Complex GEIs (haplotype × haplotype × environment) existed for all three of the male traits that we measured. In contrast, female phenotypes were in part shaped by interactions involving only the sire haplotype and environment. Furthermore, the relative contributions of additive and non-additive genetic variance underlying these same traits commonly changed across environments. We suggest that these results are likely to have implications on our understanding of several evolutionary concepts.
Genotype × Environment Interactions
Firstly, our results indicate that trait expression in male and female offspring is influenced by different levels of genetic complexity. In males, all traits were generally shaped by the specific combination of haplotypes that the offspring inherit from their parents in interaction with the environment; whereas in females, all traits were influenced by the sire effect only in interaction with the environment. Notably, across both sexes, the genetic effects were completely contingent on the environment in which the offspring were reared in. That is, in no case did we find Sire × Dam effects on F1trait expression that were general across environments. Likewise, in no case did we find additive main effects of the maternal or paternal haplotype on F1 trait expression that were general across environments.
The general lack of main effects of the parental haplotype on offspring phenotypes, and the prevalence of paternal haplotype × environment interactions for female traits, and the more complex paternal haplotype × maternal haplotype × environment interactions for male traits, adds a previously underappreciated level of complexity to our understanding of evolutionary dynamics within populations. The fact that no particular paternal haplotype resulted in greater phenotypic expression across all maternal haplotypes and environments is consistent with a resolution to the Lek Paradox (i.e. how genetic variation is maintained under directional selection) based on GEIs [62–68], as well as to the general question of whether non-additive gene interactions can contribute to maintaining genetic diversity within populations [69, 70]. A key assumption of the Lek Paradox is that traits subject to strong selection will initially harbour additive genetic variance that is eroded over time - unless there is a process that can counter this erosion. In line with this assumption, and acknowledging that GEIs might provide a resolution to the Paradox, we found that all male traits (likely to be under strong selection) harboured additive genetic variance (at least when measured at the standard thermal treatment), and that they were also all involved in G × E interactions. However, on the other hand, the existence of complex GEIs might well reduce the ability of females to accurately discriminate between males on the basis of acquiring good genes for their offspring, [1, 68, 71], particularly if the rank order in performance of a given set of male genotypes, for a given trait, changes dramatically across environments. This is because the consequences to a female of mating with a male of a particular genotype (in terms of the expression of the offspring traits measured here) will hinge on the genotype of the female involved as well as the thermal environment under which their offspring will develop. This level of complexity will presumably decrease, although not necessarily eliminate, the potential benefits to females of mate discrimination for good genes benefits, especially if there are costs involved in such choice.
Haplotype × haplotype × environment interactions are arguably more consistent with genetic compatibility models of quality, and it is worth asking whether such interactions might promote the evolution of mechanisms that allow females to discriminate between males on the basis of their genetic compatibility. Genetic compatibility models have risen to prominence in recent times [5, 7–10], with many quantitative genetic examples to suggest that Sire × Dam interactions are prevalent [18, 20–26], at least for juvenile life history traits (such as juvenile survival and development). One point worth noting is that there has previously been very little, if any, empirical attention devoted to examining whether Sire × Dam interactions (hence evidence for genetic compatibility) affect the expression of adult life history traits (reproductive parameters, morphologies and survival). In the context of the complex GEIs that we have identified in this study, if genetic compatibility models based on mate choice were to be relevant to this population of crickets in terms of offspring body and gonad size, then females would not only have to assess the quality of a focal male in relation to her own genotype, but also in relation to the environment in which the female will produce her offspring. This might therefore require the females to predict the environment in which the offspring will develop. The likelihood of such a scenario has to be assessed against how well our experimental set up reflects the natural environment of the crickets.
