Importance of Genetic Studies of Cardiometabolic Disease in Diverse Populations

Abstract

Genome wide association studies have revolutionized our understanding of the genetic underpinnings of cardiometabolic disease (CMD). Yet, the inadequate representation of individuals of diverse ancestral backgrounds in these studies may undercut their ultimate potential for both public health and precision medicine. The goal of this review is to describe the imperativeness of studying the populations who are most affected by CMD, to the aim of better understanding the genetic underpinnings of the disease. We support this premise by describing the current variation in the global burden of CMD and emphasize the importance of building a globally and ancestrally-representative genetics evidence base for the identification of population-specific variants, fine-mapping, and polygenic risk score estimation. We discuss the important ethical, legal, and social implications of increasing ancestral diversity in genetic studies of CMD and the challenges that arise from the 1) lack of diversity in current reference populations and available analytic samples, and the 2) unequal generation of health-associated genomic data and their prediction accuracies. Despite these challenges, we conclude that additional, unprecedented opportunities lie ahead for public health genomics and the realization of precision medicine, provided that the gap in diversity can be systematically addressed. Achieving this goal will require concerted efforts by social, academic, professional and regulatory stakeholders and communities, and these efforts must be based on principles of equity and social justice.

Introduction

Genome wide association studies (GWAS) have transformed our understanding of the genetic underpinnings of human health and disease. Generally relying on genotyping microarrays that assess from ~100,000 to 2.5 million genetic variants across the genome, these studies often employ imputation to infer genotypes at untyped loci based on whole genome sequenced reference panels with a larger number of variants.¹ Initial genotyping microarrays and the first available reference panels were designed to assess common variation (i.e. genetic variants with a minor allele frequency, MAF>5% in a population) in European populations, and as such our understanding of genetic variation across diverse global populations has been historically limited.² Indeed, as of 2016, most GWAS – including those of Cardiometabolic diseases (CMD) – had been conducted in European descent populations, with only 5% of participants in these studies representing Hispanic/Latino, Pacific Islander, Arab & Middle Eastern, and other Native peoples.³ The newly released GWAS Diversity Monitor tracks participant diversity in the GWAS catalog in real time, and as of this writing, non-Europeans make up between 11-24% of participants in CMD related trait GWAS, with the vast majority of non-European participants being of Asian descent (see Figure 1 for the latest data on diversity in GWAS of CMD).⁴ Although European ancestry GWAS were originally justified as a practical decision to boost power due to their relative homogeneity and comparatively large samples with genotypic data, it is now understood how problematic this lack of diversity in genetic studies is. This is particularly true for CMD, given its unequal burden across global populations. In fact, inadequate representation of individuals of diverse ancestral backgrounds in genomic studies may inadvertently undercut the potential benefits of precision medicine in the near future, and in particular for populations disproportionately impacted by CMD.⁵

Figure 1.

Participant Diversity in GWAS of Cardiometabolic Traits.

Non-Europeans make up just 11% of GWAS of ischemic stroke, 17% of GWAS of chronic kidney disease, 20% of GWAS of Type 2 diabetes, and 24% of GWAS of hypertensive heart disease, and the vast majority of non-European GWAS participants are Asian. Trait definitions note: Ischemic Heart Disease included GWAS traits: Ischemic heart disease. Hypertensive Heart Disease definition included GWAS traits: hypertension, hypertension (young onset), and Medication use (antihypertensives). Ischemic Stroke definition included GWAS traits: Ischemic stroke, Ischemic stroke (cardioembolic), Ischemic stroke (large artery atherosclerosis), Ischemic stroke (non-cardioembolic), Ischemic stroke (small artery occlusion), Ischemic stroke (small-vessel), Ischemic stroke (undetermined subtype). Type 2 Diabetes definition included GWAS traits: Type 2 diabetes, Type 2 diabetes adjusted for BMI, prevalent Type 2 diabetes. Chronic Kidney Disease definition included GWAS traits: Chronic kidney disease, Incident chronic kidney disease, Renal function and chronic kidney disease, Chronic kidney disease and diabetic kidney disease in diabetes, Chronic kidney disease and diabetic kidney disease in type 2 diabetes, Chronic kidney disease in diabetes, Chronic kidney disease in type 2 diabetes, Chronic kidney disease (severe chronic kidney disease vs normal kidney function) in type 1 diabetes, Chronic kidney disease (chronic kidney disease vs normal or mildly reduced eGFR) in type 1 diabetes, Chronic kidney disease (reduced eGFR or end stage renal disease) in type 1 diabetes, Chronic kidney disease (end stage renal disease vs. normal eGFR) in type 1 diabetes. Data from https://gwasdiversitymonitor.com, accessed March 18, 2020.⁴

The overarching goal of this review is to describe the relevance and, arguably, necessity of studying ancestrally diverse populations to gain a better understanding of the genetic underpinnings of CMD. To achieve this, we begin by summarizing key concepts regarding genetic diversity and then explain the importance of including global populations for CMD genomics research (see Key Terms Related to the Genetic Epidemiology of CMD in Box 1). We support this premise by describing the global variation of prevalence of CMD and emphasize its pertinence for the creation of a globally-representative genetics evidence base. Subsequently, we review some of the main benefits of increasing diversity in genomic studies, including the identification of CMD genetic variants that are population specific, the importance of fine-mapping, and estimation of widely-generalizable polygenic risk scores (PRS). Although our review is comprehensive with respect to the need for diversity in genetic studies of CMD and the ethical, legal and social implications for CMD research, we do not address strategies for increasing genetic resources and developing the necessary infrastructure to incorporate diversity into future genomics research, which have been described in detail elsewhere.^6-9 Indeed, the inclusion of diverse populations in genomics research has already yielded scientific insights for chronic kidney disease and low LDL in African descent populations, as well as T2D in Mexican ancestry groups (See Importance of variants specific to a population section below, and Table 1).^10-12 While our review is framed around CMD, we note that our main points are generalizable to many other complex traits and chronic diseases (e.g. Schizophrenia,^13-17 Osteoporosis,^18,19 and Asthma^20-22). We conclude that additional, unprecedented opportunities lie ahead for the realization of precision medicine assuming that the issue of diversity can be systematically addressed by concerted efforts from key stakeholders.

Box 1. Key Terms Related to the Genetic Epidemiology of CMD.

Allele - alternate forms of a gene at a particular location in the genome.

Admixture (or gene flow) - refers to the evolutionary blending of populations. This term makes the assumption that the populations were separated for many generations before individuals began to “admix”, resulting in the current population’s cultural, linguistic or ancestral characteristics.

Copy number variant (CNV) - a one kilobase or larger DNA segment present at different amounts (number of copies) compared to a reference genome.23

De novo familial variant - a genetic variant which arose in a single family through germ cell mutation.

Fine-mapping - a set of statistical procedures to identify potential causal variants in a genomic region associated with a particular trait. Diverse populations are particularly important here, as differences in linkage disequilibrium (LD) can narrow regions with fewer candidate causal variants for interrogation than in European descent populations.24

Founder population - a group descended from a comparatively small number of individuals, which due to the reduced population size does not contain all of the genetic variation of the parental population. The anatomically modern human groups who migrated out of Africa are founder populations, as are many geographic or religious isolates.

Genetic drift - random changes in the frequency of genetic variants (alleles) in a population across time.

Genetic epidemiology - the study of the role of genetic factors (and their interaction with other genetic, environmental, and socio-cultural factors) on health and disease in human populations.

Genome-wide association study (GWAS)- a study which assesses the association of genetic variants across the genome with a trait of interest, generally using a separate linear or logistic regression model for each variant and adjusting for covariates as appropriate. Principal components are used to account for population stratification.

Gene by environment interaction study- a study that quantifies either the extent that a genetic effect varies across non-genetic (environmental or socio-cultural) factors, or the extent that the effect of a non-genetic factor varies by genotype.

Gene expression - transcription and translation of genetic code into phenotype.

Genetic variation - In this review, we define genetic variation in human populations in terms of a gene pool, or a group of organisms of the same species that live in the same area (may be micro- or macro-defined).25

Genotype - the genetic makeup of an individual.

Imputation - a method for inferring genetic variants not included on genotyping arrays, based on LD patterns in known population reference panels.26

Indel - an insertion or deletion of less than one kilobase of nucleotides in a genomic region.

Linkage disequilibrium (LD) - the non-random association of variants in a population, so that variants in LD tend to be inherited together. European descent populations generally have longer blocks of LD across the genome, such that GWAS associations in these populations can tag large regions of the genome including hundreds of potential causal variants. Diverse ancestry populations have different patterns of LD, and populations with recent admixture (African Americans and Hispanic/Latinos, for example) may have much shorter LD blocks, narrowing the tagging region of GWAS associations.

Mutation - the alteration of the DNA sequence of the genome of an organism or extrachromosomal DNA. Mutations may be characterized by the functional consequences, for example missense, nonsense, nonsynonymous, etc.

Natural selection (or selective pressure) - a change in the allele frequencies across generations that is driven by differential fertility or differential survival, as processes in which individuals with certain traits are able to differentially survive and reproduce in a given environment.

Nonsense variant - a mutation that results in the premature termination of a protein coding sequence.

Penetrance - describes the proportion of a population that have a particular genetic variant and also express the phenotype associated with that variant.

Phenotype - the observable traits of an individual, including physical appearance, behavior, biomarkers, and clinical measures that are a product of the genotype interacting with other genetic, environmental, and socio-cultural factors.

