Validated Polygenic Risk Scores Complement Monogenic Testing in South Asian Patients With Familial Hypercholesterolemia

Familial hypercholesterolemia (FH) is a hereditary disorder characterized by elevated low-density lipoprotein cholesterol (LDL-C) levels and increased risk of premature atherosclerotic cardiovascular disease typically caused by variants in LDLR, APOB, and PCSK9 genes.¹ However, a substantial proportion of clinically diagnosed individuals lack these variants and have polygenic hypercholesterolemia driven by multiple common variants, quantified as polygenic risk scores (PRS) derived from genome-wide association studies.² Although PRS can predict LDL-C levels and disease risk, most models are derived from European populations, limiting generalizability. We validated 5 low-density lipoprotein–polygenic risk score (LDL-PRS) models in South Asian FH samples, with/without pathogenic (P) and likely pathogenic (LP) variants, with additional evaluation in an independent set of 27 cases.

Methods

We recruited 151 unrelated participants with clinically diagnosed FH (Department of Endocrinology, AIMS, Cochin) before starting lipid-lowering therapy with fasting LDL-C ≥190 mg/dL; Dutch Lipid Clinic Network score ≥3 after excluding secondary causes. The study was approved by the institutional ethics committee (IRB-AIMS-2020 to 331) and informed consent was obtained. An independent set of 27 FH cases was used for validation. Control samples (n = 207) were derived from a published South Asian cohort with LDL-C <100 mg/dL without history of coronary artery disease and lipid-lowering therapy.³

Targeted exome sequencing of 151 samples was performed using a 23-gene FH panel with libraries sequenced on the Illumina HiSeqX platform (150 base pair paired-end reads; 80-100x coverage). The variants were called using the Sentieon tool and classified per American College of Medical Genetics and Genomics criteria. The pathogenicity was also assessed using ClinGen Familial Hypercholesterolemia Variant Curation Expert Panel recommendations. Genotyping was performed using Global Screening Array v3 with the Illumina iScan system, retaining samples with genotyping rate >95% and excluded related samples (Pi_Hat cutoff of 0.125) using PLINK tool.

QC-passed markers were imputed using Beagle v5.0 with the GenomeAsia Phase 2 reference panel.^4,5 Marker weights for 5 PRS models (PGS000814, PGS003405, PGS000115, PGS004783, and PGS004784) were obtained from the PGS Catalog database.⁶ Among these, PGS004783 and PGS004784 were integrated PRS, namely PRSmix and PRSmix+, respectively.⁷ The PRSmix combines all the available PRS for hypercholesterolemia while PRSmix+ combines all available PRS models in cardiometabolic diseases including hypercholesterolemia.⁷ As these models were predominantly derived from European populations, the raw PRSs were normalized for ancestry using 500 South Asian samples by principal component analysis.

Age- and sex-adjusted PRSs were derived using generalized linear models in R stats package to address for differences between case and controls. Pearson's correlation between PRS and measured LDL-C and total cholesterol were assessed and ORs were estimated across PRS quintiles (Q1-Q5), with Q3 as reference. Student's t-test was used to evaluate the differences in LDL-PRS across variant groups. The area under the curve (AUC)-receiver operating characteristic curve was calculated, roc_auc_score function in scikit-learn 1.8.0 Python package. The variance explained by each PRS model was estimated using Nagelkerke R².

Results

Genotyping and targeted exome data were generated for 151 FH cases with ancestry-matched controls (n = 207) from a published cohort. The diagnostic yield was 12% (18 of 151), identifying 18 cases with 11 P/LP variants mainly in LDLR (P = 9, LP = 1) and 1 in the ABCG8 gene. In addition, 22 unique variants of uncertain significance (VUS) were identified in 29 of 151 cases, most frequently in LDLR (n = 7) and APOB (n = 5), with fewer variants in ABCG5, PPP1R17, SREBF2, ABCG8, APOE, PCSK9, and LDLRAP1.

Across all 5 LDL-PRS models, cases had higher median PRS than controls, with variant-negative individuals having the highest scores. Integrated models (PGS004783 and PGS004784) demonstrated the highest risk stratification, with OR of 2.78 (95% CI: 1.5-5.2) and 2.9 (95% CI: 1.54-5.48), respectively, for the highest quintile, compared with 1.75 (95% CI: 0.95-3.23)-2.68 (95% CI: 1.44-4.97) for the remaining models (Figure 1A). A large proportion of cases clustered in higher PRS quintiles, particularly in integrated models (eg, 74.1% [112 of 151] and 72.8% [110 of 151] in Q4-Q5 for PGS004783 and PGS004784), with a consistent increase in median PRS across quintiles.

Figure 1.

