Inter-chromosonal Differences in Fraction of CpG Dinucleotides under Stabilizing Selection in Human Genome

Next generation sequencing of human parent-offspring genomes has provided an increased amount of data on de novo mutation rates in several populations. These mutation rates differ from those derived from previous interspecies genome comparisons, especially for highly mutable CpG dinucleotides. Such substitution data predicts a much slower genome-wide decline of CpG sites than suggested from de novo mutation data. One explanation is that CpG dinucleotides are preserved by negative selection. We attempted to estimate the fraction of CpG sites experiencing negative selection in the human genome using three datasets of de novo mutation data and a Markov Chain substitution model based on genomic data of three species: Homo sapiens, Pan troglodytes and Pan paniscus. Ourproposedmathematicalevolutionarymodelinputsgenomicalignmentfragmentsofclosely related species and estimates the most likely context-dependent point substitution rates that would produce the observed statistics of column variation. Importantly, this model takes multiple substitutions of the same site into account. Our model outputs the evolutionary lengths of phylogenetic tree branches and the probability rate of 256 types of point substitutions (we use three nucleotide contexts, with the middle nucleotide subject to three possible mutations and one non-mutation). These substitutions rates are then compared to observed de novo mutation rates. Basedonthiscomparisonweestimatethatthefractionofsitesexperiencingnegativeselection issignificantlylargerforCpGthannon-CpGsites. Thisexcessis21.2%(95%COI18%to24.3%), 23.7% (95% COI 20.9% to 26.5%), and 25.7 (95% COI 17.1% to 34.3%) for Iceland, Yuen and GONL de novo mutation datasets. Given the recent estimate that 8.2% (95% COI 7.1 to 9.2%) of the human genome being constrained, we conclude that at least 25% CpG sites are under negative selection.