Polygenic risk scores (PRS) - a composite measure of multiple genetic risk factors for a particular trait or disease, typically calculated by summing the number of risk variants across the genome, with or without weighting by their effect sizes. PRS have primarily been developed to summarize genome-wide data from European descent populations, and their portability to other populations is complicated by differences in LD across populations.

Population bottleneck - a sharp reduction in population size, which can lead to a reduction in genetic variation. Bottlenecks can be caused by environmental conditions (natural disasters), disease (epidemics), or human behavior (migration or genocide). The Out-of-Africa migration of anatomically modern humans is an example of a bottleneck, where the relatively small groups that migrated out of Africa for Europe and Asia contained only a subset of the genetic diversity present in the entire continent of Africa.27

Reference panel - a well-defined population, typically comprised of individuals whose four grandparents were all from a specific geographic area, used for imputation of genotype data in GWAS.

Single nucleotide polymorphism (SNP) - variation at a single nucleic acid position in the genome that occurs at a frequency of greater than 0.5% in a population. SNPs are the most common and well characterized type of genetic variation.28

Single nucleotide variants (SNV) - variation at a single nucleic acid position in the genome with no frequency restrictions.

Table 1.

Allele Frequencies from gnomAD browser v2.1.1 (as accessed December 19, 2019) and summary of cohorts included in the cited publications for variants discussed in this review article (in order of appearance).

						Allele Frequencies, by Ancestral Population (as defined by gnomAD)*
Annotated Gene	rsID	Publication	Included cohorts	Amino Acid Change	Trait and Additional Notes	African	Ashkenazi Jewish	East Asian	European (Finnish)	European (non-Finnish)	Latino	Other	South Asian
PCSK9	rs28362286	12	Resequencing study in 128 subjects from the Dallas Heart Study (50% African American) with low plasma levels of LDL, subsequent genotyping in n=3553 from Dallas Heart Study	p.Cys679Ter	Plasma Lipids	0.80%	0.00%	0.00%	0.00%	0.00%	0.02%	0.01%	0.00%
PCSK9	rs67608943			p.Tyr142Ter	Plasma Lipids	0.24%	0.00%	0.00%	0.00%	0.00%	0.01%	0.00%	0.00%
CD36	rs3211938	88-93	Pleiotropic variant, with associations discovered in a number of single study efforts and meta-analyses including at least some African ancestry participants	p.Tyr325Ter	Plasma Lipids	8.93%	0.00%	0.00%	0.00%	0.00%	0.27%	0.24%	0.01%
APOC3	rs76353203	94-96	Discovered in a small cohort (n= 809) of Lancaster Old Order Amish individuals, then followed up in meta-analyses of large population based European cohort studies for impacts on lipids and vascular disease	p.Arg19Ter	Plasma Lipids; common (~5%) in the Lancaster Old Order Amish	0.01%	0.00%	0.01%	0.02%	0.05%	0.08%	0.12%	0.18%
GYPC	rs28515082	5	Multi-ancestry GWAS meta-analysis of non-European populations from the Population Architecture using Genomics and Epidemiology (PAGE) study (n= 49,839)		Blood Pressure	7.84%	7.78%	0.00%	3.24%	6.68%	3.30%	5.14%	0.00%
GPR20	rs111409240				Blood Pressure	20.13%	0.69%	0.00%	0.00%	0.07%	2.13%	1.11%	0.00%
JRKL	rs145054295	5	Multi-ancestry GWAS meta-analysis of non-European populations from the Population Architecture using Genomics and Epidemiology (PAGE) study (n= 49,839)		Hypertension	2.47%	0.00%	0.00%	0.00%	0.00%	0.00%	0.09%	0.00%
PRKCH	rs2230500	100-102	Association with stroke reported in several studies and meta-analyses from Japanese and Chinese populations	p.Val374Ile	Stroke	0.22%	1.03%	26.68%	3.55%	1.11%	0.63%	1.87%	1.23%
SLC16A11		11	Reported in a meta-analysis of Mexican studies (n= 3,848 cases and 4,366 controls)	Haplotype association, frequencies not in gnomAD	Type 2 Diabetes
HBB	rs334	104-111, 122, 123, 134-136	Sickle cell anemia SNP, cited papers discuss effects on Stroke, HbA1c, chronic kidney disease in a number of African American cohorts	p.Glu7Val	HbA1c	4.49%	0.00%	0.00%	0.00%	0.01%	0.21%	0.17%	0.06%
TBC1D4	rs61736969	113	Discovered in Greenlandic Inuit population (n=2,575)		Type 2 Diabetes; 17% minor allele frequency in Greenland	2.70%	0.00%	0.00%	0.00%	0.00%	0.08%	0.06%	0.00%
FADS gene cluster	rs7115739	114	Candidate SNPs identified by selective pressure analyses in Greenlandic population, follow-up in three cohorts of Greenlanders or Greenlandic individuals who live in Denmark (n=2733, n=1331, and n=541)		Type 2 Diabetes	73.40%	95.17%	72.81%	96.00%	96.71%	61.56%	91.18%	0.00%
GP2	rs78193826	116	Meta-analysis of four type 2 diabetes GWAS (36,614 cases and 155,150 controls of Japanese ancestry)	p.Val282Met	Type 2 Diabetes	0.17%	0.01%	7.89%	0.44%	0.10%	0.03%	0.69%	3.46%
CPA1	rs77792157			p.Ala341Thr	Type 2 Diabetes	0.00%	0.00%	0.23%	0.00%	0.00%	0.00%	0.01%	0.00%
GLP1R	rs3765467			p.Arg131Gln	Type 2 Diabetes	0.22%	0.19%	23.34%	0.03%	0.27%	4.53%	1.87%	8.63%
PAX4	rs2233580	117	Exome sequencing in n=12,940 individuals from five ancestry groups, East Asian specific association	p. Arg192His	Type 2 Diabetes	0.00%	0.00%	10.94%	0.00%	0.00%	0.01%	0.25%	0.03%
CREBRF	rs373863828	118,119	Discovery of BMI association in 3,072 Samoans, replication in 2,102 additional Samoans	p.Arg457Gln	Body Mass; common in Samoan populations (frequency ~25%)	0.00%	0.01%	0.01%	0.00%	0.00%	0.00%	0.03%	0.00%
G6PD	rs1050828	124,125	Association with HbA1c discovered in trans-ethnic meta-analysis of n=159,940 (n=7,565 African ancestry participants), replicated in whole genome sequencing study including n=10,338 (n=3,123 African Americans)	p.Val98Met	HbA1c	11.64%	0.00%	0.00%	0.00%	0.02%	0.41%	0.30%	0.04%
G6PD	rs76723693	125	HbA1c association discovered in whole genome sequencing study including n=10,338 (n=3,123 African Americans)	p.Leu353Pro	HbA1c	0.54%	0.00%	0.00%	0.00%	0.00%	0.07%	0.04%	0.00%
HBA1/HBA2	esv2676630	127,128	Association reported in two African American cohort studies	structural variant, not in gnomAD	HbA1c
APOL1	rs73885319	10, 130-133	Cited papers discuss APOL1 risk haplotype effects on various forms of chronic kidney disease in African ancestry and Hispanic/Latino populations		Chronic Kidney Disease	22.76%	0.00%	0.00%	0.00%	0.01%	0.68%	0.58%	0.01%
APOL1	rs71785313			Indel (G2 allele), not in gnomAD- approximate frequencies listed from 1000 Genomes		12.90%		0.00%		0.00%	0.80%		0.00%
CYP2C19	rs4986893	180	Cited article discusses differential allele frequencies by population and legal/ethical implications for these well known pharmacogenetic variants	p.Trp212Ter	Response to clopidogrel bisulfate	0.04%	0.00%	6.26%	0.01%	0.03%	0.02%	0.24%	0.40%
CYP2C19	rs4244285			p.Pro227Pro	Response to clopidogrel bisulfate	17.76%	13.20%	30.75%	17.50%	14.68%	10.12%	15.95%	32.40%

The gnomAD browser is available at: https://gnomad.broadinstitute.org/.

^*The above populations are listed in alphabetically order and reflect several ancestral super-populations as defined by gnomAD: African (AFR), Ashkenazi Jewish (ASJ), East Asian (EAS), Finnish (FIN), Non-Finnish European (NFE), American Admixed /Latino (AMR), and South Asian (SAS). Other (OTH) is comprised of individuals who did not cluster with any super-population group.

How do we Describe Human Genetic Variation?

Human variation or variability refers to the range of all possible values for any phenotype, and can be attributed to genetics, environmental factors, and their interactions. It is now understood that the role of genes, although crucial, is dynamic and modifiable. This perspective contrasts with the misconception that all genetic inheritance is static and entirely deterministic. Genetic variation can take many forms and broadly refers to differences in structure (e.g., chromosomal rearrangements or abnormalities) and composition (e.g., DNA sequence) of the genome between individuals and populations. Genetic variation occurs in both germ cells and somatic cells, but only variation that arises in germ cells can be inherited. Types of common human genetic variation include single nucleotide variants (SNV) or polymorphisms (SNPs), insertions and deletions (indels), substitutions, inversions, and copy number variants (CNVs).²⁹ Genetic variants can be found at different frequencies in a population, ranging from private (the only copy), de novo familial (a few copies), rare (MAF< 1.0%), low-frequency (MAF=1% - 5%), and common (MAF>5%). Most variants are hypothesized to be neutral in terms of functional implications.²⁹ While no empirical information exists about the functional impact of the vast majority of the estimated 10 million SNPs in the human genome, prediction algorithms such as PolyPhen-2,³⁰ SIFT,³¹ FATHMM-XF,³² MutationTaster,³³ and Combined annotation Dependent Depletion (CADD)³⁴ have been developed for this purpose and are freely available. It is important to acknowledge that more information exists for common variants, due to the design of genotyping microarrays and the greater statistical power to detect common genetic variant associations. With the decreasing cost of sequencing, the field of genetic epidemiology is increasingly able to identify low frequency, rare and de novo familial variants for CMD, as well as aspects of genomic structural variation.

