Decoupling of allele frequency and predicted pathogenicity in the tandem repeat filaggrin (FLG) gene

Abstract

Tandemly repetitive genes present a paradox in that their duplicated architecture may buffer against mutation yet often exhibits strong position-dependent functional constraints. The filaggrin gene ( FLG ), with its massive repetitive core, provides a key model to examine how repeat architecture and demographic forces shape variation in such recalcitrant genomic regions. We integrated population-scale data from gnomAD ( n  > 140,000) with deep learning predictions from AlphaMissense to map FLG ’s variant landscape. Constraint metrics indicated relaxed selection (missense o/e = 1.63), yet a clear spatial discordance emerged: predicted pathogenic variants were enriched in terminal repeats, whereas observed allele counts in human populations were concentrated in a few recurrent truncating variants in central repeats. These prevalent alleles exhibited strong alignment with population ancestry, while common variants (AF > 0.01) lacked definitive pathogenic annotations, highlighting the primacy of demographic history over uniform selection. Amino acid composition across repeats revealed significant divergence, with biases in Ser, Gly, and His correlating with predicted pathogenicity, providing a mechanistic basis for repeat-specific evolutionary constraints. Together, these findings demonstrate that variant interpretation in large tandem repeats requires explicit integration of repeat structure and population history. This framework has implications for ancestry-aware assessment of FLG variation and for the broader analysis of repetitive genes in human disease genetics.