Environmental Plasticity in the Intersexual Correlation and Sex Bias of Gene Expression

Abstract

Intersexual genetic correlations are expected to constrain the evolution of sexual dimorphic traits, including the degree of sex-biased gene expression. Consistent with that expectation, studies in fruit flies and birds have reported that genes whose expression has a strong intersexual genetic correlation (r_MF) show a lower level of sex-biased expression (SBE). However, it is known that both r_MF and SBE can be affected by the environment. It is therefore unclear whether there is a consistent relationship between these 2 quantities across multiple environments. In this paper, we study this relationship in the African malaria mosquito Anopheles gambiae. We show that both r_MF and SBE change between environments. The change in SBE across environments is significantly correlated with d_N/d_S: greater changes in SBE are associated with higher values of d_N/d_S. Furthermore, the relationship between r_MF and SBE is sensitive to the environment. We conclude that this relationship is sufficiently plastic that environmental effects should be considered in future studies.

Introduction

Females and males share a genome but differ in almost every phenotype. The phenotypic differences between females and males are thought to result from selection acting differently on females and males (Fisher 1958; Darwin 1871; Arnqvist and Rowe 2005). Because the genome of males and females is shared, many traits show a positive intersexual genetic correlation constraining the independent evolution of the sexes (Lande 1980; Bonduriansky and Chenoweth 2009; Kirkpatrick 2009; Poissant et al. 2010). Sexual dimorphism in gene expression is no exception.

The intersexual genetic correlation (r_MF) quantifies the similarity of the effects of genetic variants on a given trait in males and females. If the effects are the same in males and females, r_MF is equal to 1, while if effects are independent in the 2 sexes, r_MF is equal to 0. Griffin et al. (2013) analyzed the intersexual genetic correlation for gene expression in 40 inbred lines of Drosophila melanogaster using data collected by Ayroles et al. (2009). They estimated a median intersexual genetic correlation of approximately 0.4 for those loci that have significant genetic variation for expression. This suggests that the evolution of sex biased gene expression in fruit flies may be constrained.

For a single trait, theory shows that sexual dimorphism can evolved when r_MF is less than 1 (Lande 1980). As the number of traits grows, however, the potential for evolutionary constraints increases rapidly (Hansen and Houle 2008; Walsh and Blows 2009; Kirkpatrick 2009). Consider the toy example of a set of n traits in which the genetic correlation between half the pairs is r, while in the other half it is –r. Then there will be an absolute genetic constraint (that is, the genetic covariance matrix becomes singular) when r = 1 / (n – 1). So while a constraint requires r = 1 for a pair of traits, with 10 traits a constraint occurs with r = 0.11, while with 100 traits a correlation of only r = 0.01 suffices. Thus, modest genetic correlations can place limits on adaptation. Even in the absence of an absolute constraint, selection in one sex generates a correlated response in the other, which can slow the rates at which each sex approaches its optimal phenotype.

Consistent with these theoretical facts, several studies have reported a negative correlation between the intersexual genetic correlation and the degree of sexual dimorphism for different types of traits including morphology, behavior, physiology (Bonduriansky et al. 2005; McDaniel 2005; Fairbairn et al. 2007; Poissant et al. 2010). Similarly, a negative correlation has been reported in Drosophila melanogaster between r_MF for gene expression and the degree of sex-biased gene expression (Griffin et al. 2013). A negative relation between an intersexual genetic correlation and sexual dimorphism therefore seems to be a quite general pattern.

However, this relationship between the intersexual genetic correlation and sexual dimorphism may be sensitive to the environment in which the traits are expressed. In general, genetic correlations can be environment-dependent because of genotype-by-environment interaction (Falconer and Mackay 1996; Lynch and Walsh 1998). Effects of environments on genetic correlations can be comparable to those that result from many generations of genetic divergence (Wood and Brodie 2015). Large environmental influences on r_MF have been reported for several traits and species (Lyons et al. 1994; Vieira et al. 2000; Punzalan et al. 2014; Berger et al. 2014). In some cases, the environment even changes the sign of an intersexual genetic correlation (Lyons et al. 1994; Punzalan et al. 2014).

The degree of sexual dimorphism can also change with the environment (Bonduriansky et al. 2005; Bonduriansky 2007). For example, sexual dimorphism in the wing and thorax sizes was reported to increase with development temperature in the fruit fly D. melanogaster (David et al. 1994). A study of D. melanogaster reported that about 10% of genes alter their degree of sex-biased expression depending on the diet on which individuals are raised (Wyman et al. 2010).

Given that r_MF and the degree of sexual dimorphism can both be affected by the environment, it is perhaps not surprising that the relation between these two variables can also be environment-dependent. Indeed, in Silene latifolia, the correlation between r_MF and sexual dimorphism ranges from a negative value at high density to a positive value at low density (Lyons et al. 1994).

