A High-Throughput Programmatic Pipeline for Computational Reclassification of Variants of Uncertain Significance in LDLR: Structural Evidence for Disruption of the LDL-Receptor Class A Ligand-Binding Repeats

Abstract

Familial hypercholesterolemia (FH) is a common autosomal-dominant disorder of lipoprotein metabolism, most frequently caused by loss-of-function variants in the low-density lipoprotein receptor gene (LDLR). Clinical genetic testing for FH is increasingly limited not by sequencing throughput but by interpretation throughput: a large fraction of LDLR variants deposited in ClinVar are classified as Variants of Uncertain Significance (VUS), which cannot be used to guide diagnosis or cascade screening. This study presents a fully programmatic, command-line pipeline that reclassifies LDLR VUS by integrating population frequency, dual in silico deleteriousness prediction, and protein structural context. ClinVar GRCh38 records were filtered with bcftools, using a delimiter-anchored gene-symbol match, to isolate 1,187 LDLR-annotated variants of uncertain significance on chromosome 19; 1,101 of these were valid single-nucleotide variants and were queried against the Ensembl REST API (VEP) for SIFT, PolyPhen-2, and gnomAD allele-frequency estimates. Variants were retained as high-priority candidates if they were rare (gnomAD AF < 1×10⁻⁴), predicted deleterious by SIFT (< 0.05), and predicted probably damaging by PolyPhen-2 (> 0.90); 222 variants (20.2% of annotated SNVs) met all three thresholds. Protein coordinates were mapped onto the full UniProt canonical precursor sequence (P01130, 860 residues, signal peptide 1–21) and cross-referenced against curated UniProt domain boundaries. The five highest-confidence candidates all localize to LDL-receptor class A (LA) repeats 3–5 of the N-terminal ligand-binding domain, and four of the five (p.Cys132Trp, p.Cys149Tyr, p.Cys161Gly, p.Cys214Ser) directly substitute one of the six invariant cysteines that form the disulfide-bonded, calcium-coordinating fold of an LA repeat. These results provide a testable, structurally grounded hypothesis for pathogenicity in a substantial subset of currently unresolved LDLR VUS and demonstrate a reproducible, freely available computational workflow that could be extended to other Mendelian disease genes. As with all in silico approaches, these findings constitute prioritization evidence rather than clinical proof and require orthogonal functional or segregation data before use in diagnostic reclassification.