A Multi-Ancestry Polygenic Risk Score for Coronary Heart Disease Based on an Ancestrally Diverse Genome-Wide Association Study and Population-Specific Optimization

Johanna L. Smith, Catherine Tcheandjieu, Ozan Dikilitas, Kruthika Iyer, Kazuo Miyazawa, Austin Hilliard, Julie Lynch, Jerome I. Rotter, Yii-Der Ida Chen, Wayne Huey-Herng Sheu, Kyong-Mi Chang, Stavroula Kanoni, Phil Tsao, Kaoru Ito, Matthew Kosel, Shoa L. Clarke, Daniel J. Schaid, Themistocles L. Assimes, Iftikhar J. Kullo
doi: https://doi.org/10.1101/2023.06.02.23290896

Abstract

Background Predictive performance of polygenic risk scores (PRS) varies across populations. To facilitate equitable clinical use, we developed PRS for coronary heart disease (PRSCHD) for 5 genetic ancestry groups.

Methods We derived ancestry-specific and multi-ancestry PRSCHD based on pruning and thresholding (PRSP+T) and continuous shrinkage priors (PRSCSx) applied on summary statistics from the largest multi-ancestry genome-wide meta-analysis for CHD to date, including 1.1 million participants from 5 continental populations. Following training and optimization of PRSCHD in the Million Veteran Program, we evaluated predictive performance of the best performing PRSCHD in 176,988 individuals across 9 cohorts of diverse genetic ancestry.

Results Multi-ancestry PRSP+T outperformed ancestry specific PRSP+T across a range of tuning values. In training stage, for all ancestry groups, PRSCSx performed beter than PRSP+T and multi-ancestry PRS outperformed ancestry-specific PRS. In independent validation cohorts, the selected multi-ancestry PRSP+T demonstrated the strongest association with CHD in individuals of South Asian (SAS) and European (EUR) ancestry (OR per 1SD[95% CI]; 2.75[2.41-3.14], 1.65[1.59-1.72]), followed by East Asian (EAS) (1.56[1.50-1.61]), Hispanic/Latino (HIS) (1.38[1.24-1.54]), and weakest in African (AFR) ancestry (1.16[1.11-1.21]). The selected multi-ancestry PRSCSx showed stronger association with CHD in comparison within each ancestry group where the association was strongest in SAS (2.67[2.38-3.00]) and EUR (1.65[1.59-1.71]), progressively decreasing in EAS (1.59[1.54-1.64]), HIS (1.51[1.35-1.69]), and lowest in AFR (1.20[1.15-1.26]).

Conclusions Utilizing diverse summary statistics from a large multi-ancestry genome-wide meta-analysis led to improved performance of PRSCHD in most ancestry groups compared to single-ancestry methods. Improvement of predictive performance was limited, specifically in AFR and HIS, despite use of one of the largest and most diverse set of training and validation cohorts to date. This highlights the need for larger GWAS datasets of AFR and HIS individuals to enhance performance of PRSCHD.

Introduction

Coronary heart disease (CHD) is a leading cause of death in the United States (U.S.) and worldwide 1. CHD has an estimated heritability of 40-60% and the majority of the heritable risk is atributable to a polygenic component, i.e., the aggregation of modest effects across many genetic variants 2. Polygenic risk scores (PRS) capture a proportion of that heritability and are typically constructed by summing the products of the effect-size and the number of risk alleles at associated loci 3,4. PRS for CHD have evolved over the last decade as progressively larger genome wide association studies (GWAS) have been reported 5-8. These PRS have been evaluated in several studies and are associated with incident CHD independent of conventional risk factors such as hypertension, hypercholesterolemia, diabetes, and smoking as well as family history of CHD 8-10.

Most PRS for CHD have been developed, optimized, and validated in cohorts consisting largelyof individuals of European (EUR) ancestry (here and throughout the manuscript ‘ancestry’ refers to genetic ancestry) 11-14. Furthermore, the portability of these PRS to non-EUR groups is impacted by differences in allele frequencies (AF), effect sizes, and linkage disequilibrium (LD) paterns across ancestry groups, typically resulting in reduced predictive performance as studied populations diverge in these factors; an observation most notable between EUR and African (AFR) ancestry populations 6,11,15. We previously observed significantly lower performance of several EUR-derived PRS for CHD in AFR ancestry individuals 16,17. To prevent exacerbation of health disparities in the context of genomic medicine, there is a need to improve performance of PRS for CHD for non-EUR populations.

In this study, we leveraged a large scale, ancestrally diverse genome-wide meta-analysis for CHD to construct PRS for CHD optimized for EUR, AFR, Hispanic/Latino (HIS), East Asian (EAS), and South Asian (SAS) ancestries. To this end, we utilized two PRS derivation methods, pruning and thresholding (P+T) and the continuous shrinkage prior based PRS-CSx 8,18. We assessed the performance of the multi-ancestry PRS in individuals with diverse ancestry belonging to 8 independent validation cohorts. Finally, a PRS was selected for clinical implementation in the electronic Medical Records and Genomics (eMERGE) network phase IV study in which PRS-informed risk profiles for several common conditions are being returned to participants 19.

Methods

GWAS Summary Statistics for PRS Development

We developed PRS using both ancestry-specific and multi-ancestry meta-analysis summary statistics from a large-scale multi-ancestry GWAS for CHD including 1.1 million diverse participants with 243,392 CHD cases 17. This diverse meta-analysis included 17,202 AFR, 6,378 HIS, 29,319 EAS, and 190,776 EUR individuals with CHD belonging to four cohorts including the Million Veteran Program (MVP), the UK Biobank (UKBB), CARDIoGRAMplusC4D Consortium (2015 release), and Biobank Japan (BBJ) (Figure 1) 17,20-22.

Figure 1.
Figure 1.

Polygenic Risk Score development using independent MVP cohorts of diverse ancestry.