In this study, we examine the effect of environment on the relationship between the intersexual genetic correlation for gene expression (r_MF) and sex bias in gene expression (SBE) using transcriptome data from the African malaria mosquito Anopheles gambiae. We find that SBE is environmentally dependent. Further, the relationship between r_MF and SBE is environmentally dependent. Genes that show greater plasticity in SBE across environments also have a greater d_N/d_S ratio. These results imply that the strength of genetic constraints placed by the shared genome varies across environments, as suggested in earlier studies (Lyons et al. 1994; Punzalan et al. 2014).

Methods

Gene Expression Profiling in Mosquitoes

Gene expression data were collected using both sexes of the five strains that were established from a population of An. gambiae s.s. in Cameroon, 2010 (Cheng 2014). For each environmental treatment, total RNA was extracted from four replicate pools of 20 adults. Samples were collected at the same age (one week old), and at the same time of the day, to minimize the effects of age and circadian rhythm. The accumulated mortality on the date of RNA extraction was less than 10% for both “dry” and “wet” conditions. We followed the protocol described in Cassone et al. (2011, 2014) for RNA processing, microarray data acquiring, and raw data quality control. The data was filtered and processed following Cassone et al. (2011, 2014): A transcript was tagged as not expressed if its expression values was in the bottom 25% of expression values across all genes and arrays; only expressed transcripts were used for subsequent analysis. The log₂ ratio of male and female gene expression was estimated by using the R Bioconductor packages (Gentleman et al. 2004). Sex bias is measured as:

$$\Delta =\left(M-F\right)/\left(M+F\right)$$

(Storer 1966), where M is the mean expression in males and F is the mean expression in females. Our index Δ is highly correlated with the commonly used log₂ ratio between male and female expression (r = 0.9).

Environmental Dependent Sex Bias in Gene Expression

The R package nlme was used to partition variation in gene expression using a linear mixed model with two main fixed factors and their interactions (sex, environment) and lines as a random factor. The significance of each factor was defined at a false discovery rate (FDR) of 0.05, for this and all other analyses unless otherwise specified.

Environment Dependent Relationship Between Intersexual Correlation and Sex Bias in Gene Expression

A permutation test was performed to test if the relationships between intersexual correlation and sex bias are the same between the 2 environments. The environment tags were shuffled 1000000 times, generating a distribution of the difference of correlation between intersexual correlation and the degree of sex bias in two environments. The P value was calculated using the frequency that the difference we observed was more extreme compared to the differences from simulated results.

Results

We compiled a large transcriptome data set collected in 2 different environments from malaria mosquito An. gambiae (Cheng 2014). The transcriptional abundance of 11490 genes was assayed in both female and male adults using Nimbelgen gene-chips. We included 5 strains of An. gambiae established from natural populations in Cameroon, West Africa. Gene expression was assayed in two conditions: 27 °C with 80% relative humidity (“wet”); and 30 °C with 60% relative humidity (“dry”). The two conditions simulate different climatic conditions experienced by the mosquitoes in their geographic range (Holstein 1954). The intersexual genetic correlations were calculated across the 5 strains.

Environment Dependent Sex Bias in Gene Expression

For each transcript, we used analysis of variance (ANOVA) to partition variation in abundance into effects associated with sex, environment, and the sex × environment interaction. We measured SBE as the difference between male and female expression, normalized by their average expression. We denote this measure as Δ. We estimated Δ and its associated P value separately for each environment.

The results are shown in Figure 1. Sex-biased expression is pervasive: 93% of expressed genes had a significant sex term at a false discovery rate (FDR) of 5%. Among these genes, 2982 were strongly biased (2-fold or greater difference in expression between sexes). Many of the genes we found to be strongly sex biased were reported by Baker et al. (2011) to specifically expressed in reproductive tissues. The strongly male biased genes were enriched for genes with tissue specific expression in the testes and male accessory glands (Fisher exact test, P < 0.002), while the strongly female biased genes were enriched for genes expressed in ovary (Fisher exact test, P = 6 × 10^–14). Strongly male biased genes were significantly under represented on the sex chromosome (0.2-fold, P < 2 × 10^–16), while strongly female biased genes were over represented on the sex chromosome (1.3-fold, P = 4 × 10^–6). Similar to previous results from D. melanogaster (Ayroles et al. 2009), we also found significant differences of overall transcript abundance among the chromosomes for both sexes (P < 2 × 10^–16), with the lowest average expression on the sex chromosome. Gene ontology terms and pathway analysis using the Database for Annotation, Visualization and Integrated Discovery (DAVID) annotation tool are provided in the Supplementary Materials.

Figure 1.

