A Genome-First Study of Familial Hypercholesterolemia Comparing African and European Ancestry Individuals

Abstract

BACKGROUND:

Familial hypercholesterolemia (FH) is an inherited disorder characterized by lifelong elevated LDL-C (low-density lipoprotein cholesterol) and increased risk for premature myocardial infarction. FH research has focused on European populations and, consequently, estimates of global FH burden primarily reflect this ancestry, with limited data available from other groups.

METHODS:

We examined the prevalence and clinical outcomes of FH among 104 300 African ancestry individuals enrolled in 3 US-based cohorts: the National Institutes of Health’s All of Us, Mount Sinai’s BioMe, and Geisinger’s MyCode. Genetic variants were evaluated according to standards provided by the Clinical Genome Resource’s FH Variant Curation Expert Panel and grouped as pathogenic variants or variants of unknown significance (VUSs). Participants were assigned to European and African ancestry groups based on genetic similarity to reference populations. Clinical outcomes were LDL-C and myocardial infarction. Analyses were adjusted for age and sex. Results were meta-analyzed across cohorts.

RESULTS:

The prevalence of a pathogenic variant was similar in the African (1 in 306) and European (1 in 273) ancestry groups. The LDL-C elevation associated with pathogenic variants was 20.81 mg/dL (95% CI, 16.17–25.45) greater in individuals with African ancestry compared with their counterparts with European ancestry. Individuals with African ancestry had 1.61 (95% CI, 1.42–1.83; P=1.2×10^-12) higher odds of having a VUS compared with individuals with European ancestry and a 10.01 mg/dL (95% CI, 6.13–13.88) greater elevation in LDL-C compared with individuals with European ancestry with a VUS. Although pathogenic variants in both ancestries conferred a 2- to 3-fold increased risk of myocardial infarction, having a VUS only conferred increased risk among the African ancestry group (odds ratio, 1.91 [95% CI, 1.18–3.10]).

CONCLUSIONS:

The prevalence of pathogenic variants did not significantly vary between African and European ancestry groups. VUSs were both more prevalent among individuals of African ancestry and associated with an increased risk of myocardial infarction equivalent to that of a pathogenic variant. These findings suggest that dependence on existing resources for variant classification could contribute to underdiagnosis of FH in individuals of African ancestry.

Clinical Perspective.

• Familial hypercholesterolemia (FH) is an underdiagnosed and undertreated cause of premature cardiovascular disease, and yet it remains poorly understood outside European populations.

• We report the prevalence and natural history of molecularly defined FH in a large cohort of African ancestry individuals ascertained from 3 health care–based populations in the United States.

• Compared with European ancestry individuals, we find that individuals of African ancestry present with a more severe hypercholesterolemia phenotype in the presence of a pathogenic FH variant and are more likely to have a suspicious variant of unknown significance.

• Individuals of African ancestry with a pathogenic FH variant typically develop a more severe form of hypercholesterolemia than those of European ancestry, highlighting the need for closer monitoring and tailored treatment to meet cholesterol goals.

• Although individuals of African ancestry may be more likely to receive inconclusive results from FH genetic testing, the elevated risk of myocardial infarction associated with a variant of unknown significance in this population suggests they could benefit from an intensified treatment approach.

Familial hypercholesterolemia (FH) is an underdiagnosed and undertreated cause of premature atherosclerotic cardiovascular disease (ASCVD).^1–4 The heterozygous form of FH is caused by a pathogenic variant in the low-density lipoprotein receptor (LDLR), apolipoprotein B (APOB), or proprotein convertase subtilisin/kexin type 9 (PCSK9) genes.^5,6 In one of the largest systematic reviews and meta-analyses performed to date, the population prevalence of FH was estimated to be 1:311 (95% CI, 1:250 to 1:397).⁷ The generalizability of this prevalence is unclear as it was largely derived from European and high-income nations and not all studies included in the meta-analyses relied on genetic confirmation of a diagnosis. There is a notable lack of data on FH from regions outside of Europe, and few studies exist on individuals of African ancestry.^8,9

Given the actionability of an FH diagnosis, particularly in young patients, equitable access to accurate genetic information is needed to ensure that diagnostic gaps do not exacerbate existing health disparities. Identifying a pathogenic variant enables cascade screening for family members and facilitates insurance authorization for specialized therapies, such as PCSK9 inhibitors. Recently, studies of biobanks including All of Us, TOPMed (Trans-Omics for Precision Medicine), UK Biobank, and the Million Veteran Program and others have reported on FH across global ancestries.^10–16 They show in all ancestries the excess ASCVD risk associated with FH beyond measured LDL-C (low-density lipoprotein cholesterol), pervasive gaps in the use of lipid lowering medication, and a lower attainment of treatment targets. A limitation of published biobank studies that include participants from diverse ancestry groups is that the ancestry-specific analyses largely focused on the prevalence of known pathogenic variants, with limited evaluation of clinical outcomes or variants of unknown significance (VUSs). Although the Million Veteran Program study conducted ancestry-specific analyses, FH ascertainment was restricted to a limited set of pathogenic variants present on a genotyping array, capturing only a fraction of the pathogenic variation present in the population. Furthermore, disparities among African ancestry individuals with FH remain inadequately investigated. Given heterogeneity in participants across biobanks with respect to demographics, medical history, and ascertainment, integration of data from multiple cohorts is needed to assess the generalizability of any detected disparities.¹⁷

For many monogenic disorders, our understanding of their genetic architecture has been drawn largely from cohorts of European ancestry. A recent study of dilated cardiomyopathy found that individuals of African ancestry are less likely to clinically actionable variants and more likely to have a VUS,¹⁸ a difference that may be driven by the underrepresentation of African ancestry data in genetic databases for variant interpretation.¹⁹ Elevations in VUS classifications among African ancestry cases may reflect a disparity in genetic testing accuracy.^20,21 In the case of FH, the distribution of genetic variation in African ancestry populations is not well described.