We used two distinct methods to construct PRS, namely, pruning and thresholding (P+T) and the continuous shrinkage prior based PRS-CSx 8,18. Ancestry-specific PRS were defined from ancestry-specific GWAS summary statistics (i.e., EUR specific summary statistics were used to develop a EUR specific PRS), and multi-ancestry PRS were defined as PRS derived from multi-ancestry summary statistics. These PRS were then trained and optimized in a separate set of individuals from the MVP and externally validated in several diverse cohorts including the Atherosclerosis Risk in Communities (ARIC) 23, Multi-Ethnic Study of Atherosclerosis (MESA) 24, Cardiovascular Health Study (CHS) 25, Women’s Health Initiative (WHI) 26, eMERGE Phases I-III genotyped cohort 27, Biobank Japan (BBJ) 28, Osaka Acute Coronary Insufficiency (OACIS) study 29, the TAICHI Consortium 30, and individuals of SAS ancestry from the UKBB 31 (Table S1; Supplemental File 1).

Pruning and Thresholding (P+T)

We derived two independent sets of PRS (ancestry-specific and multi-ancestry PRS) in two sequential steps: First, we excluded from the base GWAS summary statistics, correlated single nucleotide variants (SNVs) by LD pruning, applying 4 different R2 thresholding values (0.2, 0.5, 0.8, and 0.9) and 2 different window distances (250kb and 500kb) within which these R2 were applied. LD pruning for ancestry-specific PRS was performed based on reference panels comprised of 4,000 participants from each respective ancestry (EUR, AFR, HIS, and ASN), selected among MVP participants included in the large-scale GWAS for CHD. The LD pruning for the multi-ancestry PRS was performed on the full subset of 16,000 individuals from EUR, AFR, HIS, and ASN as the reference panel. This step generated 8 ancestry-specific summary statistics and 8 multi-ancestry summary statistics for PRS development. Second, for each newly generated summary statistic from step 1, we applied 16 different p-value thresholds (5×10−08, 1×10−04, 0.001, 0.005, 0.01, 0.05, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, and 1) (Figure S1; Supplemental File 1). These led to 128 summary statistics within each ancestry, which were used to train the ancestry-specific PRS. Similarly, we obtained 128 multi-ancestry-based summary statistics to train the multi-ancestry PRS (PRSP+T).

Continuous shrinkage (PRS-CSx)

We applied a continuous shrinkage method, PRS-CSx (PRSCSx), on the effect sizes of a subset of 1.4 million well curated HapMap SNVs on each ancestry-specific summary statistic. To identify the optimal shrinkage parameter, we applied 4 different global shrinkage phi parameters (1, 1e−02, 1e−04, and 1e−06). LD reference panels used were EUR, AFR, AMR and EAS from the 1000 Genomes project. The multi-ancestry PRS were constructed from the meta-analysis of ancestry-specific summary statistics obtained after applying the global shrinkage phi. For each ancestry, 4 ancestry-specific newly derived summary statistics were obtained to train ancestry-specific PRS and 4 newly derived multi-ancestry summary statistics were obtained for train the multi-ancestry PRS (Figure S2; Supplemental File 1). A total of 12 ancestry-specific PRS (one for each global shrinkage parameter value used for each ancestry group and 4 multi-ancestry PRS) were chosen for further development (Figure S3; Supplemental File 1).

PRS Training

Following the construction of the ancestry-specific and multi-ancestry PRSP+T and PRSCSx across a range of training specifications, we proceeded to assess their performance in an independent set of prevalent cases and controls from the MVP (Figure 1B, PRS Training) using multivariable logistic regression with adjustment for age at CHD event for cases and age at the last visit in the electronic health record (EHR) for controls, year of birth, sex, and the first 5 principal components (PCs). We compared parameter training on the multi-ancestry reference panel set versus population-specific reference panel. Ancestry-specific PRS were evaluated in the corresponding ancestry, whereas the multi-ancestry PRS were evaluated in each ancestry. PRS with the highest observed odds ratio (OR) for CHD per 1 standard deviation (SD) increase were deemed to have the optimal training parameter values across ancestry populations and subsequently advanced for validation.

PRS Validation in the Million Veteran Program and Additional External Cohorts

Ancestry-specific and multi-ancestry PRSP+T and PRSCSx trained for each genetic ancestry group were validated in an independent cohort from the MVP and several additional diverse cohorts (Figure 1C, Diverse Cohorts for PRS Validation). The MVP validation cohort was restricted to incident cases of CHD occurring after enrollment, and random controls, in a ratio of 1:10 (Figure 1C) as previously described 17. Four prospective cohorts, namely ARIC, MESA, CHS, WHI, a subset of the UKBB comprised of individuals of SAS ancestry, and additionally eMERGE Phases I-III, contributed CHD incident cases and controls of EUR, AFR, HIS, and SAS ancestry for PRS validation. Validation for EAS ancestry included individuals from multiple case-control studies, namely Han Chinese participants from Taiwan as a part of the TAICHI consortium, as well as Japanese participants from the BBJ and OACIS studies who were not part of the multi-ancestry discovery GWAS 30.

Within MVP, we used diagnosis and procedure codes to identify individuals with any clinical manifestation of CHD as previously described (Supplemental File 1) 17. This definition included both ‘hard’ (e.g., myocardial infarction, revascularization) and ‘sotf’ outcomes (e.g., angina, non-invasive study positive for ischemia). In the 4 external validation NHLBI cohorts and the eMERGE cohort, cases were restricted to myocardial infarction and revascularization. Prevalent cases were defined as all other cases meeting diagnosis/procedure code criteria at the time of enrollment. Additional study details are included in Supplemental File 1.

We calculated OR per 1-SD increase in PRS using multivariable logistic regression across all validation cohorts. The dbGaP, eMERGE, and UKBB cohorts were adjusted for genetic ancestry using a continuous correction further defined in the Supplemental File 1 (Figure S4). The two EAS case-control studies were meta-analyzed using a fixed effect inverse-variance weighted model 32. For all external validation cohorts, we additionally estimated OR for CHD for participants in the top 5% of PRS distribution compared to the rest, as well as area under the curve (AUC) discrimination statistic.

Calibration was also assessed using the calibration function in the rms package in R to assess portability to cohorts that were not available for meta-analysis (i.e., the non-EAS cohorts) (Figure S5, Supplemental File 2) 33,34.