The thermal treatments that we used fall within the thermal range of the native distribution of T. oceanicus , as well as within the range of natural seasonal variation at the field collection site of these crickets in Canarvon, Western Australia. Given that T. oceanicus breeds year round [72, 73], it is realistic to expect that adults and developing nymphs in the wild at Carnarvon might experience temperatures within the range of our thermal treatments. Thus, our experiment shows that complex haplotype × environment interactions can indeed affect the expression of offspring phenotypes when thermal environments differ consistently by a few degrees, and such temperature differences may well be represented in the wild. That said, it is highly likely that crickets in the field are able to better offset variation in thermal environments through microhabitat selection than they are when in the lab (the crickets live in cracks in the ground or deep in the leaf litter where the microclimate remains warm). But note however, that we are unable to use this result to predict whether smaller thermal differences, or fluctuating temperatures, will affect the expression of parental haplotypes to the same extent. Further research is thus required to assess how the reaction norms tied to particular parental haplotypes respond to more subtle changes in the prevailing environment.
Although the traits that we assessed in this study were all morphological, the available evidence suggests that each is under selection and closely tied to fitness in males and females (see Methods). In this regard, these traits therefore provided a good opportunity to assess the relevance of genetic models, such as the good and compatibility genetic models to phenotypic expression, particularly given the paucity of information that currently exists as to the relevance of genetic compatibility models on adult trait expression (almost all previous studies have examined compatibility effects on juvenile fitness components). Models of good genes and genetic compatibility are almost always posed in the context of sexual selection, typically addressing the question of whether females might acquire good or compatible genes through mate choice. Therefore, it is worth asking whether the set of morphological traits that we measured might be used in mate choice or male-male competition for matings. Body and pronotum size are clearly traits whose expression might be readily assessed by the opposite sex, and as previously mentioned, studies in the closely related species T. commodus show that male body size is indeed under sexual selection [47, 48]. On the other hand, there is no evidence for an effect of male body size on fighting ability in T. Oceanicus . Body and pronotum size in females, however, is under natural selection (large females produce larger ovaries, r = 0.37 , thus more eggs [49, 52]. On the other hand, gonadal traits in crickets lie embedded within the body cavity where they cannot be directly assessed during mate choice. However, testes size is clearly under sexual selection in several species [50, 51], and its expression is phenotypically correlated to body size in T. oceanicus (r = 0.37, n = 504), thereby providing females with a mechanism by which to choose males on the basis of large ejaculate investment. Moreover, molecules produced by the testes clearly have important downstream effects on female reproductive biology in Teleogryllus. In T. Commodus, the testes generate an enzyme complex, Prostaglandin synthetase, that contributes to the production of the chemical messenger Prostaglandin, which is transferred to the females in the seminal fluid . These prostaglandins exert strong physiological effects on females, for example by stimulating oogenesis and elevating oviposition [76, 77], as well as by accumulating within the female's neural tissue (specifically in the head region , and possibly also suppressing the phonotaxis response to sexually signalling males ). Hence, we conclude that all of the morphological traits used in this study were relevant candidates for genetic quality models based on mate choice.
When it came to assessing the genetic architectures of the traits at hand, we generally found additive genetic variance for each male trait in the lab standard thermal environment (27°C), but not in the cooler environment (23°C). However, in females, we generally detected no such additive variance, with the exception of ovary mass at 23°C. This result is generally encouraging for good genes models based on mate choice, because it suggests that male candidate traits that are under sexual selection also exhibited an underlying genetic architecture consisting of standing additive variance. The disclaimer is that this genetic additively was only detected at 27°C, meaning that females might only draw benefits of discriminative mate choice if they can control or predict the environments in which their offspring are raised.
Previously, Simmons (2003) used a paternal half-sibling analysis to measure the coefficients of additive genetic variance in male testes and pronotum width, and female ovary mass and female pronotum width in T. oceanicus. While Simmons's results for pronotum width were consistent with our results in males and females , the measured CVAs for testes and ovary mass were different in magnitude to those measured in the current study. Specifically, at similar temperatures to that of the high temperature treatment in this study, Simmons found lower levels of additive variance for testis mass (CVA = 7.18 compared to 14.23). In contrast, he did find higher additive variance for ovary mass than we did in our study (32.05 vs. 10.06). These discrepancies are not surprising, and can be reconciled by considering that the two studies were separated in time by about a decade, and each study exhibited differences in their respective technical designs (e.g. in the breeding designs employed, the age and mating history of the focal individuals, and in sample size of captured genotypes from the wild type population).