The objective of this study was to investigate the genetic architecture and clinical presentation of FH among individuals of African ancestry. Building on previous work, our study integrates exome- and whole-genome sequence data from multiple US biobanks to compare FH in African and European ancestries across diverse health care settings, including large rural and urban systems. It is important to note that our study leverages these biobanks to advance the understanding of FH in African ancestry populations by examining the interplay among ancestry-specific FH genetic architecture, gaps in clinical variant databases, and disparities in patient outcomes. To do this, we conducted a genome-first study of FH in >100 000 individuals with African ancestry among 3 large biobanks based in the United States, and almost 540 000 individuals of European ancestry identified from the same biobanks were used as a comparison group. We applied a framework for classifying variants in FH genes based on standards adapted from the Clinical Genome Resource FH Variant Curation Expert Panel.^22,23 Based on these data, we present a comprehensive characterization and comparison of FH in populations of African and European ancestry. First, we compared the prevalence and genetic architecture of FH between the 2 ancestry groups. Next, we evaluated differences in the severity of the hypercholesterolemia phenotype and the associated risk of myocardial infarction (MI). Last, we assessed the proportion of individuals with VUSs and their associations with clinical outcomes.

METHODS

Data Availability

All analyses reported in this article were performed on existing genomic and phenotypic datasets. All sequencing data used in this study are available on the All of Us Researcher Workbench in the v8 release. Researchers can register to access this resource at https://www.researchallofus.org/. The MyCode and BioMe datasets and scripts used in analyses can be made available by contacting the investigators directly.

Study Cohorts

Geisinger’s MyCode Community Health Initiative is a large, rural US health care cohort of >360 000 consented individuals unselected for age, sex, or clinical diagnosis.²⁴ This study includes a subset of 157 067 MyCode individuals with exome sequences linked to electronic health record (EHR) data, known as the DiscovEHR cohort. Informed consent was obtained from adult participants and from the parents or guardians of pediatric participants and assent was obtained from those pediatric patients >7 years old. The Geisinger MyCode Governing Board and Institutional Review Board approved the study.

The All of Us Research Program is an National Institutes of Health–funded research effort to enroll 1 million or more individuals from the United States to contribute a DNA sample to a large biobank and linkage to their longitudinal health data.²⁵ Informed consent for all individuals in All of Us was conducted in person or through an eConsent platform that includes primary consent, Health Insurance Portability and Accountability Act Authorization for Research use of EHRs and other external health data, and Consent for Return of Genomic Results. The protocol was reviewed by the All of Us Institutional Review Board. For this study, we analyzed genetic and clinical data from 447 278 individuals in All of Us. For our study, EHRs, surveys, and genetic information were extracted from the All of Us v8 controlled tier dataset in 2025.

The BioMe Biobank is a large biorepository of DNA and plasma samples and phenotypic (questionnaire-based and EHR-linked) and genomic data collected from ancestrally diverse participants across the Mount Sinai Health System in New York City.²⁶ Since BioMe’s launch in 2007, >60 000 participants have been recruited from >26 outpatient sites, of whom 55 328 individuals’ genetic and linked clinical data were analyzed in this study. Informed consent was obtained from the participants at the time of study enrollment to permit the use of samples and de-identified linkable past, present, and future clinical information from EHRs for research purposes.

Genetic Data

Exome sequencing of the MyCode cohort was performed in collaboration with the Regeneron Genetics Center as described previously.^27,28 Exome capture was performed using Integrated DNA Technologies xGen kits for the individuals according to the manufacturer’s protocol. Multiplexed samples were sequenced using 75×75 base-pair paired-end sequencing on an Illumina v4 HiSeq 2500 or the NovaSeq 6000 platform. All reads were aligned to the GRCh38 reference genome using the Burrows-Wheeler Alignment tool.²⁹ Variants were called using weCall v1.1.2,³⁰ followed by joint calling and harmonization across freezes using GLnexus.³¹ Variants were excluded based on genotype quality (<70), depth of coverage (<20), and allelic balance (<0.2). Six pathogenic variants identified in six individuals below these thresholds were independently confirmed through Geisinger’s return‑of‑results pipeline and were included in the present analyses.³²

Genome sequencing of the All of Us cohort was performed by the All of Us Genome Center as described previously.³³ Pooled libraries are loaded on the Illumina NovaSeq 6000 instrument. After demultiplexing, genome analysis was performed using the Illumina DRAGEN platform.³⁴ The DRAGEN pipeline consists of highly optimized algorithms for mapping, aligning, sorting, duplicate marking and haplotype variant calling and makes use of platform features such as compression and BCL conversion. Alignment used the GRCh38dh reference genome. In All of Us, the variant quality filters tested and recommended by the All of Us Data and Research Center for use in the cohort were used, which excluded variants based on genotype quality (<20), depth of coverage (<10), and allelic balance (<0.2).