Here, we primarily focus on the role of ancestral genetic differences in CMD. In this review, we consider the genetic origins of differences in population CMD burden and related health parameters, which may in part track with ancestry or with the socially and culturally defined constructs like race or ethnicity (Box 2 for Key Terms Related to Ancestral Diversity). To describe the importance of ancestral diversity for the quantification of the influence of genetic factors on CMD, we briefly define additional key terms related to ancestral diversity (Box 2), and acknowledge the lack of gold standard scientific definitions.^35,36 As noted above, we use ancestry to refer to the continental origin of the ancestors of an individual or population, and to a lesser extent to the population dynamics within each continent that shaped the observed patterns of genetic variation. We note that genetic ancestry is often estimated via comparison of participants’ genotypes to continental reference populations, so incomplete representativeness, availability and small sample sizes of these reference populations are important limitations for the field of genetic epidemiology.³⁷ Moreover, discrete labeling of ancestral populations by continent or other means vastly oversimplifies genetic variation.

Box 2. Key Terms Related to Describing Ancestral Diversity.

The terms ancestry, diversity, ethnicity, and race have no gold standard scientific definitions.35,36 Indeed, often terms are erroneously used interchangeably, for example race and ethnicity.40 Below, for clarity, we define these terms as used in this review.

Ancestry- We use ancestry (or ancestral background) to refer to the continental origin of the ancestors of an individual or population, and to a lesser extent to the population dynamics within each continent that shaped the observed genetic patterns. Genetic ancestry is often estimated via comparison of participants’ genotypes to continental and/or global reference populations, so incomplete availability and the small sample sizes of these reference populations is an important concern.37 Note that discrete labeling of ancestral populations oversimplifies genetic variation. However, given differences in allele frequencies and linkage disequilibrium (LD) across populations, estimating and accounting for ancestry (either discretely or continuously) is necessary for appropriately powered and calibrated genetic analyses.41-43

Diversity- references many aspects and may include genetics, ancestry, gender/sexual orientation, age, culture, abilities/disabilities, geography, socioeconomics, etc. However, in this review we use the term diversity to refer to the genetic and ancestral diversity that drive differences in health, specifically CMD burden and related health parameters. We acknowledge that genetic or ancestral diversity are often conflated with the socially and culturally defined constructs of race and ethnicity.

Ethnicity- A socially and culturally heterogeneous term that represents how individuals and/or groups of individuals identify based on shared history, cultural traditions, and ancestry, which incorporates both societal norms and an individual’s own self-perception. While ethnicity has been used as a proxy for health and disease risk at the population level, the term is heterogeneously defined worldwide; we acknowledge that within each group there is notable ancestral diversity and that individuals within the US or across the globe may prefer to use different constructs.

Race- A socially and culturally defined term used to refer to an individual and/or group of individuals. Like ethnicity, race is often used to refer to the shared history, language and ancestry of individuals and/or populations, but it does not singularly predict genetic susceptibility to disease, genotype, or drug response of an individual patient.44,45 Historically, the construct of race has been linked to visible physical attributes, and used as a justification for discrimination to establish and reinforce social inequities. As a result, environmental and sociocultural factors have been shown to differentially track with racial groupings; for example, disparities in access to goods and services has had a large impact on health and disease.46,47

When describing the burden of disease in this review, we present both country-specific burden estimates in Figure 2 and refer to the Global Burden of Disease (GBD) regions as constructed by the Institute for Health Metrics and Evaluation (IMHE). For simplicity in the text, we highlight the burden of CMD using GBD regions that we expect to have some amount of shared ancestral heritage (e.g. Western Europe, East Asia, Sub-Saharan Africa). Then we point out GBD regions that are comprised of countries, which may have more ancestral diversity given their recent demographic histories (e.g. the United States, US, and Canada, Australia or New Zealand). Nonetheless we recognize that grouping human populations in a geographic manner, e.g. by country or region, may inadvertently oversimplify human genetic diversity. Thus, in an effort to further unpack ancestral diversity within a country like the US, which is the primary focus of this review, we also refer to common categorizations for US race/ethnic minorities as proxy groupings of individuals who may have high proportions of non-European ancestry. However, we recognize that in the US many ancestrally diverse populations, such as racial/ethnic minorities or immigrant populations, may prefer other conceptualizations of race/ethnicity than those currently used in the US.³⁸ For example, in the US the term ‘Hispanic/Latino’ is defined by the Office of Management and Budget as a union of Spanish language use and heritage from Latin America and the Caribbean (only countries with Spanish cultural origins). When self-identified US Hispanic/Latinos were asked to mark their race on the 2010 US Census using five US racial categories, 48.9% of Hispanic/Latinos identified as being of “Some other Race” (30.5%) using written descriptors such as Mexican, Puerto Rican, Latin American, 5.4%, identified as being of “Two or More Races” (including the five US racial categories and “Some other Race”), and another 13.0% choose to not respond to the race question, making the non-response rate for self-identified non-Hispanic/Latinos three times higher than for the total US population.³⁹

Figure 2 A-B.

Global burden of ischemic heart disease

Colors represent binned values of age-standardized disability adjusted life-years (DALYS) from ischemic heart disease per 100,000 population for females (Panel A) and males (Panel B). Data used for the figure was downloaded from the Institute for Health Metrics and Evaluation's Global Burden of Disease (GBD) Compare Data Visualization tool and is available at http://vizhub.healthdata.org/gbd-compare.⁴⁹

Importance of global populations for CMD research.

Ancestral background may influence one’s individual CMD risk as well as the population-level differences in CMD burden seen both across and within countries. Disability adjusted life years (DALYs) are a common epidemiological measure of overall disease burden, as they account for years of life lost due to premature mortality and years of life lost due to disability from a specific condition.⁴⁸ For example, when comparing age-adjusted estimates of DALYs due to ischemic heart disease in 2017, the burden is greatest in a number of countries within the GBD regions of Oceania, Central Asia and Eastern Europe; in general males have a higher age-adjusted burden of CMD than females (Figure 2A-B).⁴⁹

Between 1990-2017 a number of GBD regions showed an intractably high burden of ischemic heart disease as measured by DALYs (e.g. Oceania, to a lesser extent South Asia), whereas others have experienced either steady declines (e.g. Central Europe, North Africa and the Middle East) or experienced intermittent declines during the same time period (e.g. Central Asia and Eastern Europe, Online Figure I).⁴⁹ Similar disparities in CMD burden are seen across time globally with respect to hypertensive heart disease, ischemic stroke, type 2 diabetes (T2D), and chronic kidney disease (CKD) (Online Table I).⁵⁰ For example, hypertensive heart disease has the highest burden in Central Sub-Saharan Africa, followed by Oceania, and the other regions in Africa and the Middle East (Online Figure II).⁴⁹ In contrast, ischemic stroke is most common in Eastern Europe, Oceania, Central and East Asia (Online Figure III).⁴⁹ T2D and CKD are both most common in Oceania, Central Latin America and Mexico, Central and Southern Sub-Saharan Africa (Online Figures IV-V).⁴⁹

Within the US, African Americans have the highest prevalence of hypertension and related conditions such as coronary artery disease (CAD), ischemic stroke, heart failure, and CKD.⁵¹ In fact, hypertension may account for roughly 50% of the disparity in life expectancy between African Americans and European Americans.⁵² Broadly speaking, the prevalence of adult obesity, T2D, and related complications are highest among individuals of Native American, African American, and Hispanic/Latino ancestry, and lowest in those of European and East Asian descent.^53,54 Even within commonly used US race/ethnic groupings, notable differences in disease susceptibility and incidence exist. For example, Puerto Rican and Mexican American adults have a greater burden of cardiovascular disease (CVD) risk factors like obesity, than South Americans,⁵⁵ and Indian and Filipino Americans have a higher burden of obesity than Chinese Americans.⁵⁶ Asian Indian and Filipino Americans have the highest prevalence of diagnosed T2D among Asians (13% and 10%, respectively), and Mexican Americans and Puerto Ricans have the highest prevalence of diagnosed T2D than any other Hispanic/Latino backgrounds (14% and 12%, respectively).⁵⁴ While lifestyle, cultural norms, healthcare access, psychosocial and socioeconomic stressors are undeniably important contributors to the disproportionate disease burden across ancestrally diverse populations, some of these health disparities persist even after accounting for differences in social and environmental exposures for diseases.^57-59 This observation further suggests that some of the susceptibility to CMD-related traits or diseases may be influenced by genetic factors, which may be ancestry-specific or have complex interactions with environmental factors that are patterned across racial/ethnic groups.⁶⁰

In light of the differing burden of CMD both globally and within countries with diverse populations, efforts to broaden the diversity of populations studied in genetic research have become an imperative for advancing clinical research and public health. The more inclusive genomic studies are, the more effective they will be at expanding the scope of known human genomic variation and bolstering our understanding of disease etiology to be able to improve population health both globally and locally.

Importance of diverse studies for the evaluation of differential allele frequencies.