Results

PRS Training

Pruning and Thresholding (P+T)

Performance of the ancestry-specific and multi-ancestry PRSP+T in each population is shown in Figure 2. The multi-ancestry PRSP+T systematically outperformed ancestry-specific PRSP+T with noticeably higher OR per SD except for the HIS ancestry group where the performance was similar (Figure 2, Supplemental Figure S2). The multi-ancestry PRSP+T, performed best in HIS population, followed by the ASN population (1.78 and 1.73 OR per SD, respectively) (Supplemental File 2). Prediction performance of the PRSP+T for each ancestry was optimal at different p-value thresholds (Figure 2, Supplemental Figure S2). The multi-ancestry PRSP+T performed best at R2 ≤ 0.8 with LD blocks of 250 kb, p-value threshold of 0.01 for AFR, 0.03 for EUR, and 0.30 for HIS. However, the differences between these PRS and the PRS optimized at R2 ≤ 0.8 and a p-value = 0.01 were marginal, and the multi-ancestry PRS with a p-value threshold of 0.01 was chosen for validation in additional external cohorts.

Figure 2.
Figure 2.

Performance of PRS-CSx (solid bars) or P+T (dashed bars) across genetic ancestry groups when utilizing the diverse MVP training cohort. The colors represent the GWAS summary statistics used to construct the PRS (green for AFR, purple for EAS, orange for EUR, and grey for the multi-ancestry meta-analysis). The Odds Ratios (ORs) per 1 standard deviation (SD) increase with confidence intervals (CIs) in the PRS are represented on the Y-axis and the populations on which the PRS is trained are on the X-axis.

Continuous shrinkage (PRS-CSx)

The performances of the 12 ancestry-specific PRSCSx and 3 multi-ancestry PRSCSx built using EUR, AFR, HIS, and EAS summary statistics at various global shrinkage phi values for tuning (1e−02, 1e−04, and 1e−06) are shown in Figure 2. For all ancestry groups, phi = 1e−02 resulted in the best predictive performance for PRSCSx and the multi-ancestry PRS outperformed ancestry-specific PRS at this phi value. For the EUR population, both the EUR-derived PRS and the multi-ancestry PRS performed similarly, but ASN and HIS populations performed best with the EUR-derived PRS, while the AFR population performed best with the multi-ancestry PRS (Figure 2, Supplemental File 2). Overall, the multi-ancestry PRSCSx for the ASN population resulted in the highest OR per/SD increase followed by EUR and HIS populations where the strength of association was similar, and lowest in the AFR ancestry.

PRS Validation

Million Veteran Program

Ancestry specific PRSP+T predictive performance (OR per 1 -SD increase) for EUR (1.52), AFR (1.19), and HIS (1.81) was compared to the ancestry-specific PRSCSx performance for EUR (1.66), AFR (1.15), HIS (1.42), and ASN (1.32) (Figure 2; Supplemental File 2). This was also compared to the multi-ancestry-based methods using the same PRS training, i.e., the multi-ancestry PRSP+T for EUR (1.57), AFR (1.22), HIS (1.78), and ASN (1.73), as well as PRSCSx for EUR (1.98), AFR (1.23), HIS (1.94), and ASN (2.06) (Figure 2; Supplemental File 2). Of all the methods assessed at this step, the best performing methods tended to be the multi-ancestry PRSCSx and multi-ancestry PRSP+T. However, there were overlapping confidence intervals (CIs) with some single ancestry methods and the single-ancestry PRSCSx for EUR performed well in other ancestries, so we decided to further assess the three methods (Figure 2).

We advanced the ancestry optimized PRSP+T and PRSCSx, for validation in an independent setof incident cases and matching controls in ancestry groups of EUR, AFR, HIS, EAS, and SAS individuals.

Predictive performances of the multi-ancestry PRS were assessed within each ancestry group in reference to a previously reported genome-wide PRS (i.e., PRSmetaGRS 10) constructed using a cohort of predominantly of EUR ancestry (Figure 3) 17. In this independent validation cohort, the multi-ancestry PRSP+T and PRSCSx had a higher predictive performance compared to metaGRS (Figure 3). The multi-ancestry PRSCSx had a relative increase in the estimated OR per 1-SD of 12% and 23% in reference to PRSP+T and PRSmetaGRS, respectively, averaged across all three genetic ancestries.

Figure 3.
Figure 3.

Comparison of a prior PRS (metaGRS) and two new PRS using multi-ancestry summary statistics for the prediction of coronary heart disease (CHD) using the ancestrally diverse training cohort of the MVP. Odds Ratios (ORs) per standard deviation (SD) with confidence intervals (CIs) are shown for each genetic ancestry group as determined in the methods as a result of metaGRS, P+T, and PRS-CSx PRS methods being performed on the MVP training cohort.

Additional External Validation Cohorts

The best performing PRSP+T were further validated in several additional cohort and case-control studies of CHD including EUR, AFR, HIS, EAS, and SAS participants (Table 1). ORs for ancestry-specific and multi-ancestry PRSP+T ranged from 1.16 in AFR to 2.75 in SAS and were comparable to published reports, despite inclusion of the diverse meta-analysis of GWAS (Supplemental File 2) 6,17,35,36. All populations had OR estimates for the top 5% vs the rest of the population ≥ 2.16 for PRSP+T except for AFR (1.68).

View this table:
Table 1.

Odds Ratios for incident CHD for multi-ancestry PRSP+T and PRSCSx in diverse ancestry cohorts.

The two best performing PRSCSx in the training dataset, a EUR-tuned PRS and a multi-ancestry PRS, both with a tuning global phi value of 1e−02, demonstrated similar performances in our validation cohorts (Table 1, Table S2; Supplemental File 1) as the multi-ancestry PRS marginally outperformed the EUR-tuned PRS in all but the AFR and HIS cohorts. Point estimates of the OR for subjects in the top 5th percentile of scores compared to the remaining participants shitied trend compared to those observed for the ORs per 1-SD for AFR, HIS, and SAS populations, but these differences were in the context of mostly overlapping 95% confidence intervals. When comparing the multi-ancestry PRSP+T to PRSCSx, the point estimates of ORs were similar but higher for the multi-ancestry PRSCSx for EUR, AFR, HIS, and EAS populations. The OR per 1-SD was lower for the multi-ancestry PRSCSx for the SAS population (Table 1).