Whole exome sequencing of the BioMe cohort was performed by the Regeneron Sequencing Center for 30 864 participants using Integrated DNA Technologies xGen v1 target kits and for 28 227 participants using TWIST Comprehensive Exome probes. Multiplexed samples were sequenced using 75x75 base pair paired-end sequencing on an Illumina v4 HiSeq 2500. The CRAM alignment files were generated using the OQFE protocol as described previously.³⁵ DeepVariant v0.10.0 was used for variant calling. All reads were aligned to the GRCh38 reference genome. Variants were excluded based on genotype quality (<20), depth of coverage (<20), and allelic balance (<0.2). The 55 328 BioMe participants used for the analysis were genotyped using Regeneron’s Global Screening Array (GSA-24v1-0_A1). All genotypes were imputed using TOPMed reference panel freeze 5 with ~65 000 samples and ~230 million variants using the Michigan Imputation server pipeline.^36,37

Classification of Genetic Variants in LDLR, APOB, and PCSK9

Single nucleotide variants in canonical FH genes in MANE (Matched Annotation from NCBI [National Center for Biotechnology Information]) and EMBL-EBI (European Molecular Biology Laboratory's European Bioinformatics Institute) select transcripts LDLR (NM_000527.5), APOB (NM_000384.3), and PCSK9 (NM_174936.4) were annotated with Ensembl's Variant Effect Predictor (version 109), NCBI's clinical variant database ClinVar (released February 6, 2024), and LDLR-specific annotations provided in the publication by the Clinical Genome Resource FH Variant Curation Expert Panel.²³ Variants identified in FH genes were placed into 10 non–mutually exclusive categories that reflect the highest strength and type of evidence supporting a molecular diagnosis of FH per variant (Table S1). Of 3991 variants annotated as missense or predicted loss-of-function, 365 met criteria for at least 1 FH variant category (Tables S2, S3A, and S3B). Variants falling into categories 1 through 3 are supported by strong evidence based on submissions of variant interpretations with multiple assertion criteria by submitters to ClinVar or a high-confidence predicted loss-of-function variant in LDLR. Variants that meet the criteria of categories 1 through 3 are referred to as pathogenic variants throughout the remainder of the text. Categories 4 through 10 represent variants with some but not sufficient evidence of disease causation. These categories were modeled on the criteria in the FH Variant Curation Expert Panel article and are based on functional impact, allele frequency, in silico classification of missense and splice variants, variation hotspots or regions of enrichment, and variants in highly conserved cysteine residues known to be critical for receptor function.^23,24,35 Table S1 shows the definition of each category and the corresponding similar criteria from the FH Variant Curation Expert Panel guidelines. Variants that do not meet the criteria for pathogenicity but fall within categories 4 through 10 are referred to as VUS in the remainder of the text. Only variants in LDLR were considered for the VUS category, as currently only LDLR has direct guidance available from the Clinical Genome Resource for variant categorization. REVEL (Rare Exome Variant Ensemble Learner) was the in silico tool used for impact of missense variants on protein function. REVEL threshold for elevated (0.773≤REVEL<0.932) and extreme (0.932≤REVEL) variants were applied based on calibration to align with American College of Medical Genetics criteria, enabling their use as supporting evidence for pathogenicity.^37,38 LDLR variants were also annotated with the recently developed function sequence map of LDLR created by saturation mutagenesis of the gene.³⁹ The scores of the function sequence map reflect the impact that coding variants have on protein function and abundance. We annotated LDLR variants identified in our cohort with the functional score as this metric is stated to meet current Clinical Genome Resource guidelines for level 1 evidence for classification of FH variants.

We did not evaluate VUSs in APOB or PCSK9. In contrast with LDLR, where loss‑of‑function variants are the primary mechanism underlying FH, pathogenic variants in APOB and PCSK9 are missense changes that cause specific functional defects resembling gain‑of‑function effects. Because these mechanisms are inherently more difficult for in silico prediction tools to assess accurately, VUSs in these genes were not included in our analysis.³⁸

Lipid-lowering variants in APOB and PCSK9 were defined according to the same criteria as the FH pathogenic variants above, with the exception of 1 PCSK9 variant, C679Ter (rs28362286).⁴⁰ C679Ter is located in the final exon of PCSK9 and would normally be expected to escape nonsense-mediated decay. As premature stop codons located near the 3′ end of a gene often have uncertain functional consequences because they avoid mRNA degradation, the resulting shortened protein may retain partial activity, and the specific functional impact of C-terminal truncating variants such as C679Ter is often difficult to predict. However, this variant has been seen previously to be more common in individuals with African ancestry.⁴¹ After investigation in our data, we see that this variant does have a large effect on LDL (low-density lipoprotein) comparable with predicted loss-of-function variants upstream (Figure S1) and so was included in our analyses as a lipid-lowering variant.