Human variation is the result of non-genetic and genetic forces, e.g. natural selection and genetic drift. For example, variation seen in current human populations has been heavily influenced by the Out-of-Africa migration of anatomically modern humans. The movement of relatively small groups out of the African continent over time is an example of a population bottleneck, where the groups that moved into Europe and Asia contained only a subset of the genetic diversity present in the entire continent of Africa.²⁷ Efforts to characterize global human genetic variation (e.g., the 1000 Genomes Project,⁶¹ H3 Africa⁶²), have demonstrated differences in allele frequencies between populations with differing continental ancestries.^61,63,64 These differences vary based on the evolutionary age of the derived variant and the demographic history of the population. Historically, population allele frequency differences were attributed to genetic processes such as natural selection; however, we now have evidence that widespread allele frequency differences were created by the out-of-Africa bottleneck.⁶⁵ Indeed, the majority of all genomic variants are rare and display allele frequency differences, including those variants that are private to a single continental population (or are population specific).^61,66 These differences have the potential to inform future treatments and recommendations, for example the discovery of PCSK9 loss-of-function variants in African Americans that led to the development of new therapies to treat high LDL, as well as others (see Importance of variants specific to a population below). Medical genomics is increasingly conducting whole exome and genome sequencing of patients to identify disease susceptibility variants. However, identifying disease-relevant sequence variants has been difficult, partly due to lack of consensus on variant annotation. One factor that influences variant annotation is allele frequency estimates. Because of this, a priority for investigators has been the development of standard approaches for sharing genomic and phenotypic data provided by clinicians, researchers, and patients through centralized databases, such as ClinVar⁶⁷ and the University of Chicago’s Geography of Genome Variation browser.⁶⁸ For example, the Clinical Genome Resource (ClinGen) Variant Curation Interface is a publicly-available curated resource for clinicians and researchers that pulls frequency data from numerous sequencing efforts, including gnomAD⁶⁹ (https://gnomad.broadinstitute.org/), PAGE⁵, 1000 Genomes Project (https://www.internationalgenome.org), and the Exome Sequencing Project⁷⁰ (ESP; https://evs.gs.washington.edu/EVS/). These frequencies are also available on the National Center for Biotechnology Information (NCBI) dbSNP database⁷¹ (ncbi.nlm.nih.gov/snp/), with the addition of the Vietnamese Genetic Variation Database⁷², Northern Sweden, the Avon Longitudinal Study of Parents and Children⁷³ (https://www.ncbi.nlm.nih.gov/bioproject/PRJEB7217), the UK10K Study⁷⁴ (https://www.ncbi.nlm.nih.gov/bioproject/PRJEB7218), and Trans-Omics for Precision Medicine (TOPMed)⁷⁵ (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA400167). There are also many efforts conducted by industry to increase diverse representation, such as Regeneron’s DRIFT Consortium⁷⁶ (https://www.regeneron.com/sites/all/themes/regeneron_corporate/files/science/DRIFT-Consortium-Factsheet-Backgrounder-July-FINAL.pdf) and 23andMe’s Populations Collaborations Program for genotype data⁷⁷ (https://research.23andme.com/populations-collaborations/). However, these last two data sources are not currently publicly available and therefore not useful to the research and clinical communities for the adjudication of risk variants based on population frequencies.

The difficulty in determining the pathogenicity of a rare variant is compounded by clinical laboratories’ labeling putatively deleterious non-synonymous calls as variants of unknown significance, a phenomenon that occurs at higher rates for individuals of non-European descent, especially since these variants have been studied and characterized less frequently.⁷⁸ Alternatively, diverse genetic data have improved clinical knowledge by leading to a reclassification of putatively causal pathogenic variants for hypertrophic cardiomyopathy, which were later determined to be benign as a result of being over-represented in African Americans.⁷⁹ For-profit companies are also venturing into the business of variant reclassification. For example, Blueprint Genetics offers a variant classification service⁸⁰ that allows the sequencing data from a previous exome to be re-analyzed, in its entirety, to search for new clinically relevant variants that may explain or contribute to a patient’s diagnosis.

Differences in allele frequencies across global populations have also driven a SNP ascertainment bias in genotype array data.^81,82 Genotype arrays, especially older ones (e.g. Affymetrix 5.0, Illumina Goldengate), were developed based on European ancestry sequence data,⁸³ a feature that has contributed to the observed biased range of allele frequencies in GWAS of non-European populations. This is becoming a major stumbling block, as results from GWAS are combined to generate PRS (Polygenic Risk Scores, sometimes also called genetic risk scores), which are now being used to generated personalized CMD risk estimates in both clinical prognosis and personalized intervention/treatment plans.⁸⁴ Developing a successful PRS depends on maximizing the proportion of the total variance for a particular phenotype that is explained by a set of identified genetic variants. In research, it has become a widely-accepted practice to incorporate all measured variants (many of which are correlated) when calculating the proportion of variance explained, as this tends to improve prediction accuracy in complex traits.^84,85 Such PRS are being widely applied, but SNP ascertainment bias can lead to a model with vastly different risk estimates across ancestries and poor prediction accuracy. Moreover, our recent work has shown that a PRS ascertained using standard methods in one population can yield unpredictable biases in the distributions of scores in other groups, with patterns fluctuating dramatically across traits.⁸⁶ This also suggests that many causal variants, particularly in non-European ancestries, remain undiscovered. The only equitable use of PRS is one that ensures that scores can be calculated accurately for everyone, meaning the genomic data used must be fully-representative of all human genetic variation. Any genetically-informed personalized medicine approach that fails to take this into account is at risk of misinterpreting the underlying data.

Importance of variants specific to a population.

Due to the historical over-representation of European ancestry individuals in current GWAS, to date we have just begun to identify associations that are uncommon in populations of European ancestry but are common in others, due to either non-genetic or genetic factors that shift allele frequencies and differences in linkage disequilibrium (LD) patterns between populations. Below, we highlight a non-exhaustive list of examples of genetic variant associations for ischemic and hypertensive heart disease, stroke, T2D and CKD and summarize them in Table 1. We then describe the generalizability of mostly European ancestry discovered variants across ancestrally diverse populations. Although there is a paucity of genomic studies in ancestrally diverse populations, several notable large genomic studies and consortia have been assembled to focus on advancing the state of genetic research in these groups. The examples below are not meant to serve as a comprehensive accounting of all ancestrally diverse genomic studies of CMD and related traits, but rather showcase the breadth of discovery that is possible in such diverse studies.

Ischemic and Hypertensive Heart Disease and Stroke.

As described above, the burden of DALYs due to ischemic or hypertensive heart disease is greatest in Oceania, and concerningly high in Central Asia and Eastern European (ischemic heart disease) and several regions in Africa (hypertensive heart disease) (Online Figures I-II).⁴⁹ In contrast, ischemic stroke is most common in Eastern Europe, followed by Oceania and Central Asia (Online Figure III).⁴⁹ Between 1990-2017, most global regions have experienced a general decline in both ischemic and hypertensive heart disease, but the specific trajectories vary greatly. Differences in genetic ancestry and variation with respect to plasma lipid levels, hypertension or other CMD-related traits may contribute to these population-level differences. Several examples are highlighted below.

Plasma lipid levels and PCSK9, CD36, and APOC3.

Ancestry-specific variants associated with blood lipid levels were first described with the seminal 2005 sequencing of African ancestry participants of the Dallas Heart Study,¹² describing several loss-of-function PCSK9 variants (e.g., rs28362286, MAF~1%, and rs67608943, MAF~0.3%) that were associated with approximately 40% lower low-density lipoprotein (LDL) cholesterol levels.¹² Both variants are found at lower frequencies in the African ancestry samples of gnomeAD (Table 1). At the time, PCSK9 had been identified as an autosomal dominant hypercholesterolemia gene (gain-of-function mutations), but the observations for PCSK9 loss-of-function variants, with resulting large decreases in CAD risk, helped support the successful development of PCSK9 inhibitors.⁸⁷

Other plasma lipid examples include a loss-of-function variant in CD36 (rs3211938, Table 1) that is common only in African ancestry populations (MAF=9%) and under selective pressure due to malaria.⁸⁸ The variant is associated with higher high-density lipoprotein (HDL)^89-91 and lower triglycerides, as well as platelet traits,⁹² red cell distribution width,⁹³ C-reactive protein,⁹⁰ and other measures relevant to CMD.

Additionally, carriers of a triglyceride lowering null mutation (rs10892151, MAF ~5%) in APOC3 are common in the Lancaster Old Order Amish, which has allowed for the identification of APOC3 loss-of-function as cardioprotective.⁹⁴ This observation has since been confirmed in large meta-analyses for rs76353203 (Table 1).^95,96

GWAS of Blood Pressure Traits.

Recent GWAS of a pooled, ancestrally-diverse sample conducted as part of the Population Architecture using Genomics and Epidemiology (PAGE) study have analyzed the genetic etiology of systolic blood pressure (SBP), diastolic blood pressure, and 24 other complex traits.⁵ They discovered a novel variant at GYPC for SBP (rs28515082, Table 1) that was most common in their sample in Native and Hispanic/Latino Americans (MAF=10-13%), and is particularly rare in East and South Asian populations (MAF<0.5%). Although this variant is common in European descent groups (MAF=16%), it was first identified in association with blood pressure in the context of a diverse genetic sample. They also described a secondary signal for SBP at GPR20 (rs111409240, Table 1), which is independent of the previously identified European signal⁹⁷ (rs34591516). The variant leading this novel secondary signal is common in African Americans (MAF=20%), low frequency in the other diverse populations analyzed (MAF<6%), and rare in European descent individuals (MAF<1%). Other large trans-ethnic electronic health record analyses and meta-analyses of ancestrally diverse samples have also highlighted the added value of diversity in studies of blood pressure variation.^98,99

JRKL and Hypertension.

