PATERNAL EFFECTS ON THE EXPRESSION OF A MALE POLYPHENISM

Authors


Abstract

Polyphenic traits are widespread, but compared to other traits, relatively few studies have explored the mechanisms that influence their inheritance. Here we investigated the relative importance of additive, nonadditive genetic, and parental sources of variation in the expression of polyphenic male dimorphism in the mite Rhizoglyphus echinopus, a species in which males are either fighters or scramblers. We established eight inbred lines through eight generations of full-sibling matings, and then crossed the inbred lines in a partial diallel design. Nymphs were isolated and raised to adulthood with ad libitum food. At adulthood, male morph was recorded for all male offspring. Using a Cockerham–Weir model, we found strong paternal effects for this polyphenic trait that could be either linked to the Y chromosome of males or an indirect genetic effect that is environmentally transmitted. In additional analyses, we were able to corroborate the paternal effects but also detected significant additive effects questioning the Cockerham–Weir analysis. This study reveals the potential importance of paternal effects on the expression of polyphenic traits and sheds light on the complex genetic architecture of these traits.

A great variety of morphological and life-history traits vary dichotomously (Moran 1992), including color (Hazel 2002), shape, or presence of morphological structures (see references in Roff 1994, 1996; Brockmann 2001), as well as behavior (Moczek and Emlen 2000), and seasonal diapause (Mousseau and Roff 1989). Single locus inheritance for dichotomous traits is well understood in the light of evolutionary game theory (Maynard Smith 1982). However, the majority of dichotomous traits are phenotypically plastic and represent a polyphenism, where a single genotype is capable of expressing alternative phenotypes under specific environmental conditions (West-Eberhard 2003). Polyphenic dimorphisms have been modeled as a conditional evolutionarily stable strategy, which can evolve and be maintained when individuals’ phenotypes are decided through conditional decision rules (Dawkins 1980; Gross 1996). Unfortunately, evolutionary game theory models do not account for the complex genetics and inheritance patterns of conditional dimorphisms.

The first contribution toward understanding the genetic architecture of conditional dimorphisms from a quantitative genetics perspective was the threshold trait concept introduced by Wright (1934a,b) and subsequently developed by Dempster and Lerner (1950). This concept assumes that dimorphic traits have an underlying continuous variable (named “liability”), coupled with a threshold mechanism that generates discontinuity on its phenotypic expression (Falconer 1989). Threshold traits are often sensitive to environmental cues, and the “environmentally cued threshold” model (henceforth ET) is the current framework for understanding the evolution of conditional strategies (Hazel et al. 1990; Roff 1994; Hazel et al. 2004). This model considers that the sensitivity to the environmental cue is in itself a polygenic quantitative trait with normally distributed variation (Tomkins and Hazel 2007), reinforcing the importance of genetic variation for the expression of conditional traits.

An increasingly large number of studies have demonstrated the evolutionary relevance of gene-by-environment interactions across distantly related taxa (references in Pigliucci 2005). This collection of empirical studies makes it clear that genetic variation for plastic traits is a very general pattern, and therefore the degree to which polyphenic traits can be conditionally expressed is expected to depend on underlying genetic variation (West-Eberhard 2003). In fact, the origin of polyphenisms has been shown to depend on genetic accommodation, a mechanism that starts with a mutation affecting development and allowing environmental variation to expose genetic variants, ultimately leading to an increased environmental sensitivity of a plastic phenotype (Suzuki and Nijhout 2006; Suzuki and Nijhout 2008). This mechanism differs from the more widely accepted notion of genetic assimilation (which ultimately results in the loss of environmental sensitivity due to the canalization of the new phenotype), because genetic accommodation might increase the phenotype's sensitivity to environmental cues (West-Eberhard 2003; Suzuki and Nijhout 2006).

Several illuminating examples of the importance of genetic effects for the expression of polyphenisms come from social insects. For example, it had been long been assumed that caste determination in the social Hymenoptera was purely environmentally determined, so that every single female larva was potentially capable of becoming a reproductively mature queen, or a sterile worker (Wheeler 1986). However, recent studies have challenged this assumption, showing that the evidence for different genotypes in a colony being similarly capable of developing into any caste is more apparent than real (Schwander et al. 2010). On the contrary, many cases of caste determination in the social Hymenoptera are affected by a variety of direct and indirect genetic effects (Schwander et al. 2010), including genetic compatibility between queens and their mates (Schwander and Keller 2008).