Clinical Outcomes

From the MyCode, All of Us, and BioMe cohorts, we extracted 1 511 828, 1 061 754, and 375 921 LDL-C measurements (LOINC [Logical Observation Identifiers Names and Codes]: 13457-7) collected during outpatient visits, respectively. The concept names used for the data pull of LDL-C measurements in All of Us are shown in Figure S2. The primary LDL-C outcome of interest was maximum EHR-recorded LDL-C. As a sensitivity analysis, we also calculated a median LDL‑C adjusted for lipid‑lowering therapy by dividing each LDL‑C value by 0.7 when a lipid‑lowering medication had been prescribed 28 to 365 days before and then taking the median of these adjusted values for each individual. No adjustment for lipid-lowering medication use was applied to the primary outcome of maximum LDL. Descriptive data about the LDL measurements used in this study are presented in Table S4. Missingness by cohort for LDL analyses are available in Table S5. Individuals without demographic LDL data were included in variant prevalence estimates.

MI/revascularization outcome was defined based on the International Classification of Diseases, Ninth Revision and Tenth Revision, codes and Current Procedural Terminology codes present in the EHR (Table S6). The MI/revascularization outcome includes both prevalent and incident diagnoses during follow-up. Cases were defined as individuals who had a diagnosis of ASCVD present in their her, and controls were those who did not. Values and codes used to pull phenotype information in the All of Us dataset are available in Table S7A to S7C.

Lipid-lowering medications were defined as 3-hydroxy-3-methylglutaryl-coenzyme A reductase inhibitors (eg, statins), PCSK9 inhibitors, nonstatin medications (ie, bile acid sequestrants, cholesterol absorption inhibitors, fibrates, and niacin), and combination medications. Lipid-lowering medications with at least 1 medication order and their corresponding RxNorm code are shown for each cohort in the Supplemental Material (Tables S7A, S8A, and S8B).⁴²

To compare the potency of lipid-lowering medications across different statin medications and doses, we calculated a standardized measure of LDL‑C–lowering potency for all compatible medication orders in the 3 cohorts. This measure, the Defined Daily Dose, estimates the relative LDL‑C reduction based on 4 statin compounds—atorvastatin, lovastatin, pravastatin, and simvastatin—and their corresponding dose.⁴³ For each statin order, we used the equivalency table of Oni-Orisan et al to determine its relative LDL‑C–lowering effect given the specific compound and dose.⁴⁴ In this framework, a Defined Daily Dose of 1.0 represents the LDL‑C–lowering effect of taking 40 mg lovastatin daily. The equivalency table was originally developed by the Food and Drug Administration and later refined through analysis of the Kaiser Permanente GERA (Genetic Epidemiology Research on Adult Health and Aging) cohort.⁴⁴ The revised table was used in the present analyses. For each individual, we calculated a median Defined Daily Dose across medication orders available for 1 of the 4 statin medications where the individual had dose information available. We note that the Defined Daily Dose estimation reflects only the LDL‑C–lowering response to statin orders and does not account for the effects of nonstatin therapies.

Assignment of Genetic Ancestry

Genetic ancestry was quantified for individuals in MyCode, All of Us, and BioMe using a standardized approach. First, we conducted principal components analysis on genotype data from the 1000 Genomes Project, including individuals from 3 ancestry groups (European, African, and East Asian) from the following populations: Yoruba in Ibadan, Nigeria (YRI), Luhya in Webuye, Kenya (LWK), Mandinka in The Gambia (MAG), Mende in Sierra Leone (MSL), Esan in Nigeria (ESN), Han Chinese in Beijing, China (CHB), Japanese in Tokyo, Japan (JPT), Han Chinese South (CHS), Chinese Dai in Xishuangbanna, China (CDX), Kinh in Ho Chi Minh City, Vietnam (KHV), Toscani in Italia (TSI), Finnish in Finland (FIN), British From England and Scotland (GBR), Iberian Populations in Spain (IBS), and Utah residents with Northern and Western European ancestry (CEU). For the principal components analysis, we selected 25 189 autosomal variants that met the following criteria: present in both the 1000 Genomes Project and all biobank cohorts, minor allele frequency >0.01 in each dataset, located outside regions of long-range linkage disequilibrium and genomic regions under recent selection,²⁵ nonambiguous, and without discordant alleles across cohorts. Samples from each biobank were then projected onto the principal components analysis space to generate principal components (Figure S3). Admixture of each individual was estimated using Rye.⁴⁵ Individuals were assigned to 1 of 4 ancestry groups—African, European, East Asian, or not classified—based on their estimated ancestry proportions. Individuals with >50% ancestry from a single continental group were assigned to that group; those not meeting this threshold were categorized as not classified. All analyses were restricted to individuals classified as either African or European ancestry.