Also in the PAGE study, an indel (rs145054295) in JRKL was associated for the first time with hypertension (ß=−0.43, P=3.70×10⁻⁹) and effect sizes at this site were shown to be different across ancestries (P=0.025).⁵ In the PAGE study sample the variant was monomorphic in European populations;⁶¹ yet, the minor allele was found at 2.4% frequency in African Americans (ß=−0.36, P=1.96×10⁻⁵), 0.4% in Hispanic/Latinos (ß=−0.52, P=5.08×10⁻³), and 0.5% in Native Americans (ß=−1.82, P=0.058). The variant is at even lower frequencies among primarily East Asians (MAF=0.01%, ß=−2.39, P=0.32) and Native Hawaiians (MAF=0.01%, ß=14.90, P=0.08) in PAGE. These differential effect sizes by ancestry are likely largely due to allele frequency differences (also reflected in Table 1), further highlighting the importance of studying diverse populations for discovery in trait mapping and at scale, as when CMD-relevant variants are this rare, large sample sizes are required to robustly estimate effect sizes.

PRKCH and Stroke.

A missense variant (rs2230500) in PRKCH has been associated with increased risk of ischemic stroke in Japanese^100-102 and Chinese populations.¹⁰² In a meta-analysis of five study populations comprised of Chinese and Japanese individuals (3,686 cases and 4,589 controls), individuals with the GA or AA genotype had approximately 34% greater odds of developing ischemic stroke that those with the GG genotype.¹⁰² The SNP is common in Asian populations (Japanese, MAF=24%; Han Chinese, MAF=18%),¹⁰⁰ but rare in European and African descent populations (MAF≤1%, Table 1). PRKCH is a member of the protein kinase C (PKC) family and is involved in the development and progression of atherosclerosis in humans.¹⁰⁰ The A allele results in the amino acid substitution Val 374 Ile within the ATP-binding site of PKC-delta, thereby enhancing PKC activity.¹⁰³

Sickle cell trait and Stroke.

Sickle cell trait (i.e. individuals who have only one copy of the sickle cell variant at rs334, which is more common in individuals of African ancestry; Table 1), has been associated in some studies with risk of stroke¹⁰⁴, though this link has been disputed by a recent meta-analysis.¹⁰⁵ More consistent associations have been observed with thrombosis and hemostasis biomarker D-dimer.^106-109 Venous thromboembolism (particularly pulmonary embolism) has also been associated with sickle cell trait;^110,111 more analysis is needed to understand these associations in larger sample sizes, as well as elucidate the underlying mechanisms.

Type 2 Diabetes.

Unlike the trends observed for heart disease burden globally, DALYs for T2D have generally remained more intractably high between 1990-2017. For example, the burden of T2D is highest for Oceania, where it increased and then plateaued from 1990-2017. High T2D burden is also found in Southern Sub-Saharan Africa, and Central Latin America and Mexico (Online Figure IV).⁴⁹ Genetic variation related to glycemic regulation, obesity, or other CMD-related traits may explain some of these global disparities in risk.

SLC16A11 and T2D.

Studies of non-European ancestry populations have also revealed T2D-related genetic variants with population frequency differences. For example, several admixed Mexican ancestry populations carry a T2D risk haplotype with variants in SLC16A11 with four amino acid substitutions; subsequent lookups of this haplotype revealed a ~50% frequency in Native American and ~10% in East Asian study participants, but rare in participants of European or African descent.¹¹ This haplotype explains roughly 20% of the increased prevalence of T2D in Mexico, and expression of SLC16A11 in heterologous cells (non-human cells that do not usually express this gene) alters lipid metabolism, causing an increase in intracellular triacylglycerol levels.¹¹² Despite T2D having been well-studied by GWAS previously, this analysis in Mexican ancestry individuals identified SLC16A11 as a novel finding possibly involved in triacylglycerol metabolism.

T2D and Obesity in Founder Populations.

A nonsense variant in TBC1D4 was associated with a large increase in T2D risk (Odds Ratio=10.3 in homozygous carriers) as well as a large decrease in glucose uptake in muscle in a founder population of Greenlandic Inuit among 2,575 participants (rs61736969, MAF=17%, Table 1).¹¹³

Another recent GWAS of a Greenlandic Inuit population has also reported a large effect size variant for height and weight in the FADS gene cluster (rs7115739), which is highly prevalent in the Inuit,¹¹⁴ likely as an adaptive mechanism to a diet rich in polyunsaturated fatty acids. Subsequent analyses revealed that this variant also influences height in European descent populations. The effect sizes for the weight finding differed between Greenlandic and a previous European ancestry GWAS,¹¹⁵ likely due to its lower frequency in Europeans (MAF<5%, Table 1).

Pancreatic Acinar Function and T2D.

In a meta-analysis of four GWAS of T2D in individuals with Japanese ancestry, three variants were found to be in moderate LD (r²>0.6) with previously unreported missense variants [p.Val282Met in GP2 (rs78193826), p.Ala341Thr in CPA1 (rs77792157) and p.Arg131Gln in GLP1R (rs3765467) – all of which are more common in East Asian versus European populations (Table 1). These variants are in genes that have been previously related to pancreatic acinar cell function (e.g. CPA1 and GP2) and insulin secretion (e.g. GLP1R).¹¹⁶ In previous work, another coding variant in PAX4 (Arg192His, rs2233580), an important transcription factor for islet function, was also more common in East Asians populations (MAF=11%, Table 1) than any other ancestral group and was found to be associated with T2D.¹¹⁷

CREBRF, T2D and Obesity.

A large-effect BMI increasing missense variant in CREBRF (rs373863828) was also recently identified, as it is common in Samoan populations (MAF ~25%) and rare in other global populations outside of Oceania (Table 1).^118,119 The variant has a larger effect size (ß=1.36 kg/m²) than many common BMI loci, including FTO [rs1558902, the largest effect size variant reported in European GWAS¹²⁰ (ß=0.39 kg/m²)], and is associated with decreased risk of T2D. In contrast to the majority of previous GWAS findings for BMI and T2D, this variant appears to increase BMI while decreasing T2D risk. Functional analyses suggest that the variant can decrease energy use and increase fat storage in adipocyte cell models. Due to its large effect size and common allele frequency in Samoans, this variant was detectable in a discovery and replication sample of ~5,000 individuals, which is much smaller than samples generally required for GWAS.

HbA1c and T2D.

There is also a growing awareness of the relationship between variants of differing allele frequencies across global populations and the accuracy of HbA1c as a measure of long term glycemic control.¹²¹ Recent analyses using assays robust to previously known assay interference effects have found that sickle cell trait (Table 1) can lower HbA1c relative to fasting glucose levels.¹²² These effects may be somewhat assay dependent, however, as in the Diabetes Prevention Program sickle cell trait was shown to be associated with higher HbA1c.¹²³ As shown in Table 1, common (rs1050828, MAF=12%) and rare (rs76723693, MAF=0.5%) missense variants at the G6PD locus in African ancestry populations may also influence the accuracy of HbA1c as a test for glycemic control.^124,125 Similar to sickle cell trait, the geographic distribution of G6PD deficiency parallels the distribution of malaria endemic regions,¹²⁶ highlighting the need for further examination of HbA1c accuracy in other populations exposed to endemic malaria, e.g. Southeast Asia. Recent analyses suggest that alpha thalassemia (based on CNV esv2676630 carrier status) may also be associated with higher HbA1c¹²⁷ and may statistically interact with sickle cell trait to influence clinical parameters.¹²⁸ The fact that the recent identification of these HbA1c findings for common genetic variants came decades after the clinical use of HbA1c as a measure of long term glycemic control¹²¹ and an initial round of GWAS analyses for HbA1c¹²⁹ and other traits illustrates the clinical significance of what has been missed by focusing GWAS on exclusively European ancestry populations.

Chronic Kidney Disease.

DALYs due to CKD dynamically changed from 1990-2017 (Online Figure V).⁴⁹ Similar to the global trends in burden for T2D, the burden of CKD is led by Oceania, followed by Central Latin America and Mexico, and Central Sub-Saharan Africa. Below, we highlight examples of how global variation in CKD burden may reflect differences in genetic ancestry.

APOL1 and CKD.

African Americans are more than twice as likely to develop end-stage renal disease as European Americans; this observation eventually led to the discovery of the G1 (rs73885319) and G2 (rs71785313) risk variants in APOL1 that are more common among individuals of African ancestry (Table 1),¹⁰ likely due to selective pressure from African trypanosomiasis.¹³⁰ These variants have a large impact on risk in carriers of two risk alleles, with a reported Odds Ratios of 17 for focal segmental glomerulosclerosis, 29 for Human Immunodeficiency Virus-associated nephropathy¹³¹, and at least 15% lifetime risk of CKD in risk allele carriers.¹³² These variants are also associated with faster progression of disease in other African-admixed groups, such as Hispanic/Latinos.¹³³ The association was first found in African Americans because the relevant variants had reached higher allele frequency in that population (rs73885319 MAF = 23%, rs71785313 MAF = 13%), yielding higher power for a given sample size. Given that Hispanic/Latinos are an ancestrally-diverse ethnic group, it is not surprising that the association replicated in some Hispanic/Latino backgrounds (e.g., specifically those with a higher proportion of African ancestry, such as individuals from the Caribbean), but not others.

Sickle cell trait and CKD.

Initial reports of differential susceptibility to CKD for sickle cell trait from small studies¹³⁴ have since been confirmed by larger cohort studies.^135,111 In fact, sickle cell trait may have a similar effect size for progression to end stage renal disease as the well-known APOL1 high risk genotypes [Hazard Ratio of 1.8 for APOL1 versus 2.0 for sickle cell trait in the REasons for Geographic and Racial Differences in Stroke study].¹³⁶

Summary.

