The length of haplotype blocks and signals of structural variation in reconstructed genealogies

Recent breakthroughs have enabled the accurate inference of large-scale genealogies. Through modelling the impact of recombination on the correlation structure between genealogical local trees, we evaluate how this structure is reconstructed by leading approaches. Despite identifying pervasive biases, we show that applying a simple correction recovers the desired distributions for one algorithm, Relate. We develop a statistical test to identify clades spanning unexpectedly long genomic regions, likely reflecting regional suppression of recombination in some individuals. Our approach allows a systematic scan for inter-individual recombination rate variation at an intermediate scale, between genome-wide differences and individual hotspots. Using genealogies reconstructed with Relate for 2 504 human genomes, we identify 50 regions possessing clades with unexpectedly long genomic spans (p < 1 {middle dot} 10-12). The strongest signal corresponds to a known inversion on chromosome 17. The second strongest uncovers a novel 760kb inversion on chromosome 10, common (21%) in S. Asians and correlated with GWAS hits for a range of phenotypes. Other regions indicate additional genomic rearrangements: inversions (8), copy number changes (2), or other variants (12). The remaining regions appear to reflect recombination suppression by previously unevidenced mechanisms. They are enriched for precisely spanning single genes (p = 5 10-10), specifically those expressed in male gametogenesis, and for eQTLs (p = 2 {middle dot}10-3). This suggests an extension of previously hypothesised crossover suppression within meiotic genes, towards a model of suppression varying across individuals with different expression levels. Our methods can be readily applied to other species, showing that genealogies offer previously un-tapped potential to study structural variation and other phenomena impacting evolution.