Statistical Methods

Linear and logistic regression was performed for LDL-C values (linear) and MI outcomes and lipid-lowering medication use (logistic regression). Covariates in all analyses were current age of the participant and sex assigned at birth. Individuals missing data on age or sex were excluded from analyses. Meta-analysis was conducted to combine results across cohorts using inverse-variance weighted fixed-effects models implemented in the “metafor” R package. Our analyses were conducted within mutually exclusive ancestry groups in each cohort and then meta-analyzed across the 3 cohorts. Each meta-analysis included at most 1 independent effect estimate per cohort per ancestry group. This design precludes correlated study-level effect sizes by avoiding sample overlap and is consistent with standard meta-analytic assumptions of independence. Additional mutually exclusive strata within each ancestry group were created for secondary analyses including sex and FH variant category. All results reported in the main text are derived from these ancestry-stratified meta-analyses. Differences in meta-analyzed effect estimates by ancestry group were compared using the Wald test. All analyses and visualizations were conducted using R version 4.4 and the following packages: data.table, dplyr, epiDisplay, forcats, ggplot2, haven, labelled, lubridate, readxl, tidyr, and stringr.

Use of All of Us data requires compliance with their data privacy restrictions, which include a prohibition on reporting individual counts of <20 or any data where a count of <20 could be calculated. As such, any counts of <20 for the All of Us cohort were reported as <20 instead of the count, including in instances where other counts had to be removed or modified to prevent the calculation of a count of <20. MI prevalence estimates and number of cases were rounded to prevent inference of cell counts <20.

RESULTS

Cohort Characteristics

The study sample encompassing the MyCode, BioMe, and All of Us cohorts includes 104 300 individuals of African ancestry and 539 184 individuals of European ancestry (Table 1). The ancestry composition differs between biobanks: MyCode individuals are primarily of European ancestry (97.5%), whereas All of Us and BioMe provide greater representation from African ancestry, 19% and 25%, respectively. The clinical characteristics of the cohorts stratified by ancestry are shown in Tables S9A and S9B. MyCode individuals were the most likely to have a record of being prescribed a lipid-lowering medication (57.7%, compared with 43.8% in BioMe and 30.8% in All of Us). Across the entire study population, the sample was 57.4% female, 38.2% having been treated with a lipid-lowering medication, and 6.3% with MI. Unless otherwise specified, the findings presented in the Results section are derived from within-biobank analyses, meta-analyzed together to report the results of the combined cohort.

Table 1.

Clinical Characteristics

Prevalence of FH by Ancestry

We identified 342 African ancestry and 2033 European ancestry individuals with a pathogenic FH variant in the combined cohort (Table 2). In total, 230 unique pathogenic variants were identified. Fifteen and 178 pathogenic variants were only found in the African or European ancestry subgroups, respectively. The prevalence of pathogenic FH variants in the European ancestry group was 1 in 273, consistent with previously reported estimates. A similar prevalence, 1 in 306, was observed in the African ancestry group. There was no significant difference in prevalence for pathogenic variants by European or African ancestry (odds ratio [OR], 0.89 [95% CI, 0.79–1.00]; P>0.05). In secondary analyses, we compared FH prevalence between ancestry groups within individual biobanks and did not find differences (P>0.05). Stratification of individuals with a pathogenic FH variant (n=2375) revealed distinct patterns of genetic architecture between groups. Individuals with African ancestry were more likely to have a pathogenic variant in LDLR (OR, 1.17 [95% CI, 1.03–1.32]; P=0.01) and less likely to have a pathogenic variant in APOB (OR, 0.17 [95% CI, 0.10–0.27]; P<2.2×10^-16) compared with individuals of European ancestry. There was no difference between individuals with African and European ancestry in prevalence of PCSK9 pathogenic variants (OR, 0.35 [95% CI, 0.01–2.30]; P>0.05), and pathogenic variants in PCSK9 were the least common of the 3 genes in both ancestry groups.

Table 2.

Variant Prevalence

LDL Cholesterol Outcomes

We compared the LDL-C levels between individuals with African and European ancestry stratified by the presence of a pathogenic FH variant (Figure 1). Among individuals without any FH variant, LDL-C was only slightly higher (1.66 mg/dL) among the European as compared with the African ancestry group, and this difference was not consistent between biobanks in direction or magnitude (Table S10A). Next, we estimated the effect size of a pathogenic FH variant on LDL-C in African and European ancestry groups. The effect size of a pathogenic variant on LDL-C among individuals of African ancestry was 86.11 (95% CI, 79.18–93.06) mg/dL compared with 65.31 (95% CI, 63.01–67.61) mg/dL among individuals with European ancestry (Figure 2A; Table S10B). A pathogenic FH variant was associated with a 20.81 mg/dL (95% CI, 16.17–25.45) greater increase in individuals with African ancestry (P=2.42×10^-8) compared with European ancestry. Pathogenic variants in the LDLR gene have been observed to cause a more severe phenotype (eg, higher LDL-C levels and ASCVD risk) than pathogenic variants in the APOB and PCSK9 genes.⁹ However, the difference in LDL-C effect size was similar (19.35 mg/dL) when the analysis was restricted to pathogenic variants within the LDLR gene (Figure S4; Table S10C). The difference in the prevalence of pathogenic LDLR loss‑of‑function variants, which are expected to have a larger effect size than pathogenic missense variants, was not statistically significant between the 2 groups (P>0.05). The association between pathogenic FH variants and LDL-C stratified by both sex and ancestry is presented in the Supplemental Material (Table S10D).

Figure 1.

Maximum LDL-C (low-density lipoprotein cholesterol) by ancestry and variant category. MyCode (A), All of Us (B), and BioMe (C) cohorts. AFR indicates African ancestry; EUR, European ancestry; and VUS, variant of unknown significance.