The above examples highlight the need for more genetic discovery studies in ancestrally diverse and admixed populations in order to identify novel susceptibility variants that may be rare or absent in GWAS of European descent populations. In addition, there is rising concern that findings in one ancestral group may not have the same effect sizes in other ancestries. The PAGE Study has explored this question for several CMD traits; in a seminal paper they presented an analysis of BMI, T2D, and lipid levels, and compared the direction and magnitude of effects for GWAS-identified variants in multiple non-European ancestry populations against European ancestry findings. They found an overall dilution of effect sizes across ancestries.⁸⁶ The PAGE study further tested for evidence of diminished effect sizes in a diverse sample at previous GWAS findings for 26 traits related to CMD and other complex diseases.⁵ This experiment demonstrated that previously-reported GWAS findings, derived from predominantly European ancestry samples, have significantly attenuated effect sizes on average in other ancestral groups. For example, the correspondence between previously-reported GWAS effect sizes and effects seen among Hispanic/Latinos was 0.86 (95% confidence interval= 0.83-0.90) and 0.54 (0.50-0.58) among African Americans.⁵ The observed weaker effect sizes in non-European populations may be a function of differential LD, allelic heterogeneity, gene-gene or gene-environment interactions, and can further widen the gap between the impact of known GWAS findings on CMD. Regardless of the origin of the differential effects, caution should be exercised in applying any genetic risk prediction model based on SNP association findings beyond the ancestry group in which they were identified.

Importance of diversity for fine-mapping

After the initial success of GWAS in identifying genomic regions that are robustly associated with a wide array of diseases and related traits, the next major challenges for genetic epidemiology are identifying the underlying causal variants and target genes and translating these findings into clinical insights. Of the thousands of genomic regions associated with complex traits, over 90% are in non-coding, potentially regulatory regions of the genome.¹³⁷ While GWAS are an effective tool for identifying genomic regions associated with a particular phenotype, they are often unable to pinpoint the causal variant or even the implicated gene. Indeed, studies of one single ancestral population, and their LD signature, provide investigators with limited ability or power to identify causal variants.¹³⁸

In many cases, the causal variants underlying GWAS signals are shared across ancestry groups, and GWAS of multiple diverse populations (i.e. transethnic meta-analysis or pooled analysis of multiple ancestral groups) can help narrow the credible set of causal variants at a given locus due to their varied LD structure. For example, African ancestry populations have on average shorter LD blocks than those found in European populations, and this characteristic has been shown to be particularly helpful for narrowing the number of candidate causal variants at a given locus and prioritizing candidate variants for functional follow-up.^24,139 In many recent analyses of CMD, transethnic fine-mapping has helped narrow lists of candidate variants for kidney function,¹⁴⁰ QT interval,¹⁴¹ lipid traits,¹⁴² and BMI.¹⁴³ Integration of transethnic fine-mapping analyses along with functional annotation can aid in variant selection for follow-up testing of differences in transcriptional activity and protein binding in vitro,¹⁴⁴ and can lead to the identification of putative causal variants at a previously described GWAS locus.

Importance of diversity in risk prediction and precision medicine.

As described above, PRS are being routinely used to aggregate effect sizes across the genome to estimate the overall contribution of genetics to the variability observed in a given phenotype. In practice, a PRS is computed for each genotyped individual in a target (testing) sample based on extant GWAS discovery results (training sample). PRSs have been used for both estimating the impact of a set of novel GWAS results in external cohorts and for providing individualized risk prediction.^84,145 To draw inferences, the distribution of risk scores is often divided into percentiles or other categorizations to compare the risk of the outcome for an individual, to others in the same sample. However, PRS often suffer from bias as the majority of the training GWAS data are derived from European ancestry populations,¹⁴⁶ leading to unpredictable differences in PRS estimation and model fit across populations due to differences in allele frequencies, effect sizes, or the underlying etiology of the trait.^145,147 There are many examples where trying to incorporate European-derived PRSs in diverse populations results in poor model fit.^{145,146,148,149} Below we describe examples of such biases in the published literature and from our own work.

CAD PRS in European Ancestry Populations.

Research has demonstrated the potential benefit of including genetic risk scores, in the form of PRS, along with more traditional assessment of risk factors, such as the Framingham Risk Score (FRS), in predicting poor cardiovascular outcomes. One of the earliest PRS, from the Myocardial Infarction-Genes randomized placebo-control clinical trial, demonstrated that for patients receiving both PRS (based on 11 SNPs associated with CAD) and Framingham 10-year CVD risk score (FRS) information had lower LDL cholesterol levels and increased statin use compared to those individuals for whom only FRSs were used.^150,151 In a more recent clinical trial, the GeneRisk study in Finland showed that providing personalized cardiovascular disease risk information, based on a combination of traditional risk data and PRS, motivated healthy behaviors.¹⁵² Physicians at Massachusetts General Hospital are launching a Preventive Genomics Clinic to help patients understand their monogenic and polygenic risk for a variety of health conditions and take steps to minimize their risk.¹⁵³ The hope is that this clinic will serve as a model for how individuals may be able to get an inexpensive report of monogenic and polygenic risk and incorporate this information into preventive measures. However, if the PRSs are developed in a single population, they will necessarily miss important genetic variants in other populations that contribute to risk.

Obesity PRS.

To understand the differences in PRS performance as a function of population architecture and epidemiology, the PAGE study assessed the performance of a recently published PRS for obesity¹⁵⁴ in the different populations of the PAGE study. Model fit diminished with genetic distance from European populations when correlating the PRS with BMI. The four PAGE populations (Hispanic/Latino, N=19,028; African American, N=16,093; Asian, N=4,155 (88% Japanese, 5% Filipino, 4% Chinese, <1% South Asian); and Native Hawaiian, N=2,502) had an adjusted R² ranging from 2.7% (Hispanic/Latino) to 0.3% (Native Hawaiian) (Figure 3A). The Asian group also had low performance with an adjusted R² of 1.7%. However, when we look at performance for predicting obesity (BMI ≥ 30 kg/m²), the Asian participants had the best model fit (although still relatively poor), with an area under the curve of 0.587 (Figure 3B). This appears to be due to the underlying distribution of BMI and obesity within the groups. Within PAGE, Asians had the lowest proportion of obesity at 11.4%, compared to 44.2% within African American, 40.4% in Hispanic/Latino and 35.5% in Native Hawaiian participants, reflecting obesity prevalence among these groups in the general US population. (Figure 3C) Therefore, the risk score distribution was better able to differentiate the few Asian individuals with high risk for obesity within the top percentiles. In contrast, even with a higher adjusted R² in Hispanic/Latinos, African Americans, and Native Hawaiians, the PRS could not accurately distinguish these risk strata, because such a large proportion of participants were obese. This exemplifies the intersection of model fit due to heterogeneity in effect sizes (often from differential LD and allele frequencies) and differences in the prevalence of a trait, which both interact to complicate the translation of PRS to other populations.

Figure 3A-C.

Performance and distribution of Polygenic Risk Score (PRS) for obesity in PAGE participants.

(A) The adjusted R² of the PRS developed by Khera and colleagues¹⁵⁴ in PAGE participants, stratified by self-identified race/ethnicity. The risk scores were standardized by self-identified race/ethnicity and outliers beyond 4 standard deviations were removed for these analyses.

(B) The performance of the PRS on obesity (BMI≥30kg/m²), stratified by self-identified race/ethnicity. The highest area under the curve was found in East and South Asian participants.

(C) The distribution of obesity (BMI≥30kg/m²) by race/ethnicity-stratified decile of the PRS, demonstrating the differential distributions of obesity by race/ethnicity.

In summary, the predictive value of PRSs depends both on the relevant characteristics of the target (testing) dataset and the statistical power of the discovery (training) dataset—specifically, the enrichment of the genome-wide distribution of association test statistics attributable to aggregate, additive genetic effects. To date, PRSs have been developed using available GWAS as the training data, which have much larger sample sizes in Europeans than any other population.¹⁵⁵ PRSs developed from these findings are not necessarily transferable to other ancestral populations.¹⁵⁶ In fact, PRS accuracy is a function of recent human demographic history, such that a greater proportion of phenotypic variance is explainable by the PRS in target populations that are genetically more similar to the population studied in the discovery GWAS. As genetic distance between two populations increases, the polygenic predictive value of PRS decreases. A practical question is how to construct polygenic scores for recently admixed individuals or for individuals who are genetically distant from those populations reflected by the largest existing GWAS of CMD. The use of transethnic results to derive appropriate weights for more ancestrally diverse samples may increase prediction accuracy,¹⁵⁷ and MultiPred is a methodologic approach that combines PRSs based on European training data with PRSs based on training data from other target populations.¹⁵⁵ Current methods development is focused on best practices for handling allele frequency and LD differences both within and across populations. Given the limitations in assessing and comparing PRS across populations, great caution is advised in interpreting differences in PRSs across ancestries.

The promise of increasing diversity for future CMD research.