Discussion

Using summary statistics from the largest multi-ancestry GWAS meta-analysis for CHD to date and 9 independent validations cohorts, cumulatively comprised of 1.1 million diverse participants including nearly a quarter of a million CHD cases of EUR, AFR, HIS, EAS, and SAS descent 17, we developed, trained, and validated multi-ancestry and ancestry-specific PRS models to address the gap in predictive performance that currently exists between EUR and non-EUR ancestries.

We observed that the use of summary statistics from a multi-ancestry GWAS meta-analysis, in comparison to the use of ancestry-specific summary statistics, improved PRS performance in majority of the ancestry groups. PRS that leveraged shared information between ancestries to estimate SNV weights (i.e., PRSCSx) modestly outperformed the P+T method (i.e., PRSP+T). Based on the multi-ancestry informed PRSCSx, individuals in the high-genetic risk group (i.e., top 5% of the PRS distribution) compared to the remaining participants in the respective ancestry group (EUR, AFR, HIS, EAS, and SAS), had 2.5-fold, 1.7-fold, 2.5-fold, 2.3-fold, and 5-fold increased risk of CHD, respectively. These results collectively highlight complementary effects of integrating summary statistics from multiple ancestries and the use of PRS derivation methods that leverage shared information and LD diversity between ancestry groups to improve polygenic risk prediction for CHD.

Although remarkable progress has been achieved to date in both genomic discovery and polygenic risk prediction among EUR cohorts 5,7-10,37-39, similar progress has not occurred among non-EUR populations due to their underrepresentation in genomic studies 11-14. In recent years, the number of large-scale multi-ancestry GWAS and polygenic risk prediction studies have increased with the establishment of ancestrally diverse biobanks and collaborations efforts 17,18,30,40-43. Several multi-ancestry genomic studies, including for glycemic, hematologic and lipid traits as well as disease phenotypes such as type 2 diabetes and CHD, have increased the number of discovered loci, and improved fine-mapping and cross-population polygenic risk prediction with inclusion of non-EUR participants 17,40-42,44. Our findings are consistent with these results in that integration of summary statistics from several distinct ancestry groups improved predictive performance of PRS for all ancestries, including EUR descent. One possible explanation for these observations is identification of potential causal variants that are more likely to be shared between ancestries but are obscured by population-specific LD paterns 14,45. Another likely contributing factor to improved PRS performance is reduced noise in SNV effect size estimates resulting from both weighted average of population-level estimates and increased total sample size 46,47.

Despite the use of the largest ancestrally diverse cohort available to date, the improvement in the predictive performance of PRSCHD was limited in individuals of AFR ancestry compared to other ancestry groups. Prior reports investigating portability of PRS between populations noted that prediction performance across a range of traits and phenotypes 6,11,15,16,48,49 decayed with increasing genetic distance between study cohorts. Among the continental ancestry groups included in this study, AFR is the most genetically distant population from EUR and hence the modest increase in prediction performance with a multi-ancestry PRSCHD compared to the ancestry-specific counterpart. A recent report showed similar heritability for CHD in the major continental ancestry groups but absence of two common haplotypes at the 9p21 locus in AFR individuals, which corresponds to the largest effect locus in EUR ancestry individuals 17. These findings suggest potentially a larger role of ancestry-specific causal variants in individuals of African origin with regards to heritability for CHD.

Although the strength of association of PRS with CHD varied between ancestry groups, it is important to consider epidemiological differences in CHD risk across these populations. In clinical practice, primary prevention guidelines for CHD use absolute risk estimates for clinical decision making, such as 10-year or lifetime risk of a CHD event 50. Individuals are typically classified into different risk groups (e.g., low, borderline, intermediate, high risk) with a correlating intensity of pursued preventive measures. In the United States, African American and South Asian populations have substantially higher atherosclerotic cardiovascular disease (ASCVD) related mortality rates compared to non-Hispanic whites 1,51. Therefore, in a future risk model for ASCVD similar to the pooled cohort equation 52, incorporation of a PRS for CHD with a narrower risk gradient in African Americans, compared to a much wider gradient in non-Hispanic whites, could have more impact on re-classification into a higher risk group as we have previously shown 6.

Implementation of PRS in the clinical seting has begun for CHD, including at Mayo Clinic, where a PRS for CHD is available in the clinical seting, based on the results of the MIGENES clinical trial 53. The eMERGE Network, in its phase IV study is returning risk assessments to participants for 11 common conditions, including CHD 19. The multi-ancestry PRSP+T for CHD validated in this study 19 will be returned to eMERGE participants. One of the major challenges in the clinical use of PRS include variable performance between genetic ancestry populations 11,15. Developing robust PRS for diverse ancestry groups is crucial to avoid worsening existing health disparities 11 and a National Institute of Health (NIH) funded initiative is addressing this as a priority 54. The active recruitment and inclusion of diverse participants and continued development of novel PRS methods that target improvement of cross-population prediction using a variety of approaches (e.g., incorporation of local ancestry 55, weighting by trans-ancestry genetic correlation 56, and informing by fine-mapping and functional annotation 57,58) will be necessary for equitable implementation of PRS. Consequently, we anticipate that PRS for CHD will continue to evolve and improve over time.

Study Limitations

Despite the large and diverse composition of our study, the external validation for the SAS ancestry was limited to a single cohort with a modest number of cases, reducing the precision of the associated risk estimates. We were not able to include smoking status or family history in the models as the data was not available for all cohorts, and this may have affected the strength of the association of PRS with CHD in our analyses.