Figure 2.

Effect of pathogenic variants by ancestry group. Effect on (A) size of the increase in maximum LDL-C (low-density lipoprotein cholesterol) and (B) risk of myocardial infarction. Blue symbols are African ancestry, and orange symbols are European ancestry. The association between the presence of a pathogenic variant and LDL-C was calculated with linear regression adjusted for age and sex performed within each cohort and ancestry group. Odds ratios (ORs) for myocardial infarction calculated with logistic regression adjusted for age and sex performed within each cohort and ancestry group. AFR indicates African ancestry; and EUR, European ancestry.

FH and Risk of MI by Ancestry

We observed 36 970 cases of MI among 495 638 individuals of European ancestry (8.2%) and ~4580 cases of MI among 98 362 individuals of African ancestry (~4.7%). After accounting for age and sex, individuals of African ancestry without a variant had 1.37-fold (95% CI, 1.33–1.43) higher risk of MI compared with individuals of European ancestry without a variant; however, this difference was not consistent between biobanks (Table S10E). A pathogenic FH variant was associated with increased risk of MI among individuals with African ancestry (OR, 1.84 [95% CI, 1.20–2.83]; P=0.005) and European ancestry (OR, 2.58 [95% CI, 2.23–2.97]; P=1.50×10^-38) (Figure 2B; Table S10F). The association between a pathogenic FH variant and MI risk was not significantly different between ancestry groups (P>0.05). We did not identify a difference in MI risk by ancestry when the cohort was further stratified by sex (Table S10G).

VUS by Ancestry

We repeated the analyses above with LDLR VUSs to determine their clinical impact and assess gaps in variant classification data in ClinVar between the 2 groups. Individuals with African ancestry had 1.61‑fold higher odds of having a VUS compared with those of European ancestry (95% CI, 1.42–1.83; P=1.2×10^-12). Looking across the 7 VUS categories (Table S1), we saw that the VUS category most common in both ancestry groups was an elevated REVEL score (Table S2), and this category was twice as common among individuals of African ancestry with a VUS (OR, 1.96 [95% CI, 1.48–2.61]; P=1.1×10^-6) compared with individuals of European ancestry with a VUS. Evaluation of VUS annotated with LDLR functional scores from Tabet et al³⁹ also showed a quantitative difference by ancestry: VUS found in individuals with African ancestry had on average a poorer functional score than those found among individuals with European ancestry (P=1.5×10^-11). The presence of a VUS was associated with a 22.57-mg/dL (95% CI, 15.36–29.77) greater LDL-C level compared with individuals without a variant in the African ancestry subgroup (Figure 3A). The effect size of a VUS was 10.01 mg/dL (95% CI, 6.13–13.88; P=0.01) higher in individuals of African ancestry as compared with European ancestry. In addition, among those with a VUS, only individuals with African ancestry had an increased risk of MI compared with those without a variant (OR, 1.92 [95% CI, 1.18–3.10]; P=0.008) (Figure 3B). In European ancestry individuals, having a VUS did not increase risk of MI (OR, 1.20 [95% CI, 0.95–1.52]; P>0.05). The locations and protein domains of VUSs identified in the African ancestry cohort within the LDLR gene are shown in Figure S5.

Figure 3.

Effect of VUSs by ancestry group. Effect on (A) size of the increase in maximum LDL-C (low-density lipoprotein cholesterol) adjusted for lipid-lowering medication use, and (B) risk of myocardial infarction. Blue symbols are African ancestry, and orange symbols are European ancestry. The association between the presence of a VUS and LDL-C calculated with linear regression adjusted for age and sex performed within each cohort and ancestry group. Odds ratios for myocardial infarction calculated with logistic regression adjusted for age and sex performed within each cohort and ancestry group. AFR indicates African ancestry; EUR, European ancestry; OR, odds ratio; and VUS, variant of unknown significance.

Sensitivity Analyses

As our results may be sensitive to methods of estimating LDL-C levels in the untreated state, we repeated all analyses using median LDL-C adjusted for lipid-lowering medications as the phenotype. The primary associations between increased LDL-C and genetic ancestry remained unchanged (Tables S10B and S10H). To further assess the impact of lipid-lowering medications on our analyses, we compared statin intensity across ancestry groups by converting statin type and dose into a Defined Daily Dose. Across the cohorts, male sex and African ancestry were associated with a higher Defined Daily Dose (Tables S11A and S11B). Among individuals without a pathogenic variant or VUS, the Defined Daily Dose was significantly higher in those of African ancestry in all 3 cohorts (Table S11B). In contrast, among individuals with a pathogenic variant or VUS, we observed no difference between the European and African ancestry groups (Table S11C). The higher Defined Daily Dose among African ancestry lipid-lowering medication users without a variant may suggest that we are underestimating the effect size of VUS or pathogenic FH variants relative to the European ancestry group. We also compared PCSK9 inhibitor use between the ancestry groups. In the All of Us cohort, individuals of African ancestry had significantly lower PCSK9 inhibitor use than their European ancestry counterparts in both the variant‑negative and pathogenic variant groups. In contrast, no ancestry‑related differences in PCSK9 inhibitor prescribing were observed in either MyCode or BioMe. When meta-analyzed across the 3 cohorts, the difference between ancestry was not significant (Table S11C). Through these analyses, we were unable to find strong evidence that differences in lipid-lowering medication prescriptions between ancestry groups had a large effect on our results.