As mentioned above, almost all of the detectable significant genetic variance for the morphological traits was found in the 27°C treatment, and we detected only CVA for ovary mass at 23°C. Although 27°C represented the standard rearing condition for crickets in our study, the crickets had only recently been sourced from the field population where they breed year round , and had not had any chance to adapt to the new rearing conditions (fewer than six generations in the lab under a heavy inbreeding protocol that meant that the genotypes were thus shaped under drift, not selection). As outlined within the Introduction, several studies prior to ours have detected differences in genetic architectures across environments [33–39], usually attributing one of the environments as the stressful or novel environment, and the other as the benign or standard environment. Given T. oceanicus breeds all year round in the field, it is clear that neither of the treatments is novel to the crickets, although it is likely that the cooler temperature was more stressful given that peak cricket abundances in the field occur during the warmer seasons . Finally, we point out that the sample sizes differed across treatments, with slightly fewer cells and replicates represented in the diallel cross at 23°C (Figure 1). This might have resulted in a general reduction of statistical power in the cooler treatment.
Reconciling the quantitative genetic and GEI analyses
The presence of haplotype × haplotype × environment interactions, and haplotype × haplotype interactions within each environment (see test statistics in Figure 2) suggested that we should have also found abundant levels of non-additive genetic variance underlying the morphological traits in each environment. This was not the case. Our non-additive genetic variance components were represented by σ2t, an estimate of the reciprocal nuclear interaction variance (dominance or epistatic genetic variance), and σ2k, an estimate of complex non-reciprocal non-additive variance (including possible interactions of paternal and maternal effects, and of nuclear and extra-nuclear effects; see methods). Breaking down the non-additive variance components into these two estimates provided us with a more detailed picture of the genetic architecture of the traits at hand. The consequence of doing this, however, is that it decreases the power of finding significant non-additive variance components (relative to other studies based on North Carolina II breeding designs that have used the Sire term to directly estimate additive genetic variance, and the Dam × Sire interaction to directly estimate the amount of non-additive genetic variance [24, 26–28]. This explains the discord between our GEI and the Biomodel results. In contrast, the Biomodel actually has greater power to detect additive genetic variance than similar models that simply use the Sire effect as an indicator of additive genetic variance (as in the GEI analyses), because it uses information from both sires and dams to calculate the additive variance . Accordingly, while we did not find additive sire effects at 27°C for any of the male traits measured (as indicated by the Sire × Treatment terms), we did detect additive genetic variance for all male traits at 27°C using the Biomodels. This indicates that additive variance is indeed present, but is swamped by more complex Sire × Dam × Treatment interactions in the GEI analyses.
Relevance to the natural world
Our study offers valuable insights into the likely complexity of gene by environment interactions in natural populations, and the evolutionary implications of these interactions. To generate these insights, we originally captured eight random parental genotypes from a wild population of T. oceanicus, and used these to generate 36 focal genotypes. Although these 36 genotypes represent a random sample of the total genetic variation segregating within the T. oceanicus population in question, it is important to realize that these genotypes nonetheless constitute only a tiny fraction of the total pool of possible genotypes within the population. This has implications. First, it is entirely possible that the parental genotypes that acted additively in our set of diallel lines, may not act additively in the wild population when matched against the complete pool of possible genotypes. Similarly, genotypes acting non-additively in our sample of genotypes, might exhibit additivity among the total pool of genotypes . Although this point is pertinent to most quantitative genetic studies, it is particularly relevant to diallel designs which start with a small set of parental genotypes. Second, in diallel breeding designs, the parents used are inbred (although the F1 offspring genotypes that are subsequently tested are completely outbred). This might mean that presence and magnitude of parental effects are not representative of the magnitude of effects that occur in wild populations, in which the parents are likely to be outbred. Thus, while the results we present are accurate for the specific combinations of haplotypes used here, we must be cautious when scaling these results up to the genetic effects that occur within the wider natural population, let alone the species. Having said that, our motivation was not to calculate the absolute value of each genetic variance component to each trait in this species, but rather to address the potential magnitude by which genetic variances and gene expression might vary across environments.