Transferability of genetic loci and polygenic scores for cardiometabolic traits in British Pakistani and Bangladeshi individuals

Abstract

Individuals with South Asian ancestry have a higher risk of heart disease than other groups but have been largely excluded from genetic research. Using data from 22,000 British Pakistani and Bangladeshi individuals with linked electronic health records from the Genes & Health cohort, we conducted genome-wide association studies of coronary artery disease and its key risk factors. Using power-adjusted transferability ratios, we found evidence for transferability for the majority of cardiometabolic loci powered to replicate. The performance of polygenic scores was high for lipids and blood pressure, but lower for BMI and coronary artery disease. Adding a polygenic score for coronary artery disease to clinical risk factors showed significant improvement in reclassification. In Mendelian randomisation using transferable loci as instruments, our findings were consistent with results in European-ancestry individuals. Taken together, trait-specific transferability of trait loci between populations is an important consideration with implications for risk prediction and causal inference.

Fig. 1. Summary of study design, research questions and analyses conducted.

The coloured boxes indicate input data. Within the white boxes, the black text indicates the analyses we used to address the questions in blue. BPB British Pakistanis and Bangladeshi ancestry, EUR European ancestry, SAS South Asian ancestry, CAD coronary artery disease, BMI body mass index, SNP single nucleotide polymorphism, GWAS genome-wide association study, MR Mendelian randomisation, PGS polygenic score, UKBB UK Biobank. Datasets and discovery GWAS that were used in each analysis are provided in Supplementary Data 4.

Fig. 2. SNP heritability and trans-ancestry genetic correlations for cardiometabolic traits.

a SNP heritability was estimated using GCTA in G&H (orange) and eMERGE (grey) for cardiometabolic traits, namely coronary artery disease (CAD; n = 17,348 and 32,816 unrelated samples from G&H and eMERGE, respectively), body-mass index (BMI; n = 13,926 and 37,160), high-density lipoprotein cholesterol (HDL-C; n = 11,316 and 16,049), low-density lipoprotein cholesterol (LDL-C; n = 12,856 and 15,856), triglycerides (TG; n = 11,125 and 14,384), systolic blood pressure (SBP), and diastolic blood pressure (DBP; n = 15,908 and 11,864 for blood pressure). Medication data are not available in eMERGE so the highest measurements for LDL-C, SBP, and DBP are used. Error bars represent 95% confidence intervals in both plots. b Genetic correlations were estimated using Popcorn based on GWAS summary statistics generated from G&H and European-ancestry individuals from UK Biobank. Red indicates that the genetic correlation is nominally significantly lower than 1 (p-value = 0.02 for BMI; two-sided and not adjusted for multiple comparisons). Medication-adjusted lipid and blood pressure levels are used. For r_g estimates of 1 (TG and DBP), the method cannot derive confidence intervals. Sample sizes of GWAS for cardiometabolic traits in G&H are in Table 1.

Fig. 3. Comparison of the predictive accuracy of polygenic scores in people of British Pakistani and Bangladeshi versus European ancestry.

Incremental AUC is shown for coronary artery disease (CAD; n = 17,348 and 32,816 unrelated samples from G&H and eMERGE, respectively) and Incremental R² is shown for its continuous risk factors, namely body-mass index (BMI; n = 13,926 and 37,160), high-density lipoprotein cholesterol (HDL-C; n = 11,316 and 16,049), low-density lipoprotein cholesterol (LDL-C; n = 12,856 and 15,856), triglycerides (TG; n = 11,125 and 14,384), systolic blood pressure (SBP), and diastolic blood pressure (DBP; n = 15,908 and 11,864 for blood pressure). Grey indicates European-ancestry (EUR) individuals from eMERGE and orange British Pakistani and Bangladeshi (BPB) individuals from G&H. Error bars represent 95% confidence intervals estimated by bootstrap resampling of samples. The highest measurements for LDL-C, SBP, and DBP are compared between eMERGE and G&H.

Fig. 4. Net reclassification index (NRI) for coronary artery disease with the addition of a polygenic score to QRISK3.

Estimates for categorical NRI for the integrated score compared to QRISK3 in all samples (n = 420 unrelated cases and 7702 unrelated non-cases) as well as in age-by-sex subgroups (n = 207 and 2779 in males aged 25–54; n = 51 and 4187 in females aged 25–54; n = 114 and 344 in males aged 55–84; n = 48 and 392 in females aged 55–84) are shown. Red indicates NRI in cases and blue in controls. The error bars indicate 95% confidence intervals estimated using the bootstrap method.

Fig. 5. Mendelian randomisation estimates of risk factors on coronary artery disease in European (eMERGE) and British South Asian (G&H) ancestry individuals.

Two-sample Mendelian randomisation (MR) estimates for the causal effects are presented based on genetic instrument variables identified from EUR discovery GWAS for each risk factor. All independent genome-wide significant loci were used as instruments for eMERGE and only the transferable loci for G&H. Effect estimates are presented as odds ratios with 95% confidence intervals per standard deviation increase in the reported unit of the trait: triglycerides (TG), systolic blood pressure (SBP), low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), diastolic blood pressure (DBP), body mass index (BMI). The two-sided p-value (P; not adjusted for multiple comparisons) and the number of single nucleotide polymorphism instruments (N SNPs) included in the MR analysis are shown for each exposure. GWAS for CAD was performed in n = 22,008 (1110 cases) samples from G&H, and n = 32,816 (6815 cases) unrelated samples from eMERGE.