The PAGE Study and Authors of this paper have been involved in various initiatives that aim to promote the characterization of genomic variation in ancestrally diverse populations (e.g. the Hispanic Community Health Study/Study of Latinos¹⁵⁸), to develop statistical methods to analyze ancestrally diverse data^159,160 and to improve the accuracy of PRS. Other important efforts are also working to bring the promise of precision medicine to ancestrally diverse populations. One such longitudinal effort is the Multi-Ethnic Study of Atherosclerosis (MESA), which was designed to assess subclinical CVD and progression to incident CVD in a diverse population-based sample. At six recruitment centers, MESA recruited 6,814 participants from 2000-2002 across four race/ethnic groups (European (39%), African American (28%), Hispanic/Latino (22%), and Chinese American (12%)).¹⁶¹ Key findings from the study include the predictive power of coronary artery calcification (CAC) across ancestry for coronary events¹⁶², the association of air pollution with CAC progression¹⁶³, and extensive explorations of CVD biomarkers, such as lipoprotein-associated phospholipase A2 (Lp-PLA2)¹⁶⁴ and lipoprotein(a)¹⁶⁵, and optimal CVD risk thresholds for these biomarkers by race/ethnic group.¹⁶⁵ MESA and PAGE have been leaders in collaborative efforts in CVD genetic epidemiology, such as the National Heart Lung and Blood Institutes’ Trans-Omics for Precision Medicine program (TOPMed)⁷⁵ and the Cohorts for Heart and Aging Research in Genetic Epidemiology (CHARGE)¹⁶⁶, and MESA has been a pioneer in the generation of multi-omics data (such as gene expression and methylation) for multi-ethnic populations.¹⁶⁷

Additional efforts are ongoing to recruit new cohorts (https://www.theruralstudy.org/about/)¹⁶⁸ and biobank studies^169-172 with better representation of US and global populations. These include All of Us, funded by the US National Institutes of Health, which is building a large-scale biomedical data resource with the goal of reflecting the diversity of the US population.¹⁷³ The Million Veteran’s Program is also recruiting a large and ancestrally diverse cohort of US Veterans.¹⁷⁴ The population genetics company Color also recently announced plans to enroll 100,000 volunteers from under-represented groups to better assess the risk of myocardial infarction from low coverage WGS. As described above, it is especially critical that we study African descent populations⁶⁴, as early human migrations out of Africa (both forced and voluntary) took a portion of genetic diversity into Europe, East Asia and eventually into the Americas. Therefore, large genetic studies of African descent populations will likely improve the accuracy of PRSs for all populations, as well as the ability of precision medicine to reach those facing the highest CMD burden.

As the cost of sequencing the human genome continues to trend lower, whole genome sequencing is becoming available on many thousands of individuals worldwide; Table 2 presents a non-exhaustive list or global genome sequencing projects. Importantly, the ancestry biases present in GWAS arrays will no longer be a concern with sequencing data, and considerable effort should be placed on bringing these data together for novel discoveries and public health utility. Many countries (i.e., Australia, China, Dubai, Denmark, Estonia, France, Japan, Qatar, Saudi Arabia, Singapore, Turkey), and continental (H3Africa, Genome Asia 100K) efforts are also working hard to increase diversity of available genome sequence data worldwide. Of course, accomplishing the goals of coordinated discovery and sharing of data will likely take many years and follow the long timeline of the GWAS that preceded them. In addition, it will take the concerted efforts and diligence on the part of researchers and funders to prioritize these resources and ensure that they are used to their fullest extent to ameliorate bias in genomic studies and benefit human health globally.¹⁷⁶

Table 2.

Non-exhaustive list of global whole genome sequencing efforts complete or currently underway.

Study name	Location	Sample size	Ascertainment	Website
AFRICA
Africa Genome Variation Project	Sub-Saharan Africa	320	Population Genetics	https://ega-archive.org/studies/EGAS00001000960
H3 Africa Whole Genome Sequencing Study	Africa	350	Population Genetics	https://h3africa.org
Southern African Human Genome Programme	South Africa	24	Population Genetics	https://www.sahgp.org/index.php
ASIA
China Precision Medicine Initiative	China	100,000,000	Population-based	http://en.most.gov.cn/eng/organization/Mission/index.htm
Genome Asia 100K	Asia	100,000	Population-based and clinical studies	https://genomeasia100k.org
Japan Genomic Medicine Program	Japan	100,000	Population-based and clinical cohorts, drug discovery	https://www.amed.go.jp/en/program/index05.html
Singapore 10k	Singapore	10,000	Random sample for Population genetics	https://www.sciencedirect.com/science/article/pii/S0092867419310700
Singapore Sequencing Malay Project	Singapore	100	Random sample	https://blog.nus.edu.sg/sshsphphg/singapore-sequencing-malay/
AUSTRALIA
Genomics Health Futures Mission	Australia	200,000	Population-based, clinical studies, and rare diseases	https://www.health.gov.au/initiatives-and-programs/genomics-health-futures-mission#what-are-the-goals-of-the-genomics-health-futures-mission
Study name	Location	Sample size	Ascertainment	Website
EUROPE
FarGen - Faroe Genome Project	Denmark	1,500	Population Genetics	https://www.fargen.fo/en/home/
Genome Denmark	Denmark	~60,000	Population-based and clinical studies	http://www.genomedenmark.dk/english/
Estonian Genome Project	Estonia	52,000	Cohort study	https://genomics.ut.ee/en/about-us/estonian-genome-centre
Genomics Medicine Plan 2025	France	235,000 per year	Cancer, rare and common diseases	https://aviesan.fr/fr/aviesan/accueil/toute-l-actualite/plan-france-medecine-genomique-2025
RADICON-CL	Netherlands	*Could not confirm sample size	Rare diseases	https://www.wgs-first.nl/en/project
Genomics England (GEL)	United Kingdom	100,000	Cancers, rare diseases	https://www.genomicsengland.co.uk
Scottish Genomes Partnership	United Kingdom	2,588	cancers, rare diseases, population study	https://www.scottishgenomespartnership.org
MIDDLE EAST
Dubai Genomics	Dubai, UAE	3,000,000	Population-based and clinical studies	https://www.dha.gov.ae/en/Pages/DubaiGneomicsAbout.aspx
Qatar Genome Project	Qatar	20,000	Cohort study	https://qatargenome.org.qa
Saudi Arabia Genome Project	Saudi Arabia	>100,000	Population-based and clinical studies	https://genomics.saudigenomeprogram.org/en/
Turkish Genome Project	Turkey	100,000	Population-based and clinical studies	https://www.bbmri-eric.eu/news-events/turkish-genome-project-launched/
Study name	Location	Sample size	Ascertainment	Website
UNITED STATES
All of Us	United States	1,000,000	Cohort study	Sequencing tentative
NHGRI Genome sequencing project	United States	141,000	Mixed trait	https://www.genome.gov/Funded-Programs-Projects/NHGRI-Genome-Sequencing-Program
NHLBI TopMED	United States	140,000	Mixed trait	https://www.nhlbiwgs.org
NYCKidSeq	United States	1,130	Undiagnosed neurologic, cardiac, or immune disorders	https://nyckidseq.org
Undiagnosed Diseases Network	United States	1,600	Undiagnosed rare diseases	https://undiagnosed.hms.harvard.edu

Ethical, Legal and Social implications of genetics research.

In addition to the implications of genetic research more generally, there are specific ethical, legal, and social concerns surrounding research in underrepresented populations (see Brothers and Rothstein for a comprehensive review).¹⁷⁷ These authors explain that the increase in genomics-enriched health information and the potential of personalized approaches to exacerbate health disparities are key issues that require attention. Moreover, it is critical that these discussions and eventual policy decisions prioritize the issues of privacy, potential for discrimination, and changes in physician-patient relationships and liability. Specifically, it is argued that the current availability of genotype-phenotype information, rising costs, and diminished access to health care, as well as information technology, are all possible sources of increasing health disparities.

As described above, studying diverse populations is necessary for scientific understanding and improved equity and inclusion in the field. However, researchers must carefully consider how they define and approach specific populations in order to avoid essentializing race and racism and hindering engagement. One prominent recent example of the essentialization of race in cardiovascular medicine was the US Food and Drug Administration’s approval of BiDil to treat heart failure in African-Americans only¹⁷⁸, in the absence of pharmacogenomic variation to support its race-specific marketing.⁴⁴ This decision was met with widespread criticism, as the authoritative position of the US Food and Drug Administration could have given the biological reification of race an appearance of legitimacy, even though there was no biological basis for the race-specific approval.¹⁷⁹ A second example comes from the non-disclosure that individuals with certain CYP2C19 alleles do not respond equally to Plavix (clopidogrel bisulfate), an inhibitor of platelet aggregation, as they have a reduced hepatic capability for 2C19 to convert it to its active metabolite.¹⁸⁰ The states of Hawaii and New Mexico have filed civil suits against Bristol-Myers in 2014 and 2016, respectively, for wrongfully acquiring profits from sales within their borders. The alleles in question [CYP2C19*2, rs4244285 (c.681G>A) and CYP2C19*3, rs4986893 (c.636G>A)] are found at higher prevalence in the populations of these states –specifically East Asians, Native Hawaiians, Pacific Islanders, Native Americans and Hispanic/Latinos – than in European descent populations (Table 1). Moreover, there is rising concern that race-specific drug development, labeling, or marketing, may reinforce race/ethnic health disparities by increasing drug costs and demands for evidence to support efficacy and necessity in under-represented groups.⁴⁴

Some social scientists have also expressed concerns about etiologic research that investigates the genetic origins of disease and how they differ by race or ethnic groupings, as proxies of shared ancestral background. They worry that this research may inadvertently prompt the public to think of racial groups as biologically distinct categories. If genetic differences at least partially impact differential risk for disease, the public may think that this must also be true for other human traits.¹⁸¹ This hypothesis was tested by Phelan and colleagues¹⁸² using a nationally-representative survey where respondents were randomized to one of four conditions, which consisted of reading a different vignette followed by an assessment of beliefs in biologically essential racial differences. The four conditions included 1) the “Backdoor” vignette describing a genetic variant that is more strongly associated with heart attack in African Americans than in European Americans, 2) a vignette describing race as entirely socially constructed 3) a vignette describing racial groups as broadly genetically distinct from one another, 4) and a no-vignette control group. Results showed that endorsement of essential racial differences was greater for individuals assigned to the “Backdoor” vignette than to the “race as a social construction” vignette or to the non-vignette control group. There was also no difference in endorsement of essential racial differences between the groups assigned the “Backdoor” vignette and the “race as essential biological category” vignette.¹⁸² These findings imply that researchers should anticipate misinterpretations of findings and explain clearly the conclusions that can and cannot be drawn from them.

