Allele frequency change is consistent with asymmetric mitonuclear incompatibilities
By measuring allelic frequency change from the initial F2 hybrids to the replicated hybrid lines that evolved under each mitochondrial background, we can scan the genome for regions likely under selection, which should be consistent across replicates, from regions under genetic drift, which should show no or inconsistent change across replicates. Although the choice of sequenced lines is debatable, we assume that the three chosen replicates were adequate for testing our hypothesis because they all show increased in population size, and survivorship has recovered to parental fitness for multiple generations (Fig. 2D). In agreement with selection, we found some consistency among replicate populations of each reciprocal cross (Fig. S2). This was particularly pronounced on the SC lines, where we found more extreme P-values rejecting the null hypothesis of genetic drift (Fig. 3); also reflected by the higher lineage-specific thresholds based on a z-score of 2 (-log10 (P-value) of 4.17 for SC and 1.73 for SD). Together with our observation that F2 breakdown in survivorship is stronger in the SC background (Fig. 2C), this suggests that the strength of selection acting against mitonuclear incompatibilities was stronger in this mitochondrial background relative to SD. Such asymmetries in the strength of mitonuclear incompatibilities is also supported by reciprocal crosses of multiple populations of T. californicus(Ganz & Burton 1995; Peterson et al. 2013).
By comparing the genomic regions responding to selection in lines evolving under alternative mitochondrial backgrounds, we can distinguish regions responding to uniform selection (i.e. where the same nuclear allele dominates), from regions responding to divergent selection (i.e. where different alleles dominate). Consistent with uniform selection in 6.5% of the genome (Table 1), we find that the SC allele dominates in much of chromosome 7 and a small segment of chromosome 8. Uniform selection could be explained by BDMIs between 2 or more nuclear factors (nuclear-nuclear instead of mitonuclear incompatibilities), purging of slightly deleterious alleles that have been fixed in one of the parental populations, or by adaptation to laboratorial conditions favoring alleles of one of the populations. These possibilities are also not mutually exclusive, and previous work in this species lend support to any of these. Studies with reciprocal F2 hybrids from multiple population crosses have found skews towards the same allele in one or two chromosomes (Foley et al. 2013; Lima et al. 2019), suggesting a modest role of nuclear-nuclear incompatibilities inT. californicus . Previous genomic studies of natural populations found that effective population size in SD is less than half that of SC (Pereira et al. 2016), and thus this population is expected to accumulate deleterious mutations twice as fast. Moreover, experimental studies have revealed that these populations are differentially adapted to temperature (Willett 2010; Hong & Shurin 2015; Pereira et al.2017) and to salinity (Leong et al. 2017), and that rearing temperature used in this study favors SC alleles, relative to SD (Willett 2010). While these three selective regimes are impossible to distinguish under our experimental design, future evolve-and-resequencing experiments varying environmental factors (such as in Griffiths et al. (2020), but with reciprocal mitochondrial backgrounds) will shed light on the relative role of ecological adaptation driving this signature of uniform selection.
Consistent with divergent selection, we find that allele frequencies differ in 33% of the genome, depending on the mitochondrial background. Yet, concerning the targets of divergent selection, there is no region where opposite alleles dominate, as would be expected if the same genomic region was under divergent selection in both mitochondrial backgrounds, i.e. if mitonuclear incompatibilities had a symmetric genetic architecture. Instead, the location of genomic regions consistent with divergent selection strongly differs among mitochondrial backgrounds. For example, the SC mitochondria seems to interact strongly with regions in chromosomes 2, 3, 8, 9, 11 and 12, while the SD mitochondria seems to interact with regions in chromosomes 4, 6 and 9 (Fig. 3). In agreement, previous studies using F2 hybrids from reciprocal crosses have reported allele frequency changes during a single generation in multiple chromosomes, both in response selection for rapid development (5 chromosomes, in Healy & Burton 2020) or simply as a consequence of differential survivorship during development (4 to 5 chromosomes, in Lima et al (2019)). In contrast to those studies, where skews consistent with divergent selection were small in magnitude (up to ~15%) and were consistent across entire chromosomes, by evolving hybrids over multiple generations, we were able to identify genomic regions that span between half to small portions of the chromosomes (62 Kbp – 6.2 Mbp; Fig. 3). Nevertheless, our regions under divergent selection still contain hundreds or thousands of genes, and thus prevent identifying the causal genes underlying mitonuclear incompatibilities. Future “evolve-and-resequence” studies increasing effective recombination rate, either by increasing the size of experimental populations or the number of generations under experimental evolution, will provide higher resolution into the asymmetric and polygenic genomic architecture of mitonuclear incompatibilities described here.
Consistent with our hypothesis, the direction of divergent selection favoring mitonuclear adaptation is strongly supported, as nuclear alleles matched the mitochondrial background in 88.6 and 87% of the cases (Table 1). Some of the remaining windows with skews towards the mismatched allele can be explained by non-biological reasons. For example, in SD lines, all windows showing skews towards the mismatched SC allele are physical linked to a large region of chromosome 7 where the SC allele is favored in both mitochondrial backgrounds, and thus can be explained by the stochasticity of recombination events during this experiment. In contrast, in SC lines, windows with the mismatched allele can only be explained by antagonistic selection, as these are located in half of chromosome 11, which shows no change in the reciprocal cross. Although this pattern is usually not expected for mitonuclear incompatibilities, empirical studies in F2 hybrids show that it is not uncommon. In studies with F2 hybrids, Lima et al 2019 found patterns consistent with antagonistic selection in one to three chromosomes, depending on the population cross, and Healy & Burton (2020) found such pattern in F2 hybrids between SD and SC also in chromosome 11.
Taken together, our results offer new insights into the genomic architecture of mitonuclear incompatibilities. We find that mitonuclear incompatibilities are highly asymmetric, both in the magnitude of selection and in the location of targeted nuclear regions, supporting theoretical predictions of BDMIs involving uniparentally inherited genes (Turelli & Moyle 2007). Despite the small size and limited gene content of the mitochondrial genome, we find that it exerts significant divergent selection in many genomic regions located in multiples chromosomes of the nuclear genome. Our finding that mitonuclear interactions predominantly favor coadapted gene complexes imply that such early BDMIs can result in barriers to gene flow that are widespread across the genome. The rapid evolution of the mitochondria and coevolving nuclear genes during periods of geographic isolation, may thus constitute important reproductive barriers that persist after taxa establish secondary contact or sympatry.