Evidence for strong fixation bias at 4-fold degenerate sites across genes in the great tit genome

Abstract

It is well established that GC content varies across the genome in many species and that GC biased gene conversion, one form of meiotic recombination, is likely to contribute to this heterogeneity. Bird genomes provide an extraordinary system to study the impact of GC biased gene conversion owed to their specific genomic features. They are characterised by a high karyotype conservation with substantial heterogeneity in chromosome sizes, with up to a dozen large macrochromosomes and many smaller microchromosomes common across all bird species. This heterogeneity in chromosome morphology is also reflected by other genomic features, such as smaller chromosomes being gene denser, more compact and more GC rich relative to their macrochromosomal counterparts - illustrating that the intensity of GC biased gene conversion varies across the genome. Here we study whether it is possible to infer heterogeneity in GC biased gene conversion rates across the genome using a recently published method that accounts for GC biased gene conversion when estimating branch lengths in a phylogenetic context. To infer the strength of GC biased gene conversion we contrast branch length estimates across the genome both taking and not taking non-stationary GC composition into account. Using simulations we show that this approach works well when GC fixation bias is strong and note that the number of substitutions along a branch is consistently overestimated when GC biased gene conversion is not accounted for. We use this predictable feature to infer the strength of GC dynamics across the great tit genome by applying our new test statistic to data at 4-fold degenerate sites from three bird species - great tit, zebra finch and chicken - three species that are among the best annotated bird genomes to date. We show that using a simple one-dimensional binning we fail to capture a signal of fixation bias as observed in our simulations. However, using a multidimensional binning strategy, we find evidence for heterogeneity in the strength of fixation bias, including AT fixation bias. This highlights the difficulties when combining sequence data across different regions in the genome.