Genotyping sequence-resolved copy-number variations using pangenomes reveals paralog-specific global diversity and expression divergence of duplicated genes

Abstract

Genetic analysis of copy number variations (CNVs), especially in complex regions, is challenging due to reference bias and ambiguous alignment of Next-Generation Sequencing (NGS) reads to repetitive DNA. Consequently, aggregate copy numbers are typically analyzed, overlooking variation between gene copies.Pangenomes contain diverse sequences of gene copies and enable the study of sequence-resolved CNVs. We developed a method, ctyper, to discover sequence-resolved CNVs in NGS data by leveraging CNV genes from pangenomes. From 118 public assemblies, we constructed a database of 3,351 CNV genes, distinguishing each gene copy as a resolved allele. We used phylogenetic trees to organize alleles into highly similar allele-types that revealed events of linked small variants due to stratification, structural variation, conversion, and duplication. Saturation analysis showed that new samples share an average of 97.8% CNV alleles with the database. The ctyper method traces individual gene copies in NGS data to their nearest alleles in the database and identifies allele-specific copy numbers using multivariate linear regression on k-mer counts and phylogenetic clustering. Applying ctyper to 1000 Genomes Project (1kgp) samples showed Hardy-Weinberg Equilibrium on 99.3% of alleles and a 97.6% F1 score on genotypes based on 641 1kgp trios. Leave-one-out analysis on 39 assemblies matched to 1kgp samples showed that 96.5% of variants in query sequences match the genotyped allele. Genotyping 1kgp data revealed 226 population-specific CNVs, including a conversion on SMN2 to SMN1, potentially impacting Spinal Muscular Atrophy diagnosis in Africans.Our results revealed two models of CNV: recent CNVs due to ongoing duplications and polymorphic CNVs from ancient paralogs missing from the reference. To measure the functional impact of CNVs, after merging allele-types, we conducted genome-wide Quantitative Trait Locus analysis on 451 1kgp samples with Geuvadis rRNA-seqs. Using a linear mixed model, our genotyping enables the inference of relative expression levels of paralogs within a gene family. In a global evolutionary context, 150 out of 1,890 paralogs (7.94%) and 546 out of 16,628 orthologs (3.28%) had significantly different expression levels, suggesting divergent expression from original genes. Specific examples include lower expression on the convertedSMNand increased expression on translocatedAMY2B(GTEx pancreas data). Our method enables large cohort studies on complex CNVs to uncover hidden health impacts and overcome reference bias.