Toll-related immune genes are evolving recurrently in D. innubila likely due to strong pathogen pressures
We next sought to identify genes and functional categories showing strong signatures of adaptive evolution, suggesting recurrent evolution as opposed to recent local adaptation. We reasoned that if the population differentiation seen in antifungal genes and cuticle development proteins (Figure 2 & 3, Supplementary Figure 4) was due to local adaptation also acting over longer time periods, we would expect to see signatures of adaptation in those categories. Furthermore, Hillet al. used dN/dS-based statistics to show that genes involved in some immune defense pathways were among the fastest evolving genes in the D. innubila genome (Hill et al. 2019). We also sought to identify what genes are evolving due to recurrent positive selection in D. innubila in one or all populations, and if this is associated with environmental factors. To this end we calculated the McDonald-Kreitman based statistic direction of selection (DoS) (Stoletzki and Eyre-Walker, 2011) and SnIPRE selection effect (Eilertson et al., 2012) to identify an excess of selection. We then fit a linear model to identify gene ontology groups with significantly higher DoS or selection effect than expected. In this survey we find cuticle genes and antifungal genes did have some signatures of adaptive evolution (DoS > 0 and selection effect > 0 for 80% of genes in these categories) but as a group showed no significant differences from the background (GLM t-value = 1.128,p -value = 0.259, Supplementary Table 4). In fact, we only found two functional groups significantly higher than the background, Toll signaling proteins (GLM t-value = 2.581 p -value = 0.00986, Supplementary Table 3) and antimicrobial peptides (AMPs, GLM t-value = 3.66 p -value = 0.00025, Supplementary Table 3). In a previous survey we found that these categories were also the only functional groups to have significantly elevated rates of amino acid divergence (Hill et al. 2019). These results suggest that this divergence is indeed adaptive.
Interestingly, D. innubila is burdened by Drosophila innubila nudivirus (DiNV), a Nudivirus that infects 40-50% of individuals in the wild (Unckless 2011a; Hill and Unckless 2020). A close relative of the virus suppresses Toll-regulated AMPs in D. melanogaster (Palmer et al. 2018; Hill and Unckless 2020), which might explain why the Toll pathway and AMPs are fast evolving in D. innubila. Five AMPs showed consistently positive DoS and selection effect values (which are also among the highest in the genome): four Bomanins and Listericin . All are AMPs regulated by Toll signaling (and additionally JAK-STAT in the case of Listericin ) (Hoffmann 2003; Takeda and Akira 2005). Listericin has been implicated in the response to viral infection due to its expression upon viral infection (Dostert et al. 2005; Zambon et al.2005; Imler and Elftherianos 2009; Merkling and van Rij 2013). For all immune categories, as well as cuticle proteins and antifungal proteins, we find no significant differences between populations for either MK-based statistics, and no significant differences in the distribution of these statistics between populations (GLM t-value < 0.211, p-value > 0.34 for all populations, Supplementary Table 3). Thus, perhaps selection at these loci is ubiquitous and genes flow between populations homogenizes that signature.
Mutation rates, efficacy of selection and population structure can vary across the genome, which can confound scans for selection (Charlesworth et al. 2003; Stajich and Hahn 2005).To work around this, we employed a control-gene resampling approach to identify the average difference from the background for each immune category (Chapman et al. 2019). Consistent with our results previous results, we find no signatures of recurrent positive selection in antifungal genes (Supplementary Figure 10, 61% resamples > 0) or cuticle genes (Figure 6, 54% resamples > 0) but do again find extremely high levels of positive selection in AMPs (Figure 6, 100% resamples > 0) and Toll signaling genes (Figure 6, 99.1% resamples > 0). Segregating slightly deleterious mutations can bias inference of selection using McDonald-Kreitman based tests (Messer and Petrov 2012). To account for this bias, we also calculated asymptotic α for all functional categories across the genome (Haller and Messer 2017). To this end we calculated the asymptotic α for all functional categories across the genome (Haller and Messer 2017). As before, while we find signals for adaptation in antifungal and cuticle proteins (asymptotic α > 0), we find no evidence of higher rates of adaptation than the background (Supplementary Figure 10, permutation test Antifungal p -value = 0.243, Cuticlep -value = 0.137). Again, the only categories significantly higher than the background are Toll signaling genes (Permutation testp -value = 0.033) and AMPs (Permutation test p -value = 0.035). Together these results suggest that while genes involved in antifungal resistance and cuticle development are evolving adaptively, it is not recurrent across the whole functional category, instead only occurring in one or two specific genes. Alternatively, the adaptation may be too recent to detect signal using these metrics. Long-term recurrent adaptation appears to be driven by host-pathogen interactions (likely with DiNV (Hill et al. 2019)) as opposed to local adaptation.
Figure 6: McDonald-Kreitman based statistics for immune categories in D. innubila and cuticle development. The left two plots show estimated statistics (Direction of Selection and Selection Effect) for each gene. The right two plots show the difference in average statistic (Direction of Selection and Selection Effect) for each gene and a randomly sampled nearby gene.