A crucial next step for understanding the genetics of polyphenisms is investigating the polygenic nature of a threshold trait's sensitivity to its environmental cue. We studied this question in the acarid mite Rhizoglyphus echinopus, a species in which males are conditionally fighters or scramblers. Male dimorphism in R. echinopus is environmentally cued by body size and male density (Radwan 2001; Tomkins et al. 2011), but there is a great deal of overlap in the sizes of males that become either fighters or scramblers, even under the same male density. This overlap suggests that there is either a large genetic variation for the switch point that links male morph expression to the body size cue, or alternatively some genotypes in the population are entirely canalized to expressing one of the morphs regardless of body size and male density (Lively et al. 2000; Buzatto et al. 2012). In a previous study, we established inbred lines to scrutinize the genetic variation for the switch point between male morphs, and we found that inbred lines had heterogeneous levels of conditionality that depicted genetic variation for the switch point in the base population (as predicted by the ET model), but also suggested a mixture between canalized and conditional strategists in R. echinopus (Buzatto et al. 2012). These results revealed that genes of major effect that canalize morph expression might overlay conditionality in this polyphenic trait, suggesting that the genetic architecture of this threshold trait deserves further attention.

In the present study, we investigated the genetic architecture of the male polyphenism in R. echinopus, applying a partial-diallel cross-classified design using eight inbred lines. Because our experimental design included reciprocal crosses between some of our inbred lines, we were able to partition the additive, nonadditive genetic, maternal, and paternal variances. We found important paternal effects for the expression of dimorphic males in R. echinopus, but weak additive and nonadditive genetic, and no maternal effects. These paternal effects could result from direct sire genotypic effects (Y-linked genes, e.g.) or indirect genetic sire effects (genetically determined effects of paternal origin that are environmentally transmitted). To the best of our knowledge, this is the first time such a powerful tool from the field of quantitative genetics has been used to examine paternal effects in the genetic architecture of a dimorphic trait that is mainly environmentally cued. Our results shed light on the poorly understood interaction between environmental cues and genetic variation for conditionally expressed traits.

Materials and Methods

MODEL ORGANISM

The mites used in this study were derived from a base colony of R. echinopus that we originally collected from an infested organic onion in August 2005. The base colony have been kept in six petri dishes at 22°C and >90% humidity for over 50 generations, with a standing adult population of a few thousand individuals (more details in Buzatto et al. 2012). Rhizoglyphus echinopus presents two distinct male morphs coupled to alternative mating behaviors (Radwan 2009). Fighter males possess a very thick and sharply terminated third pair of legs, which these males use to kill rivals and monopolize females. In contrast, scramblers’ legs are all equally thin and without a sharp tip, and scramblers search for unguarded females to mate with (Radwan 1993; Radwan 2009). Male dimorphism in the bulb mite R. echinopus is environmentally cued by body size and male density (Radwan 2001; Tomkins et al. 2011).

ESTABLISHING INBRED LINES

To initiate the inbred lines, we isolated approximately 100 larvae from the base colony and reared them individually with ad libitum food. Using the resulting virgin adults, we paired 40 couples separately. Next, we isolated 20 larvae produced by each of these 40 couples, and reared them to adulthood. Subsequently, virgin full siblings were paired and their offspring reared in the same way (more details of how we reared individual larvae and paired virgin adults can be found in Buzatto et al. 2012). Throughout inbreeding, approximately 14% of full-sibling pairings failed to produce any offspring rendering the lines extinct. We successfully repeated the procedure of pairing full siblings and raising their offspring individually in eight inbred lines, for eight generations, leading to an expected inbreeding coefficient of F= 0.826.

CROSSING INBRED LINES IN A PARTIAL DIALLEL DESIGN

We paired a virgin adult female to a virgin adult male in all the combinations of inbred lines selected for our partial diallel design (Table 1). We attempted to perform each selected cross four times, but the number of crosses that were successful at producing offspring varied from 0 to 4 among the different combinations of inbred lines. Our design generated 28 independent combinations of full siblings, four independent combinations of reciprocal full siblings, 36 independent combinations of maternal half siblings, 36 independent combinations of paternal half siblings, 88 independent combinations of reciprocal half siblings, and 214 independent combinations of unrelated individuals. We used scramblers as sires when scramblers were more common than fighters in the sire line and fighters as sires when fighters were more common than scramblers in the sire line. For the lines in which scramblers and sires were similarly common, we randomly used fighters or scramblers as sires in different replicates of the crosses. Successful pairings produced approximately 200 eggs. As soon as these eggs started to hatch, we isolated 50 larvae produced by each pair, and raised them to adulthood individually, as described in Buzatto et al. (2012). At adulthood, we scored all the male offspring for morph. During the whole experiment, we kept all adults and juveniles at 22 °C (Binder KB 240 cooled incubator) and >90% humidity.

