pan-ukb-us-east-1.s3.amazonaws.com/sumstats_flat_files/continuous-6033-both_sexes-irnt.tsv.bgz.

Code availability

All custom code and analysis scripts written in MATLAB, R, and BASH for this project are available on GitHub. The repository, located at https://github.com/smsinks/Multi-Ancestry-GWAS-of-Cardiovascular-Traits, contains the necessary files to replicate the phenotypic comparisons, multi-trait analyses, and post-GWAS processing described herein. Publicly available software used in this study (JASS, FUMA, LAVA, LDSC) are cited in the Methods section. A persistent archived copy of the code has been deposited on Zenodo here: 10.5281/zenodo.19073142⁹⁷.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s42003-026-09977-1.