Previous studies have shown that individuals of African ancestry are more likely to have large-effect LDL-C–lowering variants in the PCSK9 gene and may be protective for hypercholesterolemia and ASCVD events.⁴⁶ We examined the impact of large effect lipid-lowering variants in the APOB and PCSK9 genes on the association with pathogenic FH variants and clinical outcomes. Across the 3 cohorts, we identified 19 large‑effect lipid‑lowering variants in 2756 individuals of African ancestry and 66 large‑effect lipid‑lowering variants in 14 314 individuals of European ancestry (Table S12). Removing individuals with large effect lipid-lowering variants did not have an effect on our analyses of LDL-C or MI (Tables S13A and S13B).

DISCUSSION

This genome-first study of FH integrates data from 3 US biobanks to characterize the prevalence, genetic contributors, and phenotypic presentation within the African ancestry population. Although the prevalence of pathogenic FH variants was similar between individuals of African and European ancestry in the US population, elevation in LDL-C was more severe among those of African ancestry. An analysis of the genetic architecture of FH among the African ancestry group revealed an enrichment of pathogenic variants in the LDLR gene, which are associated with a more severe phenotype than APOB or PCSK9. This is consistent with previous findings that pathogenic APOB variants are elevated among individuals of European ancestry in the All of Us cohort.¹³ However, the increased prevalence of LDLR variants in the African ancestry group did not fully account for the larger increase in LDL-C relative to European ancestry individuals. Individuals of African ancestry were more likely to have a VUS in the LDLR gene than those of European ancestry. The presence of a VUS among the African ancestry group was associated with an increased risk of FH-related phenotypes including severe hypercholesterolemia and MI, suggesting that VUSs found predominantly in individuals with African ancestry are more likely to be unrecognized FH-causing variants. Supporting this conclusion, we found that VUS detected in individuals of African ancestry were associated with poorer functional scores, based on a recently published sequence-function map of the LDLR gene, than VUS identified in individuals of European ancestry.

The global burden of FH on cardiovascular health is incompletely understood, largely because of limited research beyond Western populations.⁴⁷ This and other factors have led to variant submissions and interpretations in publicly available databases such as ClinVar that provide variant-level classifications and, in some cases, strength of evidence information that is not representative of the variation found in non-European populations.^48,49 The lack of globally representative population data in resources for interpreting genetic variants among non-Europeans is an impediment to accurate molecular diagnoses in these groups. In genetic studies of disease cohorts, including various types of cancer and cardiomyopathies, VUSs in disease-related genes were more prevalent among non-Europeans than their European counterparts.^21,50 In our study, we found that a population-based cohort of African ancestry individuals were at 1.61 higher odds to have a VUS in the LDLR gene compared with individuals of European ancestry.

The presence of a VUS may represent a missed opportunity for precision care, yet the lack of clinical studies of patients with a VUS leaves the scope of that opportunity unclear. In our study of an African ancestry population, VUSs in FH genes were nearly as frequent as pathogenic variants and associated with a significant increase in LDL‑C levels and a higher risk of MI, patterns not observed in the European ancestry population. These findings indicate that relying on current clinical databases likely underestimates FH prevalence in African ancestry populations. It is more important to note that these findings highlight a potential barrier for individuals of African ancestry with monogenic FH in obtaining a molecular diagnosis through genetic testing. Future updates to the Clinical Genome Resource's (ClinGen) Variant Curation Expert Panel guidelines should incorporate large-scale functional analyses of FH gene variants to improve clinical variant interpretation across global populations. The few studies conducted on FH outside the United States and European nations, including Japan and China, have reported pervasive underdiagnosis and undertreatment.^51,52 Similarly, individuals of African ancestry in FH patient registries are both less likely to be prescribed lipid-lowering medication and less likely to achieve LDL-C treatment goals such as an LDL-C level <100 mg/dL compared with individuals of European ancestry.^53,54 In this analysis, we did not detect differences in lipid-lowering medication prescriptions between ancestry groups that would explain the heterogeneity in effect size associated with a pathogenic variant. The observation of a larger effect size of pathogenic FH variants in those of African ancestry requires careful interpretation. A first possibility is a difference of genetic architecture of FH between individuals of European and African ancestry. We did observe a smaller proportion of pathogenic APOB variants in those of African descent. A related hypothesis addresses pathogenic LDLR variants themselves; the distribution of these in individuals with African ancestry may be enriched for more severe variants. This highlights the value of ancestry-specific analyses of monogenic disorders, because differences in genetic architecture across populations may affect disease severity. A second possibility is that there is more pervasive undertreatment of FH among those of African ancestry that was unaccounted for in our estimates of untreated LDL-C. In particular, nonstatin therapies of high potency may be more heavily prescribed in those of European ancestry. A third possibility is that there may also be lower compliance with prescribed statin therapy. These differences in treatment would not be detected by our correction factor for lipid-lowering treatment. Our findings may be suggestive that individuals of African ancestry with a pathogenic FH variant have higher LDL-C levels and may need additional interventions to achieve the management goals.