Table 1.  The partial diallel design used to cross eight inbred lines and analyze the genetic architecture of morph determination in the bulb mite Rhizoglyphus echinopus. Shaded cells represent crosses with reciprocals (e.g., sire line 2 × dam line 6 and sire line 6 × dam line 2). Sample sizes are shown for the number of male offspring generated in each cross, and numbers in the same cell (separated with commas) represent replicate crosses. Numbers in bold represent offspring with fighter sires, whereas the remaining numbers represent offspring sired by scramblers. The distribution of missing crosses in our design is fairly homogenous, and these missing cells should not bias the estimation of variance components (as in Roff and Sokolovska 2004).
Sire lineDam line
12345678Missing
1    2834, 40, 832, 9, 3930, 303
2 0      36, 40, 35, 31, 24, 37, 4
       37, 18 37 33, 36  
3 24 18, 15,      16, 29, 27, 24, 3
   18      33, 45 31  
4 0 15, 18, 21, 17      28, 23, 4
  17, 21      23, 3  
51613, 21,24, 18, 4, 18    3
   15, 17 19, 19       
6   44, 20, 23, 27, 28, 19 24     3
  17, 1121      
7  19, 17,35, 26020, 34,  4
   26, 25  30   
8     27, 20 0 38, 42, 38, 23   4
       35, 35    
Missing53335333 

DATA ANALYSIS

After scoring the male offspring of all crosses for their morph, we used the Cockerham–Weir model (called “biomodel” by Cockerham and Weir 1977; Lynch and Walsh 1998) to model the probability of an individual offspring becoming a fighter and estimate the six components of genetic and environmental variance. The equation we used was

image

in which Yijkl is the probability of becoming a fighter for the lth son from the kth replicate of the cross between sire i and dam j, μ is the mean proportion of fighters in the population, and morphi is the fixed factor of sire i morph. The remaining terms are random effects assumed to be mutually independent and normally distributed, and to have mean zero: Ni and Nj are the haploid nuclear contribution (additive effects) from lines i and j, independent of sex; Tij is the nonadditive interaction of the haploid nuclear contributions (including dominance and epistatic effects); Mj is the maternal genetic and environmental effects of line j when used as dams; Pi is the paternal genetic and environmental effects of line i when used as sires; Kij is the sum of all nuclear–extranuclear and extranuclear–extranuclear interactions (i.e., all possible interactions between maternal effects, paternal effects, and nuclear effects); Rk(ij) is the effect of kth replicate cross within the sire i× dam j combination; and Wl(k(ij)) is the residual (within replicate cross) effect of individual l (as in Fry 2004; Ivy 2007; Bilde et al. 2008; Dowling et al. 2010). The Cockerham–Weir model also assumes that the variances of additive nuclear effects through sires and dams are the same, and that the reciprocal dominance effects, Tij and Tji are identical. However, the reciprocal effects of Kij and Kji are not necessarily equal, which accounts for the fact that cytoplasmatic elements contributed by sires and dams might differ (Lynch and Walsh 1998).

To estimate these variances, we fitted the Cockerham–Weir model using the default estimation method in the GLIMMIX procedure in SAS version 9.2 (SAS Institute 2004). We used the TYPE = LIN command to model the covariance between families as linear functions of the variances (see Fry 2004 for details of how these covariances are modeled and their biological interpretations). We also used the COVTEST statement to fit reduced models in which a given covariance parameter was set to 0, and the comparison of these models with the full model (where all parameters were allowed to have positive values) provided statistical inferences about the covariance parameters through likelihood ratio tests. We performed this analysis on a covariance matrix of 406 pairwise comparisons between the 28 families derived from our successful between-line crosses (see Table 1).