It is becoming increasingly clear that the underlying evidence base of much genetic research insufficiently represents the same populations historically underrepresented in biomedical research – and that both the quantity and quality of the available information may be issues.¹⁸³ Skepticism about and suspicion around the research enterprise by potential research participants^184,185 has been offered as one source of the persistence of genomics-associated disparities in data availability. This is a critical issue because the development of precision health care and ‘learning healthcare systems’ require the inclusion of health and genomics information from currently underrepresented populations.¹⁸³ Principles of implementation science are likely well suited in this domain, aiming to improve inclusion and representation of the genomics evidence base in CMD and health care in general.¹⁸⁶

Generally, biomedical literature has focused on the ethical implications of genetic research as it relates to appropriate disclosure and accurate interpretation of primary research results and unintended findings¹⁸⁷ In fact, there is rising concern that the use of genetic testing for CMD prevention may not result in measurable changes in either increased motivation for lifestyle change, or healthier lifestyles with respect to diet or physical activity.¹⁸⁸ Racial/ethnic minorities already face structural barriers to healthy lifestyles and unhealthy built environments in the US.¹⁸⁹ High PRS estimates for obesity, or a clinician’s perception thereof, have been shown to result in higher-quality patient-provider interactions as compared to patients with more environmentally driven obesity.¹⁹⁰ Increasing use of genetic testing may overemphasize the relative importance of the biological versus social determinants of health,¹⁹¹ which may divert limited resources to provide assistance to those without insurance or fail to address the social or structural determinants of health, which both differentially impact certain US groups.¹⁸⁹ Recent research demonstrates that a common prediction algorithm used to identify patients at ‘high need’ underestimated African American patients’ risk compared to that of European Americans. This was because the algorithm used health care costs as a proxy for health care needs and on average less resources are spent on African Americans.¹⁹² If uptake of CMD genetic testing becomes more common among European Americans, their average healthcare costs will increase and so will the underestimation of need among African American patients.

CMD genetic testing, PRS estimation, and other precision medicine activities using genetic information may have differential implications for racial/ethnic groups or other vulnerable populations who may have experienced discrimination or stigmatization in the past.¹⁸⁹ First, as CMD genetic research is deployed more frequently to predict risk and classify individuals as high risk, to identify those who can benefit from behavior change or therapeutic interventions,¹⁹³ or to find those individuals who may or may not respond to standard therapies,¹⁹⁴ such labels may compound the discrimination and stigmatization already experienced by individuals in underrepresented groups.¹⁸⁹ Second, the portability of the resulting predictions from PRS across ancestries and environmental exposures is a major concern.^145,147 In part, this has led to direct-to-consumer prediction algorithms for complex conditions being advertised as restricted to one race.¹⁹⁵ Moreover, for ancestrally diverse individuals and populations, common methodologic approaches to estimate ancestry, identify ancestral outliers, and/or stratify samples into relatively homogenous ancestral groups, may conflict with their own self-identities,¹⁹⁶ and appear to endorse the essentialization of racial categories.

Additional legal challenges may arise as the range and scale of genetic testing advances. Currently healthcare maintenance organizations are working to integrate genetic screenings into their routine clinical care and medical record systems by following the American College of Human Genetics list of 59 genes (including several cardiovascular diseases with high penetrance) as actionable incidental findings¹⁹⁷ It stands to reason that routine genetic predictions for complex CMD traits and diseases may soon be added and this could lead to further legal challenges to the Genetic Information Nondiscrimination Act in the US, or similar legislation in other countries,¹⁹⁸ or be used to adjust health insurance premiums or deny coverage of the relatives of genotyped patients.

A number of specific ethical concerns remain around how individual genetic prediction using PRS is applied outside of boundaries of clinical research ¹⁹⁹. Some direct-to-consumer genetic testing companies that provide these predictions may offer lower quality genetic testing and counseling than otherwise provided by laboratories used in clinical care. As an example, Genomic Prediction, with a Clinical Laboratory Improvement Amendments-certified laboratory, currently provides PRSs for a number of CMD diseases and traits (e.g. T2D, CAD, myocardial infarction, hypercholesterolemia and hypertension) for embryos prior to implantation.²⁰⁰ The broad use of PRS for embryonic selection for multifactorial traits and diseases like CMD still remains an ethical question for the field,²⁰¹ as it could further reinforce social intolerance of diversity.¹⁹¹

CONCLUSIONS

Understanding how genetic, environmental, and socio-cultural factors interact to influence disparities in CMD risk is imperative for improving individual and public health. There is a critical need for larger, more diverse genetic studies of CMD so that we can minimize the research gap between the global burden of CMD and the samples and findings available in current genetic resources and databases. CMD genetic research has great potential to inform prevention and personalized medicine. However, this potential will only be realized if CMD research findings, including PRS estimation, are derived from samples that reflect the true ancestral diversity of the target population. Similarly, the potential benefits of tailored lifestyle interventions or therapies cannot be fully realized until equitable representation across ancestrally diverse populations is achieved. Such an inclusive approach necessitates international collaboration, such as that exemplified by the GIANT²⁰² (Genetic Investigation of Anthropometric Traits - https://portals.broadinstitute.org/collaboration/giant/index.php/GIANT_consortium), MAGIC²⁰³ (Meta-Analyses of Glucose and Insulin-related traits Consortium - https://www.magicinvestigators.org), GLGC²⁰⁴ (Global Lipids Genetics Consortium - http://lipidgenetics.org), and CHARGE consortia, as well as other efforts to increase diversity in genetic epidemiology studies.

Inadequate representation is a pressing limitation of current CMD genetic research, and it poses an additional challenge related to engagement of individuals from underrepresented populations. Many scholars have outlined several important steps to advancing engagement in genetic research for minority communities in the US and Canada, many of whom face CMD disparities.^194,205,206 These steps include the need to acknowledge barriers to participation and to incorporate community-based participatory research approaches and benefits-sharing models to avoid possible injustices around commercialization of data and medical innovations. Moreover, additional resources are needed to address barriers to higher education and to nurture the genetics careers of individuals from underrepresented groups in the field.^207-211

The robust identification of genomic regions associated with CMD and other health outcomes has created major opportunities for the improvement of individual and population health. At the same time, it has generated new challenges that may affect the utility of this information and its equitable integration into the broader healthcare system. In this review, we identify the following key challenges: 1) lack of diversity in the genotype-phenotype evidence base for CMD; 2) lack of diversity in genomic reference panels that may be hampering our ability to identify causal variants for CMD; 3) lack of clearly defined designations for the classification of race/ethnicity and ancestry, which complicate analyses of the genetic contributions to disease; 4) the unequal generation of health-associated genomic data and prediction accuracies that may further exacerbate CMD disease disparities. Despite these challenges, unprecedented opportunities lie ahead for reducing the burden of CMD and improving health via the equitable generation and use of genomics data in diverse populations.

As we have documented in this review with respect to CMD, large scale sharing and harmonization of GWAS data is already occurring. However, these large scale initiatives are just beginning and it is our responsibility as researchers and stakeholders to insist on data integration. Related to this, and also of relevance for GWAS studies, is the issue of data harmonization around self-identified race/ethnicity. Although we have argued in this review that race and ethnicity are socially and culturally defined constructs, the continued use of self-identified categories in genetic studies may capture unexplained phenotypic variability. Novel machine learning approaches have been developed to address this problem^212,213, for example the HARE framework developed for the MVP.²¹³ HARE provides ancestry clustering that benefits from both self-identified labels and genetically-informed ancestry. Briefly, HARE combines self-identified race/ethnicity and supervised genetic clustering (determined via support vector machines) to improve clustering beyond that derived from either self-identified or genetic ancestry only.²¹³ This allows researchers to build predictors for consistent levels of ancestry, allowing for realistic harmonization of labels even within biobanks where self-identification may be limited or inconsistent.

We must make both the (1) increased generation and (2) increased integration of data in diverse populations a public health priority for CMD and other complex diseases. Achieving this goal will require concerted efforts by social, academic, professional and regulatory stakeholders, as well as the communities bearing the highest burden of CMD, and must be based on principles of equity and social justice.

Non-Standard Abbreviations and Acronyms:

BMI

Body mass index

CAD

Coronary artery disease

CADD

Combined annotation Dependent Depletion

CKD

Chronic kidney disease

CMD

Cardiometabolic disease

CNV

Copy number variant

CVD

Cardiovascular disease

DALY

Disability adjusted life years

FRS

Framingham Risk Score

GWAS

Genome-wide association analysis

HDL

High density lipoprotein

LD

Linkage disequilibrium

LDL

Low density lipoprotein

MAF

Minor allele frequency

MESA

Multi-Ethnic Study of Atherosclerosis

PAGE

Population Architecture using Genomics and Epidemiology

PKC

Protein kinase C

PRS

Polygenic risk score

SBP

Systolic blood pressure

SNP

Single nucleotide polymorphism

SNV

Single nucleotide variant

TOPMed

Trans-Omics for Precision Medicine program

T2D

Type 2 diabetes

US

United States