Conclusions

We demonstrated that incorporation of summary statistics from diverse genetic ancestry groups, as opposed to individual ancestry groups alone, and leveraging shared information between these populations, led to improved performance of PRSCHD in majority of the ancestry groups. Despite utilization of one of the largest and most ancestrally diverse set of training and validation cohorts to date, the gain in predictive performance for AFR was limited. Ongoing work is needed to narrow the persistent performance gap for AFR ancestry individuals. Increasing AFR representation at each stage of PRS development is necessary to lessen performance disparities, and such efforts should be a priority for the community of genomics researchers.

Data Availability

All data produced in the present work are contained in the manuscript

Sources of Funding

This work was supported by grants from the Polygenic Risk Methods in Diverse Populations (PRIMED) Consortium through the National Human Genome Research Institute (NHGRI): grant U01 HG11710, the electronic Medical Records and Genomics (eMERGE) Network funded by the NHGRI: grant U01 HG06379, a National Heart, Lung, and Blood: grant K24 HL137010, the Clinical Genome Resource (ClinGEN) funded by the NHGRI: grant HG09650, and R35 GM140487.

Disclosures

Conflict of Interest

The authors declare that they have no conflict of interest.

Human and Animal Rights and Informed Consent

This article used data from previously published human studies.

Acknowledgements

We acknowledge the investigators and participants of the electronic Medical Records and Genomics (eMERGE) Network. Infrastructure for the CHARGE Consortium is supported in part by the National Heart, Lung, and Blood Institute (NHLBI) grant R01HL105756. This work was also supported in part by the National Institutes of Health, National Heart, Lung, Long and Blood Institute (NHLBI) contract 1R01HL151855, R01HL146860, and the National Institute of Diabetes and Digestive and Kidney Diseases contract UM1DK078616.