Next, we used the observational variance components (represented by sigma's) to estimate the causal variance components (represented by V's; according to Bilde et al. 2008), always assuming that our inbred lines represent a random sample from the base population, and that epistasis is small:

  • (1) σN2nuclear additive variance, VA= 2σN2/F, where F is the inbreeding coefficient.
  • (2) σT2:  nuclear interaction (dominance) variance, VDT2/F2.
  • (3) σM2:  maternal effect variance VM, which can result from maternal genotype or maternal environment effects.
  • (4) σP2:  paternal effect variance VP, which can result from paternal genotype or paternal environment effects.
  • (5) σK2:  interaction variance VK, which can result from interactions between paternal and maternal effects, and/or interactions between nuclear and extra-nuclear effects.
  • (6) σR2:  among replicate crosses variance.
  • (7)  σW2:  within replicate crosses variance.

Bilde et al. (2008) noted that σR2 and σW2 could only directly represent the environmental variance VE if the parental lines were fully inbred (F= 1). Therefore, to estimate VE, we first summed all the observational components of variance to obtain the total phenotypic variance (VTOT), and then subtracted all the other causal components of variance from VTOT.

All data used in this study is available as supplementary information.

Results

SINGLE LOCUS INHERITANCE

The proportion of fighters in the offspring of all crosses ranged from 0% to 76.2% (mean ± standard deviation = 26.3 ± 19.8%, n= 79 crosses, Fig. 1). To investigate the possibility of single-locus inheritance, we examined graphically the morph ratios of offspring sired by fighters and scramblers (as in Tomkins et al. 2004). If the sire phenotype is dominant (with possible genotypes AA and Aa), the male offspring he produces when mated to a female of unknown genotype (AA, Aa, or aa) should express the same phenotype of their father with expected ratios of 1:0, 3:1, or 1:1. On the other hand, if the sire phenotype is recessive (necessarily aa), the male offspring he produces when mated to a female of unknown genotype (AA, Aa, or aa) should express the same phenotype of their father with expected ratios of 0:1, 1:1, or 1:0. The hypothesis that male morph possess single-locus inheritance in R. echinopus can clearly be rejected by Figure 2, in which the offspring of scrambler and fighter males are plotted against the expected ratios.

Figure 1.

The frequency of fighters produced in the offspring of each cross in a partial diallel design in the bulb mite Rhizoglyphus echinopus. Each bar represents the average frequency of fighters in all replicates of that cross. The within-line crosses (the diagonal) was used to illustrate the frequency of fighters in each parental line, but was not included in the partial diallel analysis.

Figure 2.

Square-root numbers of fighters and scramblers in the offspring of fighter (A, C) and scrambler (B, D) sires. Square-root transformation was used to facilitate visual interpretations of the ratios in plots in which both axes have the same scale. Dashed lines represent the expected proportions in the offspring under single gene inheritance. If fighter phenotype is dominant and scrambler phenotype is recessive (upper panels A and B), the offspring of a fighter (AA or Aa) should present fighter ratios of 1:0, 3:1, or 1:1 when mated with a female of unknown genotype (AA, Aa, or aa), whereas the offspring of a scrambler (aa) should present fighter ratios of 1:0, 1:1, or 0:1. Conversely, if fighter phenotype is recessive and scrambler phenotype is dominant (lower panels C and D), the offspring of a fighter (aa) should present fighter ratios of 0:1, 1:1, or 1:0 when mated with a female of unknown genotype (AA, Aa, or aa), whereas the offspring of a scrambler (AA or Aa) should present fighter ratios of 0:1, 1:3, or 1:1. Open circles represent data that are consistent with at least one of the expected ratios, and full circles represent data that are significantly different (P < 0.05 in χ2-tests on untransformed values) to all expected ratios.

COCKERHAM–WEIR ANALYSIS

In the Cockerham–Weir model fitted in SAS, the fixed effect of sire morph on offspring morph determination was not significant (P= 0.32). The Cockerham–Weir model also failed to detect any variance in morph determination due to additive genetic effects, maternal genotype and maternal environment effects, or to interactions between paternal and maternal effects, or between nuclear and extra-nuclear effects (Table 2). The variance attributable to interactions between nuclear haploid genomes (indicative of dominance or epitasis) was responsible for about 11% of the phenotypic variation in male morph, but was not statistically significant. Paternal effects on male morph in the offspring, on the other hand, were significant, accounting for over 21% of the phenotypic variation in this trait (Table 2). We also detected variance among replicate crosses and a large amount of variance among males within-replicate crosses (accounting for 68% of the total variance), which is expected in an environmentally cued trait.

