Using a Read-Based Approach to Estimate the Accumulation of Deleterious Variation in Domestic and Wild Canids from Low-Coverage Sequencing Data

Whilst many bioinformatic tools and pipelines have been developed for variant calling, such approaches typically rely on databases of known variant sites and/or high coverage data, which is often not available in non-model taxa. Moreover, if more than one taxon is being compared, biases can be introduced if there is variation in the numbers of samples sequenced (e.g. with joint calling), sequencing depth and/or sample sizes between groups.  Here we introduce a read-based method that estimates average heterozygosity per individual directly from the aligned sequencing reads.  Unlike SNP calling, the read based approach avoids the introduction of systematic biases, is less prone to the effects of differences in coverage across samples, and does not rely on external databases of known variants.  We apply this method to a set of 131 canid genomes in order to investigate the role of population history in shaping the patterns of deleterious mutations in domestic dogs and their wild relatives.  Overall, our results indicate that as long as the quality of sequencing is high (Q>30), read-based estimates of heterozygosity are accurate for both low (2-10x) and higher coverage genomes.  Based on these data, we show that the ratio of amino-acid changing to silent variants is negatively correlated with neutral genetic diversity, suggesting selection is weaker in small populations. Moreover, we show that breed dogs which have undergone severe bottlenecks and inbreeding have the highest ratio of amino-acid to silent variants, whereas wolves have the lowest and semi-feral village dogs intermediate.