Limitations

This study does not capture the full diversity of African ancestry populations. For historical reasons, individuals with African ancestry living in the United States are disproportionately of West African ancestry.⁵⁵ Future studies that encompass the entire African continental diversity would be needed to fully capture all FH variants that exist. In addition, the 3 cohorts used in the study, although they are all large US-based biobanks, differ in several respects including ancestry, likelihood of MI, and data completeness. Our analyses are based on estimating LDL-C in the untreated state among patients prescribed lipid-lowering medications. Although we find consistent results across 2 approaches that account for treatment and find no systematic difference in the dose or intensity of statins prescribed between ancestry groups, we are unable to account for group differences in compliance in medication, lipid-lowering medications prescribed outside the primary health system, the effect of nonstatin lipid-lowering medications on LDL-C. FH dramatically increases risk for premature cardiovascular diseases. As population biobanks overrepresent middle-age and older adults, the early mortality associated with FH may contribute to a survival bias that was not accounted for in our analyses. We used a cross‑sectional design to evaluate the association between FH variants and MI. Although pathogenic variants in the African ancestry group were associated with a larger effect size relative to the European ancestry group, we did not observe a commensurate increase in MI risk. Our study measures LDL-C over an average of 6 to 11 years (Table S4) but does not account for cumulative lifetime exposure to LDL-C, which is a key driver of atherosclerotic plaque formation. As a limitation of this retrospective study, treatment effects, environmental exposures, comorbidities, and other sources of unmeasured confounding may contribute to this observed discordance. Prospective studies of individuals of African ancestry that capture untreated LDL-C levels and cumulative lifetime LDL-C exposure are needed to more accurately estimate the risk associated with pathogenic variants in this population. Last, this study did not evaluate the contribution of copy number variants, noncoding variation, or genetic variation outside the 3 canonical FH genes.

CONCLUSIONS

Previous work in biobanks has demonstrated pervasive underdiagnosis and undertreatment of individuals with pathogenic FH variants in diverse populations. In the present study, we show that the higher proportion of individuals with a VUS in the African ancestry population poses a distinct challenge for accurately estimating FH prevalence in this group. Furthermore, the elevated risk of hypercholesterolemia and MI observed among African ancestry individuals with VUS suggests that gaps in clinical variant databases, an essential resource for variant interpretation, may contribute to disparities in FH recognition and management. Further research is required to elucidate global variation in the genetic architecture of FH and to determine its contribution to cardiovascular disease burden in populations that remain underrepresented in genetic studies.

ARTICLE INFORMATION

Acknowledgments

We thank Andrea Hattenberger for her assistance with obtaining the phenotype data at Geisinger. We acknowledge the All of Us Research Program and its participant partners, whose contributions are essential to this research. This program would not be possible without their partnership. We thank Dr. Kathleen Ferar for her assistance with obtaining data from the Mount Sinai Data Warehouse and Mount Sinai’s BioMe biobank. We thank Drs. Eimear Kenny and Chelsea Lowther from Mount Sinai’s Institute for Genomic Health for valuable feedback.

Disclosures

S.S.G. is a consultant for Esperion Therapeutics. L.K.J. is an employee of Amgen and holds company stock. The other authors report no conflicts.

Supplemental Material

Tables S1–S13

Figures S1–S5

Supplementary Material

cir-153-1928-s001.pdf ^{(6MB, pdf)}

cir-153-1928-s002.pdf ^{(213.7KB, pdf)}

Funding Statement

A.H.W., M.A.K., A.S.F.B., D.C., V.P., S.S.G., and M.T.O. were supported in part by National Health, Lung, and Blood Institute grant No. R01HL159182. M.G.S., T.B., and V.P. were also supported in part by National Institutes of Health/National Human Genome Research Institute grant No. R01HG013350. This work was supported in part through the Minerva computational and data resources and staff expertise provided by Scientific Computing and Data at the Icahn School of Medicine at Mount Sinai and supported by Clinical and Translational Science Awards grant No. UL1TR004419 from the National Center for Advancing Translational Sciences. Research reported in this publication was also supported by the Office of Research Infrastructure of the National Institutes of Health under award Nos. S10OD026880 and S10OD030463. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. The All of Us Research Program is supported by the following grants: National Institutes of Health, Office of the Director: Regional Medical Centers: 1 OT2 OD026549; 1 OT2 OD026554; 1 OT2 OD026557; 1 OT2 OD026556; 1 OT2 OD026550; 1 OT2 OD 026552; 1 OT2 OD026553; 1 OT2 OD026548; 1 OT2 OD026551; 1 OT2 OD026555; IAA #: AOD 16037; Federally Qualified Health Centers: HHSN 263201600085U; Data and Research Center: 5 U2C OD023196; Biobank: 1 U24 OD023121; The Participant Center: U24 OD023176; Participant Technology Systems Center: 1 U24 OD023163; Communications and Engagement: 3 OT2 OD023205; 3 OT2 OD023206; and Community Partners: 1 OT2 OD025277; 3 OT2 OD025315; 1 OT2 OD025337; 1 OT2 OD025276.