Table 2.  The results from fitting the Cockerham–Weir model to the data obtained with a partial diallel design to investigate the genetic architecture of morph determination in the bulb mite R. echinopus: the observational variance component estimates (with standard errors in parenthesis and P-values obtained with likelihood ratio tests; see section Methods), and the causal variance component estimates. Percent represents the proportion of total phenotypic variance explained by each parameter of the Cockerham–Weir model (see section Methods).
ObservationalCausal
Variance componentEstimate (SE) P-valueVarianceBiological interpretationEstimatePercent
σN20 V A Nuclear additive00
σT2 0.1488 (0.1303) 0.4242 V D Nuclear dominance 0.2181 10.85
σM20 V M Maternal effects00
σP2 0.4237 (0.4015) 0.0089 V P Paternal effects 0.4237 21.08
σK20 V K Interactions between the above00
σR2 0.5357 (0.1512) <0.001 V E Environmental 1.3685 68.07
σW20.9021 (0.02921) 
    V TOT   2.0103  

ADDITIVE EFFECTS

Although the Cockerham–Weir model did not detect significant additive effects, such effects on morph ratio can be revealed with regressions of average offspring values on midparent values (Falconer 1989). Indeed, the mean proportion of fighters in the offspring of each type of cross was positively related to the average morph ratio of the parental lines (R2= 0.492, df = 22, P= 0.0001; Fig. 3A) and to the average body size of the parental lines (measured as quiescent tritonymph weight [QTW], see Buzatto et al. 2012 for details; R2= 0.326, df = 22, P= 0.0036; Fig. 3B). We also calculated the heritabilities of male morph using a modified formula of Falconer's proband method that takes into account the unequal variances of liability in the parental and offspring generation (eq. 18.3, page 303 in Falconer 1989). This formula assumes that the affected individuals (fighters in our case) are selected as parents in the parental generation, and uses the proportion of fighters in the parental and offspring generation to calculate the heritability in the underlying liability scale. Therefore, we only calculated the heritabilities for the crosses that had fighter sires, and we used the average between the morph ratios of parental lines in the formula. We found that mean heritability was 0.81 (standard error: 0.08, range: 0.16–1.50, n= 25 crosses).

Figure 3.

The mean frequency of fighters in the offspring of each cross was positively related to the average morph ratio (A) and to the average body size (measured as quiescent tritonymph weight [QTW]; B) of the parental lines. Dashed lines represent 95% confidence intervals around the regression line.

PATERNAL AND MATERNAL EFFECTS

To test for paternal and maternal effects outside the Cockerham–Weir analysis, we grouped the crosses by sire line (across different dam lines) and by dam line (across different sire lines), and looked for correlations between these values and the values of paternal and maternal lines. The average morph ratios of the crosses, when grouped by sire line, were significantly positively correlated with morph ratios in the paternal lines (R2= 0.550, df = 6, P= 0.033; Fig. 4A), but the average morph ratios of the crosses grouped by dam line were not significantly correlated with morph ratios in the maternal lines (R2= 0.406, df = 6, P= 0.089; Fig. 4C). The average body sizes of the paternal lines (again measured as QTW) were not significantly correlated with morph ratio in the offspring when the crosses were grouped by sire line (R2= 0.294, df = 6, P= 0.165; Fig. 4B). The average body sizes of the maternal lines were also not significantly correlated with morph ratio in the offspring when the crosses were grouped by dam line (R2= 0.129, df = 6, P= 0.383; Fig. 4D). Finally, the average morph ratios of the crosses grouped by sire line (across different dam lines) varied significantly more than the average morph ratios of the crosses grouped by dam line (across different sire lines; F7,7= 7.299, P= 0.018; Fig. 5).

Figure 4.

The average morph ratios of the crosses grouped by sire line were positively correlated with morph ratios in the paternal lines (A; dashed lines represent 95% confidence intervals around the regression line), but not to the average body sizes of the paternal lines (measured as quiescent tritonymph weight [QTW]; B). The average morph ratios of the crosses grouped by dam line were not correlated either with the morph ratios in the maternal lines (C), or with the average body sizes of the maternal lines (D).

Figure 5.