Footnotes

  • ↵* Co-first Authors

References

  1. 1.↵
    Tsao CW, Aday AW, Almarzooq ZI, Anderson CAM, Arora P, Avery CL, Baker-Smith CM, Beaton AZ, Boehme AK, Buxton AE, et al. Heart Disease and Stroke Statistics—2023 Update: A Report From the American Heart Association. Circulation. 2023;147. doi: 10.1161/cir.0000000000001123
    OpenUrlCrossRef
  2. 2.↵
    Kullo IJ, Ding K. Mechanisms of Disease: the genetic basis of coronary heart disease. Nature Clinical Practice Cardiovascular Medicine. 2007;4:558–569. doi: 10.1038/ncpcardio0982
    OpenUrlCrossRef
  3. 3.↵
    Euesden J, Lewis CM, O’Reilly PF. PRSice: Polygenic Risk Score sotiware. Bioinformatics. 2015;31:1466–1468. doi: https://doi.org/10.1093/bioinformatics/btu848
    OpenUrlCrossRefPubMed
  4. 4.↵
    Kullo IJ, Lewis CM, Inouye M, Martin AR, Ripati S, Chaterjee N. Polygenic scores in biomedical research. Nature Reviews Genetics. 2022. doi: 10.1038/s41576-022-00470-z
    OpenUrlCrossRef
  5. 5.↵
    Tikkanen E, Havulinna AS, Palotie A, Salomaa V, Ripati S. Genetic Risk Prediction and a 2-Stage Risk Screening Strategy for Coronary Heart Disease. Arteriosclerosis, Thrombosis, and Vascular Biology. 2013;33:2261–2266. doi: 10.1161/atvbaha.112.301120
    OpenUrlAbstract/FREE Full Text
  6. 6.↵
    Dikilitas O, Schaid DJ, Tcheandjieu C, Clarke SL, Assimes TL, Kullo IJ. Use of Polygenic Risk Scores for Coronary Heart Disease in Ancestrally Diverse Populations. Current Cardiology Reports. 2022;24:1169–1177. doi: 10.1007/s11886-022-01734-0
    OpenUrlCrossRef
  7. 7.↵
    O’Sullivan JW, Raghavan S, Marquez-Luna C, Luzum JA, Damrauer SM, Ashley EA, O’Donnell CJ, Willer CJ, Natarajan P. Polygenic Risk Scores for Cardiovascular Disease: A Scientific Statement From the American Heart Association. Circulation. 2022;146. doi: 10.1161/cir.0000000000001077
    OpenUrlCrossRef
  8. 8.↵
    Khera AV, Chaffin M, Aragam KG, Haas ME, Roselli C, Choi SH, Natarajan P, Lander ES, Lubitz SA, Ellinor PT, et al. Genome-wide polygenic scores for common diseases identify individuals with risk equivalent to monogenic mutations. Nature Genetics. 2018;50:1219–1224. doi: 10.1038/s41588-018-0183-z
    OpenUrlCrossRef
  9. 9.
    Rapati S, Tikkanen E, Orho-Melander M, Havulinna AS, Silander K, Sharma A, Guiducci C, Perola M, Jula A, Sinisalo J, et al. A multilocus genetic risk score for coronary heart disease: case-control and prospective cohort analyses. The Lancet. 2010;376:1393–1400.
    OpenUrl
  10. 10.↵
    Inouye M, Abraham G, Nelson CP, Wood AM, Sweeting MJ, Dudbridge F, Lai FY, Kaptoge S, Brozynska M, Wang T, et al. Genomic Risk Prediction of Coronary Artery Disease in 480,000 Adults: Implications for Primary Prevention. J American Coll Cardiol. 2018;72:1883–1893.
    OpenUrl
  11. 11.↵
    Martin AR, Kanai M, Kamatani Y, Okada Y, Neale BM, Daly MJ. Clinical use of current polygenic risk scores may exacerbate health disparities. Nature Genetics. 2019;51:584–591. doi: 10.1038/s41588-019-0379-x
    OpenUrlCrossRefPubMed
  12. 12.
    Manolio TA. Using the Data We Have: Improving Diversity in Genomic Research. The American Journal of Human Genetics. 2019;105:233–236. doi: 10.1016/j.ajhg.2019.07.008
    OpenUrlCrossRef
  13. 13.
    Clarke SL, Assimes TL, Tcheandjieu C. The Propagation of Racial Disparities in Cardiovascular Genomics Research. Circulation: Genomic and Precision Medicine. 2021;14. doi: 10.1161/circgen.121.003178
    OpenUrlCrossRef
  14. 14.↵
    Gurdasani D, Barroso I, Zeggini E, Sandhu MS. Genomics of disease risk in globally diverse populations. Nature Reviews Genetics. 2019;20:520–535. doi: 10.1038/s41576-019-0144-0
    OpenUrlCrossRef
  15. 15.↵
    Martin AR, Gignoux CR, Walters RK, Wojcik GL, Neale BM, Gravel S, Daly MJ, Bustamante CD, Kenny EE. Human Demographic History Impacts Genetic Risk Prediction across Diverse Populations. The American Journal of Human Genetics. 2017;100:635–649. doi: 10.1016/j.ajhg.2017.03.0041.
    OpenUrlCrossRefPubMed
  16. 16.↵
    Dikilitas O, Schaid DJ, Kosel ML, Carroll RJ, Chute CG, Denny JC, Fedotov A, Feng Q, Hakonarson H, Jarvik GP, et al. Predictive Utility of Polygenic Risk Scores for Coronary Heart Disease in Three Major Racial and Ethnic Groups. The American Journal of Human Genetics. 2020;106:707–716. doi: 10.1016/j.ajhg.2020.04.002
    OpenUrlCrossRefPubMed
  17. 17.↵
    Tcheandjieu C, Zhu X, Hilliard AT, Clarke SL, Napolioni V, Ma S, Lee KM, Fang H, Chen F, Lu Y, et al. Large-scale genome-wide association study of coronary artery disease in genetically diverse populations. Nature Medicine. 2022;28:1679–1692. doi: 10.1038/s41591-022-01891-3
    OpenUrlCrossRef
  18. 18.↵
    Ge T, Irvin MR, Patki A, Srinivasasainagendra V, Lin Y-F, Tiwari HK, Armstrong ND, Benoit B, Chen C-Y, Choi KW, et al. Development and validation of a trans-ancestry polygenic risk score for type 2 diabetes in diverse populations. Genome Medicine. 2022;14. doi: 10.1186/s13073-022-01074-2
    OpenUrlCrossRef
  19. 19.↵
    Linder J, Allworth A, Bland ST, Caraballo PJ, Chisholm R, Clayton EW, Crosslin D, Dikilitas O, DiVietro A, Esplin ED, et al. Returning integrated genomic risk and clinical recommendations: the eMERGE study. Genetics in Medicine. 2023. doi: https://doi.org/10.1016/j.gim.2023.100006
  20. 20.↵
    Van Der Harst P, Verweij N. Identification of 64 Novel Genetic Loci Provides an Expanded View on the Genetic Architecture of Coronary Artery Disease. Circulation Research. 2018;122:433–443. doi: 10.1161/circresaha.117.312086
    OpenUrlAbstract/FREE Full Text
  21. 21.
    Ishigaki K, Akiyama M, Kanai M, Takahashi A, Kawakami E, Sugishita H, Sakaue S, Matoba N, Low S-K, Okada Y, et al. Large-scale genome-wide association study in a Japanese population identifies novel susceptibility loci across different diseases. Nature Genetics. 2020;52:669–679. doi: 10.1038/s41588-020-0640-3
