GC Content and Recombination: Reassessing the Causal Effects for the Saccharomyces cerevisiae Genome

Recombination plays a crucial role in the evolution of genomes. Among many chromosomal features, GC content is one of the most prominent variables that appear to be highly correlated with recombination. However, it is not yet clear (1) whether recombination drives GC content (as proposed, for example, in the biased gene conversion model) or the converse and (2) what are the length scales for mutual influences between GC content and recombination. Here we have reassessed these questions for the model genome Saccharomyces cerevisiae, for which the most refined recombination data are available. First, we confirmed a strong correlation between recombination rate and GC content at local scales (a few kilobases). Second, on the basis of alignments between S. cerevisiae, S. paradoxus, and S. mikatae sequences, we showed that the inferred AT/GC substitution patterns are not correlated with recombination, indicating that GC content is not driven by recombination in yeast. These results thus suggest that, in S. cerevisiae, recombination is determined either by the GC content or by a third parameter, also affecting the GC content. Third, we observed long-range correlations between GC and recombination for chromosome III (for which such correlations were reported experimentally and were the model for many structural studies). However, similar correlations were not detected in the other chromosomes, restraining thus the generality of the phenomenon. These results pave the way for further analyses aimed at the detailed untangling of drives involved in the evolutionary shaping of the yeast genome.THE architecture of genomes is the result of various evolutionary forces, which can exert concerted or opposing effects. Recombination is considered to represent one such fundamental drive. Indeed, correlations with recombination were reported for a large number of structural or functional properties, such as the length of genes, the length of introns for split genes (Comeron and Kreitman 2000; Prachumwat et al. 2004), or even gene order, with the clustering of essential genes in regions of low recombination (Pal and Hurst 2003). GC content represents perhaps the most prominent property for which strong correlations with recombination were reported for the genomes of many organisms including mammals, Drosophila melanogaster, Caenorhabditis elegans, and Saccharomyces cerevisiae (Gerton et al. 2000; Marais et al. 2001; Birdsell 2002; Kong et al. 2002; Meunier and Duret 2004). On the other hand it was recently demonstrated that in Arabidopsis thaliana rate of crossover and GC content are not correlated (Drouaud et al. 2006). However, despite these numerous results, it is not clear as yet (1) whether recombination drives GC content or the converse and (2) what are the length scales for the correlations between GC and recombination.Correlations between recombination and GC content have been detected both at local scales [typically in the kilobase range (see Gerton et al. 2000)] and at much larger ones (Kong et al. 2002). Arguments were advanced in favor of context-dependent recombinational activities, with the idea that such activities could be regulated, at least in part, by global features of chromosome structure, characterized more or less directly by the GC content (for a general overview, see, for example, Eyre-Walker and Hurst 2001). In this direction, in terms of evolutionary models, mutual influences between recombination and GC were even considered at the highest organizational levels, with the proposal that the large-scale organization of mammalian genomes in terms of GC-rich isochores could be accounted for to a large extent by the integral of past recombinational activities (Duret et al. 2006; Duret and Arndt 2008).Regarding the causality relationship between recombination and GC, the biased gene conversion model (see Eyre-Walker 1993 for original formulations) proposes that recombination represents a driving force for GC variations, from local to genomewide scales (in terms of isochore structures). In this model a basic role is attributed to allelic gene conversions during meiotic recombination, as a consequence of the repair of mismatches in heteroduplex DNA. This process is supposed to be biased toward GC, leading to an increase of overall GC contents in regions with high recombination activity (Brown and Jiricny 1989; Eyre-Walker 1993; Galtier et al. 2001; Marais et al. 2001; Birdsell 2002). On the contrary, with analyses mainly based on the Saccharomyces cerevisiae genome, the supporters of the opposite causality model have suggested that it is rather high GC content that promotes recombination (Gerton et al. 2000; Petes 2001; Blat et al. 2002; Petes and Merker 2002).In this general background we here reassess various questions concerning the relationships between recombination and GC for the S. cerevisiae model system. Surprisingly, whereas S. cerevisiae has served as the system of choice for many of the original questions and models concerning recombination, it appears that various questions, debated notably in the context of mammalian genomes, were not further put to test in the S. cerevisiae genome for which the most accurate recombination data of any system have become recently available (Blitzblau et al. 2007; Buhler et al. 2007; Mancera et al. 2008).We first addressed the causality question at local scales, using the same approach as the one that was implemented in the case of mammalian genomes. At such scales, with the new recombination data for S. cerevisiae, we confirmed the strong correlations between GC and recombination. We then analyzed the patterns of substitutions that occurred in the S. cerevisiae strain S288C lineage under two evolutionary perspectives: (1) after the divergence between the S288C lineage and the lineage of another strain of S. cerevisiae, YJM789, and (2) after the divergence between the S. cerevisiae and the S. paradoxus lineages. The rationale behind such substitution analyses (Meunier and Duret 2004; Webster et al. 2005; Khelifi et al. 2006; Duret and Arndt 2008) is to address the possible effect of recombination on GC content, through the determination of the relative rates of AT to GC and GC to AT substitutions. On the basis of such analyses, we found that recombination is not directly correlated to the patterns of AT/GC substitutions in S. cerevisiae, which indicates that recombination has no detectable influence on GC content in this case.Beyond the local scales, we then considered the ranges of mutual influences between recombination and GC content in S. cerevisiae. We first extended the substitution analyses at significantly larger scales, to test the possibility that the local result could hide long-range correlations. Indeed, results demonstrating the effect of recombination on GC content in the human genome could be observed only at the megabase scale (Duret and Arndt 2008). In S. cerevisiae, however, we found no evidence for a significant effect of recombination on GC content at any scale. Concerning the large-scale influences, we tested then a model developed by Petes and Merker (2002), following which, in S. cerevisiae, recombinational activity at one given locus could be determined by the GC content of the surrounding region, over large distances. This model was elaborated on the basis of the analysis of chromosome III, but our results did not allow us to validate the generality of the hypothesis for all S. cerevisiae chromosomes.