The frequency of fighters in the offspring produced by the crosses between inbred lines of the bulb mite Rhizoglyphus echinopus. Data were grouped by sire line (across different dam lines; A) or by dam line (across different sire lines; B). Averages and standard deviations are shown, and lines were ordered according to these averages for ease of visual inspection. Variation was significantly higher between crosses grouped by sire line than between crosses grouped by dam line (see section Results), indicating that the former was a better predictor of fighters’ frequency in the offspring than the latter.

Discussion

Here we have investigated the mechanisms that influence the inheritance of conditional male dimorphism in the mite R. echinopus. With a classical cross-classified design, the partial diallel (Lynch and Walsh 1998), we attempted to disentangle the relative importance of additive, nonadditive genetic, and parental sources of variation. Below we discuss how our results provide evidence for weak additive genetic variance and strong paternal effects on the expression of this polyphenism.

ADDITIVE EFFECTS

Our first striking result was the lack of significant additive variance in the Cockerham–Weir analysis for morph determination in R. echinopus. This result was unexpected because offspring body size is positively related to the probability of becoming a fighter in juveniles (Tomkins et al. 2011, Buzatto et al. 2012) and therefore any additive genetic variance for body size would be indirectly detected by the Cockerham–Weir model as additive variance for morph determination. The evidence suggests that the lack of significant additive effects in the Cockerham–Weir model reflects an underestimation of these effects rather than their complete absence, however. First of all, the mean frequency of fighters in the offspring of each cross was positively related to the average morph ratio (Fig. 3A) and to the average body size of the parental lines (Fig. 3B). Regressions of average offspring values on midparent values provide a straightforward alternative method to the Cockerham–Weir model for detecting additive variance (Falconer 1989): under purely additive effects offspring trait values should correspond to the average of parental values; departures from this relationship indicating environmental effects, dominance, epistasis, or indirect genetic effects. Furthermore, our measures of heritability of male morph (in the underlying liability scale, Falconer 1989) using midparent values returned high values (0.81 on average). It is hence surprising that we detected this additive variance with these simple approaches, but failed to detect it with the Cockerham–Weir model.

The contrasting results from our regressions of average offspring values on midparent values and our Cockerham–Weir analysis probably reflect a conservative estimate of variance components by the latter. This could be a property of the model itself, or because our breeding design was a partial diallel, which generates fewer independent combinations of the different types of siblings than a full diallel and is therefore likely to be less powerful. Nevertheless, the missing cells were designed to be distributed throughout the diallel in a balanced way, such that the only possible bias would be uniform and in a downward direction for all the genetic variance estimates (Roff and Sokolovska 2004; Dowling et al. 2010). Our conclusion is that it is important to recognize the reduced power of the Cockerham–Weir model, and this means that great caution should be exercised when interpreting partial diallel designs in the absence of other analyses. We suggest that alternative methods for estimation of additive effects (such as regressions of average offspring values on midparent values) should always be presented alongside the results of a Cockerham–Weir model, such that reliance is not placed solely on this model.

In conclusion, clearly there is evidence for the existence of some additive genetic variance for morph determination in R. echinopus, coming from our regressions of offspring morph ratios on midparent morph ratio and midparent body size, even though the Cockerham–Weir analysis failed to detect it. However, caution must be exercised when interpreting the magnitude of such additive effects, as offspring on midparent regressions are not capable of disentangling extra-nuclear effects from additive nuclear effects.

PATERNAL EFFECTS OR SIRE MORPH EFFECTS?

Even though the lack of significant additive effects indicated by the Cockerham–Weir model reflects an underestimation, it is important to note that these additive effects are probably weaker than the paternal effects detected in the same model as the strongest source of heritable variation for morph determination in R. echinopus. These results are consistent with the fact that we detected higher variation between the frequency of fighters produced by crosses grouped by sire line than between crosses grouped by dam line, which suggests that sire line predicts fighters’ frequency in the offspring better than does dam line. Likewise, we found that the average morph ratios of the crosses grouped by sire line were positively correlated with morph ratios in the paternal lines, whereas the average morph ratios of the crosses grouped by dam line were not correlated with morph ratios in the maternal lines. We interpret these results as evidence that sire line effects are stronger than dam line effects or additive effects, pointing to the importance of the same paternal effects that we detected with the Cockerham–Weir model.