The effect of environment on the degree of sex bias of gene expression (|Δ|) on 11490 genes. Each point represents a transcript.

The environment had a strong effect on gene expression, as 31% of genes had a significant environment term (FDR = 5%; see the Supplemental Information for functional enrichment results). SBE is strongly affected by environment. Nearly half (41%) of the expressed genes show a significant interaction between sex and environment (FDR = 5%, see the Supplemental Information for functional enrichment results), with a significant under representation on the sex chromosome (0.7-fold, P = 0.0002). Among genes with a significant interaction term, 14% changed from significantly female biased in one environment (FDR = 5%) to significantly male biased in the other (FDR = 5%), or vice versa.

Characteristics of Genes With Environmentally Dependent SBE

Environmentally dependent SBE can evolve when selection favors different optimal expression levels in different environments. Alternatively, it might evolve by neutral processes (drift) if stabilizing selection favoring the same optimal expression level in males and females, but the strength of that selection is weaker in one environment than another. We hypothesize genes under stronger stabilizing selection for expression will tend to experience stronger purifying selection on their sequences. To test this idea, we quantified the strength of purifying selection using the ratio of substitution rates at non-synonymous and synonymous sites (d_N/d_S). We used values of d_N/d_S estimated by Neafsey et al. (2015) based on orthologous sequences in the An. gambiae species complex. (Similar results obtain when we use d_N/d_S values estimated by Neafsey et al. (2015) based on 16 more distantly related species.) The change in SBE between the two environments was quantified as the absolute value of the difference between Δ that we measured in each environment. We find that the change in Δ across environments is significantly correlated with d_N/d_S (correlation test, P < 0.001). Genes with a greater change in Δ have higher d_N/d_S. This result is consistent with the hypothesis whose sequences experience weaker purifying selection also experience weaker stabilizing selection on their expression levels. An alternative interpretation is presented in the Discussion.

Sex Bias Decreases With Increasing r_MF

The genome-wide average of sex-biased expression was |Δ| = 0.25 (S.E. = 0.0027) in the wet environment and 0.21 (S.E. = 0.0025) in the dry environment. The difference of |Δ| between the two environments is significant (P < 0.001, permutation test), suggesting most genes were less sex-biased in the dry environment than in the wet environment. The proportion of sex-biased genes (at a FDR = 0.05) is 82% in dry environments comparing to 89% in wet environment. (The proportions are lower than the overall fraction of 93% cited earlier because of slightly reduced power with smaller sample size when the sample is split between the 2 environments.)

Following Griffin et al. (2013), we estimated r_MF as the correlation in gene expression between males and females across our laboratory strains. These are therefore broad-sense genetic correlations, not additive genetic correlations (Falconer and Mackay 1996). The average value of r_MF increased from 0.24 (S.E. = 0.0051) in the wet environment to 0.34 (S.E. = 0.0051) in the dry environment (P < 0.001, permutation test). Thus, there is a stronger genetic constraint in the dry environment than the wet environment.

We then studied the relationship between r_MF and sex bias in gene expression separately for each environment (Figure 2). In the wet environment, there was a small but significant negative relationship between r_MF and |Δ| (correlation = −0.03, S.E. = 0.009, P = 0.002). In the dry environment, however, the relationship was not significant (correlation = 0.0002, S.E. = 0.009, P = 0.8). The correlations between r_MF and |Δ| differ significantly between the 2 environments (P = 0.034, permutation test).

Figure 2.

Linear regressions of intersexual correlation (r_MF) on the degree of sex bias of gene expression (|Δ|) based on 11490 genes. Results are plotted separately for the dry environment (upper line) and wet environment (lower line). Shaded areas show 95% confidence intervals. Individual data points were left out intentionally to highlight the trend lines. See online version for full colors.

In summary, the environment has a strong and genome-wide impact on the intersexual correlation in gene expression (r_MF), and on the degree of sex biased gene expression (|Δ|). The change in Δ across environments is significantly correlated with d_N/d_S, with greater changes in Δ associated with higher of d_N/d_S. There is a significant (albeit weak) effect of environment on the relation between the intersexual correlation (r_MF) and the degree of sex biased gene expression (|Δ|). In terms of genome-wide averages, a larger value of r_MF is associated with a smaller value of |Δ| across the two environments. This pattern is consistent with the idea that larger between-sex correlations may inhibit the evolution of adaptive sexual dimorphism.

Sexual Dimorphism in the Between-environment Genetic Correlation

Now, we view the results from a different perspective: the sexual dimorphism in the genetic correlation of expression across environments. This alternative compliments the analyses presented above. Just as we discussed earlier in the context of genetic correlations between the sexes, correlations between environments can constrain the evolution of optimal phenotypic plasticity in expression (Via and Lande 1985).