Validation, Distribution and Predictive Performance of LDL-PRS

(A) Validation of low-density lipoprotein–polygenic risk scores (LDL-PRS) across 5 PRS models. Samples were stratified into quintiles (Q1-Q5) for each model (PGS004784, red; PGS004783, orange; PGS000814, blue; PGS003405, green; PGS000115, violet). The x-axis represents ORs with 95% CIs, and the y-axis shows polygenic risk score (PRS) models and quintile bins. The dotted vertical line indicates the reference (OR: 1) corresponding to the middle quintile (Q3). (B) Distribution of LDL-PRS across genetically defined familial hypercholesterolemia (FH) groups and controls. The x-axis represents controls and FH subgroups stratified by variant status (pathogenic [P]/likely pathogenic [LP], variants of uncertain significance [VUS], none), and the y-axis shows median LDL-PRS values. (C) LDL-PRS across quintiles and association with measured cholesterol levels. Median low-density lipoprotein cholesterol (LDL-C) (orange) and total cholesterol (TC) (green) are shown across PRS quintiles (Q1-Q5). Pearson's correlation coefficients with 95% CIs between PRS and lipid levels are provided in the inset. The x-axis represents PRS quintiles and the y-axis indicates cholesterol levels (mg/dL). (D) Predictive performance of LDL-PRS models. Receiver operating characteristic curves for the 5 PRS models are shown alongside the random classifier (p.val <0.01 for all comparisons with random classifier). The x-axis represents the false positive rate, and the y-axis represents the true positive rate. (E) Variance explained by LDL-PRS. Nagelkerke's R² values for LDL-C (red) and total cholesterol (blue) across the 5 PRS models. The x-axis represents PRS models and the y-axis indicates R² values. AUC = area under the curve.

The highest LDL-PRS was found in variant-negative cases (0.56 [±0.66]; P = 2.2 × 10⁻¹⁶), intermediate in VUS (0.29 [±0.73]; P = 9.76 × 10⁻⁴) followed by control group (−0.24 [±0.83]) (Figure 1B). Notably, variant-positive cases had LDL-PRS lower than controls (−0.35 [±0.74]).

Median LDL-C and total cholesterol increased from Q1 to Q5 across all models (Figure 1C), with modest correlations between PRS and cholesterol levels, among which the highest correlation value of 0.321 (95% CI: 0.22-0.41) was observed between PRS from PGS004783 model and LDL-C. However, when cases and controls were analyzed separately, no significant differences in LDL-C were observed across quintiles likely reflecting the higher LDL-C thresholds in this cohort.

Predictive performance was highest for PGS004783 (AUC = 0.718; 95% CI: 0.66-0.77), followed by PGS004784 (AUC = 0.706; 95% CI: 0.64-0.76), whereas other models showed modest performance, ranging from AUC values of 0.644 (95% CI: 0.58-0.69) to 0.676 (95% CI: 0.61-0.73) (Figure 1D). PGS004783 (PRS mix) also demonstrated moderate explanatory power (Nagelkerke's R² > 0.2) for LDL-C and total cholesterol (Figure 1E).

Discussion

In this study, we validated 5 PRS models, including integrated models, in a South Asian FH cohort. Variant-negative individuals had higher LDL-PRS than controls and variant-positive cases, supporting the role of polygenic burden in conferring risk comparable to monogenic mutations.⁸ Although PRS integration is gaining attention; guidelines emphasize population-specific validation,⁹ particularly for South Asian individuals who face elevated cardiometabolic risk. Recent studies demonstrate that high PRS levels can increase cardiovascular risk to levels seen in severe hypercholesterolemia.^10,11

All PRS models showed modest correlation with LDL-C and total cholesterol; however, integrated PRS models (PRSmix and PRSmix+) outperformed individual models. Consistent with previous studies, integrated PRS models enhance prediction accuracy and cross-ancestry performance,⁷ helping to mitigate the limited transferability of PRS derived from the European population. These findings were validated in an independent cohort, in which most cases clustered in higher PRS quintiles, supporting the clinical utility of PRS for diagnosis and risk stratification, especially among variant-negative FH.

Study limitations

The major limitation of this study is that it was conducted at a single center with a relatively small sample size.

Conclusions

We validated 5 PRS models for FH in a South Asian population. Variant-negative individuals exhibited the highest LDL-PRS, underscoring the role of polygenic burden. Integrated PRS models, particularly PRSmix, exhibited the best discriminative performance, supporting the integration of PRS alongside monogenic testing, particularly when no mutation is identified.

Funding Support and Author Disclosures

The study was partially funded by the Amrita School of Medicine, Kochi, Amrita Vishwa Vidyapeetham, India. Dr Ramesh Menon, Akshi Bassi, Dr Anjanappa, Praveena L. Samson, Ramprasad Vedam L., Dr Gupta were employed and/or have equity in MedGenome Labs Ltd. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.