Our approach disentangles the paternal effects of sire morph from other paternal effects (such as body size, for instance) that are environmentally or genetically transmitted through the sire line. In a previous study on the closely related species R. robini (Smallegange 2011), such effects would have been confounded with sire morph effects, given that no sire traits (such as body size, for instance) were measured in that study. Therefore, any sire trait that correlates with sire morph would have been detected as a sire morph effect by Smallegange (2011). Even in the present study, if we analyze the offspring of all crosses together regardless of parental line, a t-test on arcsine square-root transformed proportions in the offspring indicates that the proportion of fighters was significantly lower in offspring sired by scramblers than in offspring sired by fighters (analysis not shown). However, using the same dataset and properly accounting for the genotypes of sires and dams with the Cockerham–Weir model fitted in SAS, we found no significant effect of sire morph on offspring morph determination. In that regard, our approach provided a very powerful way to conclude that sire morph is not in itself an important factor in determining offspring morph in R. echinopus, and perhaps generally in bulb mites. It is possible that sire morph effects detected in previous studies with male-dimorphic mites could actually represent other sorts of paternal effects that might correlate with sire morph such as size.

POSSIBLE SOURCES OF PATERNAL EFFECTS

There are numerous possible explanations for the paternal effects that are clearly operating on morph determination in R. echinopus. The first potential explanation is that a major gene that influences the expression of male morph resides on the Y chromosome. Because R. echinopus is known to possess XY sex determination (Grondziel 1975), genetic variation in the Y chromosome among our inbred lines would be evident when analyzing the offspring of different sire lines (across dam lines, see Fig. 5A), but not the offspring of different dam lines (across sire lines, see Fig. 5B), which is consistent with our results. The moving of alleles (or their regulatory elements) that are under sexually antagonistic selection to the hemizygous sex chromosome (the male Y in R. echinopus) has been seen as one of the ways to resolve intralocus sexual conflict (Stewart et al. 2010). Investment in fighting legs would likely be costly for female R. echinopus, in terms of resources diverted away from general body size and fecundity. Indeed, the fighter phenotype in R. echinopus is sex limited, as females never express a thick and sharply terminated third pair of legs and never fight. It is therefore possible that the morphology of the third pair of legs in this species went through a period of intralocus sexual conflict in its evolutionary history, and this conflict was eventually resolved with the genes coding for fighter legs being moved to the Y chromosome. Similar paternal effects due to genes linked to the sex chromosomes have been detected for the wing dimorphism in the cricket Gryllus rubens (where males are XO, and females XX; Walker 1987, but see Zera and Rankin 1989).

Nevertheless, paternal effects can also result from genes that reside in the X chromosome of sires. Even though this possibility sounds counterintuitive at first, a recent study of Drosophila melanogaster found that variation on the X chromosome of sires affected the fitness of their male offspring, demonstrating that nontransmitted paternal chromosomes can influence the phenotype of their offspring (Friberg et al. 2012). In fact, epigenetic reprogramming of DNA in any chromosome carried by sperm cells would also generate the same kind of results. This has been suggested for D. simulans, in which paternal photoperiodic conditions influence the development times of their offspring (Giesel 1986). Likewise, the temperature in which sires of D. melanogaster are raised affects wing length in the offspring (Crill et al. 1996), and the host-plant species in which sires of the seed beetle
Stator limbatus are reared influences the relationship of host-plant species and development time in the offspring (Fox et al. 1995). In our present experiment, however, we mitigated effects of paternal environment, as the environmental conditions under which sires were reared were completely controlled (standardized temperature, light cycle, population density [every mite was reared in isolation], and diet [ad libitum yeast provided]). Therefore, the only form of epigenetics that could be responsible for the paternal effects detected in the present study is genomic imprinting, where the expression of an allele differs depending upon its parent of origin (Day and Bonduriansky 2004).

Genetically determined effects of paternal origin that are environmentally transmitted represent another possible explanation for the paternal effects on morph determination in R. echinopus. In the Australian field cricket, for instance, paternal effects on offspring phenotype result from the large ejaculates or accessory gland secretions that provide resources to the female (Garcia-Gonzalez and Simmons 2007). Accessory gland secretions are also shown to enhance offspring fitness in D. melanogaster (Priest et al. 2008). As long as there is genetic variation for such accessory gland secretions among sires, this variation would generate indirect genetic effects on the offspring phenotype that would be detected as paternal effects by the Cockerham–Weir model. In bulb mites, a previous study with R. robini revealed that the fecundity of females is associated with a polymorphism in the phosphogluconate dehydrogenase (Pgdh) genotype of their mates (Konior et al. 2006). If this effect of sire genotype on dam fecundity also influences other reproductive traits of females, such as egg size for instance, genetic variation for the Pgdh genotype could also generate a paternal effect on morph determination because egg size is connected to morph expression in the offspring (Smallegange 2011 and see discussion on maternal effects below).