    OpenUrlCrossRef
  22. 22.↵
    Nikpay M, Goel A, Won H-H, Hall LM, Willenborg C, Kanoni S, Saleheen D, Kyriakou T, Nelson CP, Hopewell JC, et al. A comprehensive 1000 Genomes–based genome-wide association meta-analysis of coronary artery disease. Nature Genetics. 2015;47:1121–1130. doi: 10.1038/ng.3396
    OpenUrlCrossRefPubMed
  23. 23.↵
    The ARIC Investigators. The Atherosclerosis Risk in Communities (ARIC) Study: design and objectives. American Journal of Epidemiology. 1989;129:687–702.
    OpenUrlCrossRefPubMed
  24. 24.↵
    Bild DE, Bluemke DA, Burke GL, Detrano R, Diez Roux AV, Folsom AR, Greenland P, R. Jd, Kronmal R, Liu K, et al. Multi-Ethnic Study of Atherosclerosis: Objectives and Design. American Journal of Epidemiology. 2002;156:871–881. doi: https://doi.org/10.1093/aje/kwf113
    OpenUrlCrossRefPubMedWeb of Science
  25. 25.↵
    (CHS) MtiCHSRG, Fried LP, Borhani NO, Enright P, Furberg CD, Gardin JM, Kronmal RA, Kuller LH, Manolio TA, Mitelmark MB, et al. The cardiovascular health study: Design and rationale. Annals of Epidemiology. 1991;1:263–276. doi: https://doi.org/10.1016/1047-2797(91)90005-W
    OpenUrlCrossRefPubMed
  26. 26.↵
    Group TWsHIS. Design of the Women’s Health Initiative Clinical Trial and Observational Study. Controlled Clinical Trials. 1998;19:61–109. doi: https://doi.org/10.1016/S0197-2456(97)00078-0
    OpenUrlCrossRefPubMedWeb of Science
  27. 27.↵
    Stanaway IB, Hall TO, Rosenthal EA, Palmer M, Naranbhai V, Knevel R, Namjou-Kahles B, Carroll RJ, Kiryluk K, Gordon AS, et al. The eMERGE genotype set of 83,717 subjects imputed to ⋃40 million variants genome wide and association with the herpes zoster medical record phenotype. Genetic Epidemiology. 2019;43:63–81. doi: 10.1002/gepi.22167
    OpenUrlCrossRefPubMed
  28. 28.↵
    Nagai A, Hirata M, Kamatani Y, Muto K, Matsuda K, Kiyohara Y, Ninomiya T, Tamakoshi A, Yamagata Z, Mushiroda T, et al. Overview of the Biobank Japan Project: Study design and profile. Journal of Epidemiology. 2017;27:S2–S8. doi: 10.1016/j.je.2016.12.005
    OpenUrlCrossRefPubMed
  29. 29.↵
    Kurotobi T, Sato H, Kinjo K, Nakatani D, Mizuno H, Shimizu M, Imai K, Hori M, Group O. Reduced Collateral Circulation to the Infarct-Related Artery in Elderly Patients with Acute Myocardial Infarction. J American Coll Cardiol. 2004;44:28–34. doi: doi:10.1016/j.jacc.2003.11.066
    OpenUrlFREE Full Text
  30. 30.↵
    Assimes TL, Lee IT, Juang J-M, Guo X, Wang T-D, Kim ET, Lee W-J, Absher D, Chiu Y-F, Hsu C-C, et al. Genetics of Coronary Artery Disease in Taiwan: A Cardiometabochip Study by the Taichi Consortium. PLOS ONE. 2016;11:e0138014. doi: 10.1371/journal.pone.01380141.
    OpenUrlCrossRef
  31. 31.↵
    Sudlow C, Gallacher J, Allen N, Beral V, Burton P, Danesh J, Downey P, Elliot P, Green J, Landray M, et al. UK Biobank: An Open Access Resource for Identifying the Causes of a Wide Range of Complex Diseases of Middle and Old Age. PLOS Medicine. 2015;12:e1001779. doi: 10.1371/journal.pmed.1001779
    OpenUrlCrossRefPubMed
  32. 32.↵
    Evangelou E, Ioannidis JPA. Meta-analysis methods for genome-wide association studies and beyond. Nature Reviews Genetics. 2013;14:379–389. doi: 10.1038/nrg347233.
    OpenUrlCrossRefPubMed
  33. 33.↵
    Harrell Jr. FE. rms: Regression Modeling Strategies. R package version 6.3-0. 2022.
  34. 34.↵
    Van Calster B, McLernon DJ, Van Smeden M, Wynants L, Steyerberg EW. Calibration: the Achilles heel of predictive analytics. BMC Medicine. 2019;17. doi: 10.1186/s12916-019-1466-7
    OpenUrlCrossRefPubMed
  35. 35.↵
    Mars N, Kerminen S, Feng Y-CA, Kanai M, Lall K, Thomas LF, Skogholt AH, dellaBriota Parolo P, Project TBJ FinnGen, et al. Genome-wide risk prediction of common diseases across ancestries in one million people. Cell Genomics. 2022;2.
  36. 36.↵
    Wang M, Menon R, MSanghamitra M, Patel AP, Chaffin M, Tanneeru D, Deshmukh M, Mathew O, Apte S, Devanboo CS, et al. Validation of a Genome-Wide Polygenic Score for Coronary Artery Disease in South Asians. Journal of the American College of Cardiology. 2020;76:703–714. doi: https://doi.org/10.1016/j.jacc.2020.06.024
    OpenUrlFREE Full Text
  37. 37.↵
    Tada H, Melander O, Louie JZ, Catanese JJ, Rowland CM, Devlin JJ, Kathiresan S, Shiffman D. Risk prediction by genetic risk scores for coronary heart disease is independent of self-reported family history. European Heart Journal. 2016;37:561–567. doi: 10.1093/eurheartj/ehv462
    OpenUrlCrossRefPubMed
  38. 38.
    Ding K, Bailey KR, Kullo IJ. Genotype-informed estimation of risk of coronary heart disease based on genome-wide association data linked to the electronic medical record. BMC Cardiovascular Disorders. 2011;11:66. doi: 10.1186/1471-2261-11-66
    OpenUrlCrossRefPubMed
  39. 39.↵
    Abraham G, Havulinna AS, Bhalala OG, Byars SG, De Livera AM, Yetukuri L, Tikkanen E, Perola M, Schunkert H, Sijbrands EJ, et al. Genomic prediction of coronary heart disease. European Heart Journal. 2016;37:3267–3278. doi: 10.1093/eurheartj/ehw450
    OpenUrlCrossRefPubMed
  40. 40.↵
    Mahajan A, Spracklen CN, Zhang W, Ng MCY, Pety LE, Kitajima H, Yu GZ, Rüeger S, Speidel L, Kim YJ, et al. Multi-ancestry genetic study of type 2 diabetes highlights the power of diverse populations for discovery and translation. Nature Genetics. 2022;54:560–572. doi: 10.1038/s41588-022-01058-3
    OpenUrlCrossRef
  41. 41.
    Chen J, Spracklen CN, Marenne G, Varshney A, Corbin LJ, Luan JA, Willems SM, Wu Y, Zhang X, Horikoshi M, et al. The trans-ancestral genomic architecture of glycemic traits. Nature Genetics. 2021;53:840–860. doi: 10.1038/s41588-021-00852-9
    OpenUrlCrossRefPubMed
  42. 42.↵
    Chen M-H, Raffield LM, Mousas A, Sakaue S, Huffman JE, Moscati A, Trivedi B, Jiang T, Akbari P, Vuckovic D, et al. Trans-ethnic and Ancestry-Specific Blood-Cell Genetics in 746,667 Individuals from 5 Global Populations. Cell. 2020;182:1198–1213.e1114. doi: 10.1016/j.cell.2020.06.045
