Abstract
The human salivary amylase genes display extensive copy number variation (CNV), and recent work has implicated this variation in adaptation to starch-rich diets, and in association with body mass index. In this work, we use paralogue ratio tests, microsatellite analysis, read depth and fibre-FISH to demonstrate that human amylase CNV is not a smooth continuum, but is instead partitioned into distinct haplotype classes. There is a fundamental structural distinction between haplotypes containing odd or even numbers of AMY1 gene units, in turn coupled to CNV in pancreatic amylase genes AMY2A and AMY2B. Most haplotypes have one copy each of AMY2A and AMY2B and contain an odd number of copies of AMY1; consequently, most individuals have an even total number of AMY1. In contrast, haplotypes carrying an even number of AMY1 genes have rearrangements leading to CNVs of AMY2A/AMY2B. Read-depth and experimental data show that different populations harbour different proportions of these basic haplotype classes. In Europeans, the copy numbers of AMY1 and AMY2A are correlated, so that phenotypic associations caused by variation in pancreatic amylase copy number could be detected indirectly as weak association with AMY1 copy number. We show that the quantitative polymerase chain reaction (qPCR) assay previously applied to the high-throughput measurement of AMY1 copy number is less accurate than the measures we use and that qPCR data in other studies have been further compromised by systematic miscalibration. Our results uncover new patterns in human amylase variation and imply a potential role for AMY2 CNV in functional associations.