MATERNAL EFFECTS

At first glance, the fact that we did not detect maternal effects on morph contrasts greatly with a recent study on the congeneric species R. robini, in which there were significant maternal effects through egg size on morph determination (Smallegange 2011). However, Smallegange's (2011) results might align with ours, as the maternal effects detected by Smallegange (2011) would probably be detected as paternal effects in a diallel approach such as the one we used. The reason being that Smallegange's (2011) study also found a sire effect on the size of eggs produced by females. If a female's mate influences the size of her eggs, paternal effects could actually be the ultimate determining factor of egg size, as sires would actually be driving the differential allocation of resources into eggs by dams (Burley 1986). In dung beetles, for example, the size of male influences the sizes of the brood balls produced by their mates, which ultimately affects offspring phenotype (Kotiaho et al. 2003). Similar results have been reported for butterflies (Wedell 1996), birds (Cunningham and Russell 2000; Gil et al. 1999), and fish (Kolm 2001), all indicating that males affect female resource allocation to the offspring, and pointing toward maternal effects that are driven by paternal effects.

The egg size effect on offspring phenotype detected in bulb mites by Smallegange (2011) was only significant under poor dietary conditions. Given that all our mites developed on a rich yeast diet, the absence of maternal effects in our study could also come down to the fact that this egg size effect was in fact not relevant in our experimental conditions. Nevertheless, we still think that the results from Smallegange's (2011) study on R. robini could shed light on the proximate mechanism through which the paternal effect that we detected operates. Investigating paternal effects on egg size in R. echinopus is hence a promising next step to understand the complex gene by environment interactions that underlie the expression of male polyphenism in this species.

IMPLICATIONS FOR THE ET MODEL

By definition, polyphenisms are environmentally sensitive, and the ET model treats the sensitivity to the environmental cue as a heritable quantitative trait (Hazel et al. 1990; Tomkins and Hazel 2007). Therefore, the ET model approach is capable of identifying the value of the environmental cue that represents the mean switch point to which a population will evolve, which is the mean switch point at which the overall selection differential is zero (see Box 3 in Tomkins and Hazel 2007). Because the response to selection is a product of the selection differential and the heritability of a trait (the “breeder's equation,”Lynch and Walsh 1998), the efficiency with which selection will move the switch point mean of a population to its equilibrium is directly proportional to the heritability of switch point.

In R. echinopus, male dimorphism is environmentally cued by body size and male density (Radwan 2001; Tomkins et al. 2011), and in a previous study we suggested that there is large genetic variation for the switch point that links male morph expression to the body size cue (Buzatto et al. 2012). However, the heritability of switch point variation is constrained by the fact that male dimorphism is sex limited in its expression: genes that affect the switch point controlling male dimorphism are not expressed in females. The paternal effects on male dimorphism that we have detected might mean that sons inherit the switch point characteristics of their father rather than their mother; this may be adaptive because it is their father's switch points that are under selection while their mother's switch points (if they have them) are hidden from selection (the storage effect; Reinhold 1999). This hypothesis for an adaptive shift to paternally biased inheritance could provide an explanation for the evolution of paternal effects on the expression of male polyphenism in other taxa. A different explanation might be that alleles determining the sensitivity to environmental cues have pleiotropic effects on one or more other traits that are under sexually antagonistic selection, favoring their expression on a male limited chromosome (Stewart et al. 2010). Our data alongside those of Smallegange (2011) and Kotiaho et al. (2003) suggest that parental effects are an important aspect of the phenotypic and genetic architecture of polyphenisms and worthy of further investigation.

Associate Editor: M. Reuter

ACKNOWLEDGMENTS

We are extremely grateful to M. Penrose and T. Sanders for assistance in creating inbred lines used in this study, to the Centre for Evolutionary Biology at UWA for feedback, and to J. D. Fry and D. Dowling for discussions about the analysis of our data using the Cockerham–Weir model. This work was supported by Australian Research Council fellowships to JLT and LWS, and the University of Western Australia. BAB was funded by an International Postgraduate Research Scholarship, a C. F. H. and E. A. Jenkins Postgraduate Research Scholarship, and an Education Australia Limited Student Mobility Scholarship.

Ancillary