    OpenUrlCrossRefPubMed
  43. 43.↵
    Lu X, Liu Z, Cui Q, Liu F, Li J, Niu X, Shen C, Hu D, Huang K, Chen J, et al. A polygenic risk score improves risk stratification of coronary artery disease: a large-scale prospective Chinese cohort study. European Heart Journal. 2022;43:1702–1711. doi: https://doi.org/10.1093/eurheartj/ehac093
    OpenUrlCrossRefPubMed
  44. 44.↵
    Graham SE, Clarke SL, Wu K-HH, Kanoni S, Zajac GJM, Ramdas S, Surakka I, Ntalla I, Vedantam S, Winkler TW, et al. The power of genetic diversity in genome-wide association studies of lipids. Nature. 2021;600:675–679. doi: 10.1038/s41586-021-04064-3
    OpenUrlCrossRef
  45. 45.↵
    Evans DM, Cardon LR. A Comparison of Linkage Disequilibrium Paterns and Estimated Population Recombination Rates across Multiple Populations. The American Journal of Human Genetics. 2005;76:681–687. doi: 10.1086/4292741.
    OpenUrlCrossRefPubMedWeb of Science
  46. 46.↵
    Cavazos TB, Wite JS. Inclusion of variants discovered from diverse populations improves polygenic risk score transferability. HGG Adv. 2021;2:100017. doi: https://doi.org/10.1016/j.xhgg.2020.100017
    OpenUrl
  47. 47.↵
    Zhang Y, Qi G, Park J-H, Chaterjee N. Estimation of complex effect-size distributions using summary-level statistics from genome-wide association studies across 32 complex traits. Nature Genetics. 2018;50:1318–1326. doi: 10.1038/s41588-018-0193-x
    OpenUrlCrossRefPubMed
  48. 48.↵
    Privé F, Aschard H, Carmi S, Folkersen L, Hoggart C, O’Reilly PF, Vilhjálmsson BJ. Portability of 245 polygenic scores when derived from the UK Biobank and applied to 9 ancestry groups from the same cohort. The American Journal of Human Genetics. 2022;109:12–23. doi: 10.1016/j.ajhg.2021.11.008
    OpenUrlCrossRef
  49. 49.↵
    Fahed AC, Aragam KG, Hindy G, Chen Y-DI, Chaudhary K, Dobbyn A, Krumholz HM, Sheu WHH, Rich SS, Roter JI, et al. Transethnic Transferability of a Genome-Wide Polygenic Score for Coronary Artery Disease. Circulation: Genomic and Precision Medicine. 2021;14. doi: 10.1161/circgen.120.003092
    OpenUrlCrossRef
  50. 50.↵
    Arnet DK, Blumenthal RS, Albert MA, Buroker AB, Goldberger ZD, Hahn EJ, Himmelfarb CD, Khera AV, Lloyd-Jones D, McEvoy JW, et al. 2019 ACC/AHA Guideline on the Primary Prevention of Cardiovascular Disease: A Report of the American College of Cardiology/American Heart Association Task Force on Clinical Practice Guidelines. Journal of the American College of Cardiology. 2019;74:177–232. doi: https://doi.org/10.1016/j.jacc.2019.03.010
    OpenUrlFREE Full Text
  51. 51.↵
    Volgman AS, Palaniappan LS, Aggarwal NT, Gupta M, Khandelwal A, Krishnan AV, Lichtman JH, Mehta LS, Patel HN, Shah KS, et al. Atherosclerotic Cardiovascular Disease in South Asians in the United States: Epidemiology, Risk Factors, and Treatments: A Scientific Statement From the American Heart Association. Circulation. 2018;138:CIR.00000000000. doi: 10.1161/cir.0000000000000580
    OpenUrlCrossRef
  52. 52.↵
    Goff DC, Lloyd-Jones DM, Bennet G, Coady S, D’Agostino RB, Gibbons R, Greenland P, Lackland DT, Levy D, O’Donnell CJ, et al. 2013 ACC/AHA Guideline on the Assessment of Cardiovascular Risk. Circulation. 2014;129:S49–S73. doi: 10.1161/01.cir.0000437741.48606.98
    OpenUrlFREE Full Text
  53. 53.↵
    Kullo IJ, Jouni H, Olson JE, Montori VM, Bailey KR. Design of a randomized controlled trial of disclosing genomic risk of coronary heart disease: the Myocardial Infarction Genes (MI-GENES) study. BMC Medical Genomics. 2015;8. doi: 10.1186/s12920-015-0122-0
    OpenUrlCrossRefPubMed
  54. 54.↵
    The Polygenic Risk Methods in Diverse Populations (PRIMED) Consortium. https://primedconsortium.org/. 2021.
  55. 55.↵
    Atkinson EG, Maihofer AX, Kanai M, Martin AR, Karczewski KJ, Santoro ML, Ulirsch JC, Kamatani Y, Okada Y, Finucane HK, et al. Tractor uses local ancestry to enable the inclusion of admixed individuals in GWAS and to boost power. Nature Genetics. 2021;53:195–204. doi: 10.1038/s41588-020-00766-y
    OpenUrlCrossRef
  56. 56.↵
    Cai M, Xiao J, Zhang S, Wan X, Zhao H, Chen G, Yang C. A unified framework for cross-population trait prediction by leveraging the genetic correlation of polygenic traits. The American Journal of Human Genetics. 2021;108:632–655. doi: 10.1016/j.ajhg.2021.03.002
    OpenUrlCrossRef
  57. 57.↵
    Weissbrod O, Kanai M, Shi H, Gazal S, Peyrot WJ, Khera AV, Okada Y, Matsuda K, Yamanashi Y, Furukawa Y, et al. Leveraging fine-mapping and multipopulation training data to improve cross-population polygenic risk scores. Nature Genetics. 2022;54:450–458. doi: 10.1038/s41588-022-01036-9
    OpenUrlCrossRef
  58. 58.↵
    Amariuta T, Ishigaki K, Sugishita H, Ohta T, Koido M, Dey KK, Matsuda K, Murakami Y, Price AL, Kawakami E, et al. Improving the trans-ancestry portability of polygenic risk scores by prioritizing 1. variants in predicted cell-type-specific regulatory elements. Nature Genetics. 2020;52:1346–1354. doi: 10.1038/s41588-020-00740-8
    OpenUrlCrossRef
Back to top
Posted June 06, 2023.
Download PDF

Supplementary Material

Data/Code
Email
A Multi-Ancestry Polygenic Risk Score for Coronary Heart Disease Based on an Ancestrally Diverse Genome-Wide Association Study and Population-Specific Optimization
Johanna L. Smith, Catherine Tcheandjieu, Ozan Dikilitas, Kruthika Iyer, Kazuo Miyazawa, Austin Hilliard, Julie Lynch, Jerome I. Rotter, Yii-Der Ida Chen, Wayne Huey-Herng Sheu, Kyong-Mi Chang, Stavroula Kanoni, Phil Tsao, Kaoru Ito, Matthew Kosel, Shoa L. Clarke, Daniel J. Schaid, Themistocles L. Assimes, Iftikhar J. Kullo
medRxiv 2023.06.02.23290896; doi: https://doi.org/10.1101/2023.06.02.23290896
Twitter logo Facebook logo LinkedIn logo Mendeley logo
Citation Tools

Subject Area