We estimated r_dw as the correlation in expression between the dry and wet environments across our laboratory strains (broad-sense genetic correlations, Falconer and Mackay 1996). The genome-wide average for r_dw is 0.32 ± 0.007 in males and 0.65 ± 0.004 in females. The difference of r_dw between males and females is significant (P < 0.0001, permutation test). Thus, there are much stronger genetic constraints on the evolution of plasticity in females than in males. The genome wide average absolute change in expression between the environments, denoted |Δ_e|, is 0.0445 ± 0.0004 in females, and 0.146 ± 0.001 in males; the difference between the sexes is significant (P < 0.0001, permutation test). In females, the correlation between r_dw and |Δ_e| is insignificant (r_dw = −0.01 ± 0.01, P = 0.25), but in males, the correlation between r_dw and |Δ_e| is negative and significant (r_dw = −0.15 ± 0.01, P < 2.2 × 10^–16).

In conclusion, there is significant effect of sex on the genetic correlation of expression across environments. The average value of r_dw is greater in females than males, while the average change in expression between environments, |Δ_e|, is less in females than males. This result suggests the evolution of plasticity in gene expression is more strongly constrained in females. The lack of a significant correlation between r_dw and |Δ_e| in females could reflect that constraint: perhaps many genes have not yet reached their optimum expression levels in the two environments.

Discussion

Intersexual genetic correlations have been used to quantify the constraints on the evolution of sexual dimorphism, resulting from the shared genome of females and males (Lande 1980; Brommer et al. 2007; Kirkpatrick 2009). Theory predicts that intersexual correlations in gene expression can constrain the evolution of sex-biased expression. Negative relationships between the intersexual genetic correlation and sex bias in gene expression have been observed in fruit flies and birds, and have been interpreted as evidence supporting the theory (Griffin et al. 2013).

Our results, however, show that the intersexual genetic correlation for gene expression (r_MF) and the degree of sex bias in gene expression (Δ) depend on the environment. Further, the relationship between those variables is environmentally dependent: the correlation between r_MF and |Δ| is significantly different in the two environments. This finding suggests that results from a single environment may be insufficient to characterize r_MF, |Δ|, and the relationship between those variables.

We find that the genome-wide average value of r_MF is significantly higher in the dry environment than in the wet environment. We also observed a significantly lower degree of sex bias, on average, in the dry environment. This suggests there might be more opportunity to evolve sex biased gene expression in the wet environment. In an alternative analysis, we considered how the genetic correlation in expression across environments differs between the sexes. Females show a much greater potential for evolutionary constraints in the evolution of adaptive plasticity than males.

We also find that the genes with greater plasticity of sex biased expression across environments tend to have higher d_N/d_S ratios. One hypothesis to explain this result is that loci whose sequences are under weaker purifying selection also experience weaker stabilizing selection for a consistent level of expression across environments. These genes therefore show greater plasticity in SBE across environments. Their greater plasticity could be the result of neutral evolution (that is, sex differences in expression have evolved by drift) or adaptation (that is, selection more often favors different expression levels for the genes). An alternative hypothesis is that genes with greater sex bias in expression tend to experience more positive selection. This suggestion is made plausible by the observation that reproductive proteins are often sex-limited in expression and often show evidence of rapid adaptive evolution (Swanson and Vacquier 2002).

The biggest limitation in our study that the results are based on only 5 strains of mosquitoes. Two other studies that have analyzed intersex correlations in transcription also used modest numbers of species or populations: Griffin et al. (2013), with 40 strains, and Dean and Mank (2016), with 6 populations. (The latter study is based on correlations across breeds and species, rather than genotypes within species, and so the results are not strictly genetic correlations.) These small numbers are mitigated by the much larger number of gene transcripts that were assayed (11490 in our case). Therefore, while our individual estimates of r_MF are imprecise, we expect the overall patterns across environments will be much more robust. Further, it is difficult to imagine that estimation errors associated with small sample size should have stronger effects in one environment than another. Therefore, we expect our conclusions regarding the relations between environments, r_MF, and Δ to be valid. Ultimately, this issue can only be resolved by studies using larger numbers of genotypes.

In summary, we show that the relationship between intersexual correlation (r_MF) and the degree of sex bias in gene expression (|Δ|) is environmentally dependent in the African malaria mosquito An. gambiae. Both r_MF and Δ vary with the environment. It is likely to be difficult to predict the impact of the environment on r_MF and Δ. The genetic constraints on the evolution of sex-biased gene expression, therefore, are not rigid, but plastic.

Supplementary Material

Supplementary data are available at Journal of Heredity online.

Funding

This work was supported by grants from the Swiss National Science Foundation (grant CRSII3-147625) and the National Institutes of Medicine (grant R01-GM116853) to M.K.