Structural, Enzymatic and Population-Genetic Profiling of CYP17A1 Variants Defines the Molecular Basis of Partial 17α-Hydroxylase/17,20-Lyase Deficiency

Context: 17α-hydroxylase/17,20-lyase deficiency (17-OHD) is a rare form of congenital adrenal hyperplasia that presents with hypertension, hypokalemia and absent or incomplete sexual development. The clinical spectrum extends from complete deficiency to partial forms compatible with spontaneous puberty and fertility, and the molecular determinants of this heterogeneity remain incompletely defined. Objectives: To define the pathophysiology of 17-OHD by combining clinical phenotyping with functional, structural, computational and population-genetic characterization of seven CYP17A1 missense variants: T131P, L209P, I296T, G303D, R340H, L361P and R449H. Methods: Clinical and hormonal data were collated for ten probands. Variants were expressed in HEK293T cells and 17α-hydroxylase and 17,20-lyase activities were measured with [4-14C]-progesterone and [21-3H]-17α-hydroxypregnenolone and normalized to immunoreactive CYP17A1 abundance. Residue environments were measured directly in the experimental structures PDB 3RUK, 4NKX and 4NKZ, and model confidence was taken from AlphaFold. Structural Flexibilities were studied by molecular dynamics simulations and analyzed by principal component analysis and clustering. Variants were annotated with Ensembl VEP, SpliceAI, CADD and dbNSFP predictors, scored with AlphaGenome using adrenal-gland splicing tracks, and placed in population genetic context using gnomAD v4.1, the 1000 Genomes Project and a gene-wide ClinVar survey. Sequence constraint was assessed with a masked protein language model and with 172 to 176 vertebrate orthologues. Results: All seven variants retained substantial immunoreactive protein, but below wild type, and the set-level reduction was significant in all assays. Residual 17α-hydroxylase activity ranged from approximately 1% of wild type for L209P to approximately 50% for R340H, and residual 17,20-lyase activity exceeded hydroxylase activity for every variant. None of the seven variants altered splicing by SpliceAI or AlphaGenome, confirming a purely protein-level mechanism. AlphaMissense was the only predictor whose scores agreed with measured residual activity, yet its binary classification called I296T, R340H and R449H benign despite measured deficits and ClinVar pathogenic or likely pathogenic submissions, and CADD ranked the series in the opposite order to severity. Across gnomAD v4.1 the cumulative burden of pathogenic and likely pathogenic CYP17A1 alleles predicted a carrier frequency of 1 in 1,023 globally and 1 in 523 in East Asian individuals. Serum progesterone was raised in all 10 probands but did not distinguish partial from complete deficiency, whereas morning cortisol did. Conclusion: The pathogenicity of these CYP17A1 variants reflects specific catalytic and regulatory defects in a largely intact protein rather than loss of steady-state enzyme. Residual activity relates to clinical severity non-linearly, and current missense predictors are not calibrated for the partial-loss alleles that dominate the clinically ambiguous end of the spectrum. Direct measurement of residual activity remains necessary for prognosis and reproductive counselling.