Beauty or function? The opposing effects of natural and sexual selection on cuticular hydrocarbons in male black field crickets

Although many theoretical models of male sexual trait evolution assume that sexual selection is countered by natural selection, direct empirical tests of this assumption are relatively uncommon. Cuticular hydrocarbons (CHCs) are known to play an important role not only in restricting evaporative water loss but also in sexual signalling in most terrestrial arthropods. Insects adjusting their CHC layer for optimal desiccation resistance is often thought to come at the expense of successful sexual attraction, suggesting that natural and sexual selection are in opposition for this trait. In this study, we sampled the CHCs of male black field crickets (Teleogryllus commodus) using solid‐phase microextraction and then either measured their evaporative water loss or mating success. We then used multivariate selection analysis to quantify the strength and form of natural and sexual selection targeting male CHCs. Both natural and sexual selection imposed significant linear and stabilizing selection on male CHCs, although for very different combinations. Natural selection largely favoured an increase in the total abundance of CHCs, especially those with a longer chain length. In contrast, mating success peaked at a lower total abundance of CHCs and declined as CHC abundance increased. However, mating success did improve with an increase in a number of specific CHC components that also increased evaporative water loss. Importantly, this resulted in the combination of male CHCs favoured by natural selection and sexual selection being strongly opposing. Our findings suggest that the balance between natural and sexual selection is likely to play an important role in the evolution of male CHCs in T. commodus and may help explain why CHCs are so divergent across populations and species.


| INTRODUC TI ON
Few biologists today would challenge the importance of sexual selection to the evolutionary process. There is good reason for this general agreement: there is an abundance of convincing evidence from theoretical models (e.g. Kirkpatrick, 1996;Lande, 1981Lande, , 1982Rowe and Houle, 1996), comparative studies (e.g. Arnqvist, 1998;Cooney et al., 2019;Wickman, 1992) and experimental evolution experiments conducted in the laboratory (e.g. Hawkes et al., 2019;House et al., 2013;Hunt et al., 2012) showing that sexual selection can drive the evolution of male sexual traits, often over relatively short time frames. Moreover, sexual selection gradients in the wild are stronger than natural selection gradients, suggesting that sexual selection is also likely to be an important evolutionary force in natural populations Kingsolver et al., 2001).
However, despite some evidence that sexual traits may evolve faster than nonsexual traits (Pitchers et al., 2014), examples of contemporary evolution by sexual selection in natural populations appear to be rare (Svensson & Gosden, 2007). Although numerous explanations have been provided to explain this paucity of examples, such as a regime of fluctuating selection (Siepielski et al., 2013; but see de Villemereuil et al., 2020) and/or genetic constraints (Hansen & Houle, 2004;Merilä et al., 2001;Pitchers et al., 2014), an oftenneglected explanation is how changes in the interaction between natural and sexual selection can alter how sexual traits evolve (Svensson & Gosden, 2007).
Most (if not all) sexual traits are also likely to be targeted by natural selection meaning that how these modes of selection interact is key to understanding how sexual traits will evolve (Svensson & Gosden, 2007). Indeed, the interaction between natural and sexual selection is built into most theoretical models of sexual selection, although the exact nature of this interaction varies across models (Mead & Arnold, 2004). Classically, it has been argued that sexual selection is opposed by natural selection, at least once male sexual traits have become sufficiently elaborated (Mead & Arnold, 2004).
This occurs because sexual selection favours the greater elaboration of male sexual traits and female preferences for them, but both become increasingly costly to their bearer, reducing their survival (Fisher, 1930;Kirkpatrick, 1982;Kirkpatrick & Ryan, 1991;Lande, 1981;Price et al., 1993). Natural selection will therefore act as an evolutionary 'brake' that prevents the continued evolution of male sexual traits once they pass their naturally selected optima (Fisher, 1930;Kirkpatrick, 1982;Lande, 1981). In contrast, good-genes models of sexual selection propose that natural and sexual selection are reinforcing, with both modes of selection favouring males with the highest fitness (Mead & Arnold, 2004). This argument is based on male sexual traits being honest signals of genetic quality (Zahavi, 1975) so that females preferring males with more elaborate sexual traits produce high-quality offspring that are both more attractive and have higher viability, thereby gaining indirect benefits from their mate choice (Iwasa et al., 1991;Iwasa & Pomiankowski, 1999;Kirkpatrick & Ryan, 1991). It is possible, however, that the interaction between natural and sexual selection will be more complex than this simple dichotomy, with the outcome varying across populations (Long et al., 2012) or with different environmental conditions (e.g. Parrett & Knell, 2018). Cuticular hydrocarbons (CHCs) are an excellent trait for studying the interaction between natural and sexual selection as this trait has a clearly defined role in both processes (Blomquist & Bagnères, 2010). CHCs are organic compounds that are deposited as a waxy layer on the surface of the cuticle in most terrestrial arthropods, and the total abundance and structural composition of these compounds have been shown to play a key role in reducing evaporative water loss (Blomquist & Bagnères, 2010). Indeed, a large number of laboratory studies have now shown that the total abundance and/or the proportion of longer-chained CHCs increase when individuals are maintained at a warmer temperature (e.g. Sharma et al., 2012;Wagner et al., 2001;Woodrow et al., 2000) and a lower humidity (e.g. Sprenger et al., 2018;Woodrow et al., 2000), as well as in populations artificially selected for desiccation resistance (e.g. Ferveur et al., 2018;Gibbs et al., 1997;Kwan & Rundle, 2009). Similar changes in CHC composition with temperature and humidity have also been shown across some natural populations (e.g. Buellesbach et al., 2018;Rajpurohit et al., 2017Rajpurohit et al., , 2020, but not all (e.g. Frentiu & Chenoweth, 2010;Leeson et al., 2020). Although far fewer studies exist, there is also direct evidence that an increase in longer-chained CHCs reduces evaporative water loss through the cuticle (e.g. Gibbs et al., 1997;Toolson, 1982). More recently, CHCs have also been shown to play an important role in sexual selection, especially in the context of female mate choice (Wyatt, 2003). Whereas in some species females prefer one or a small number of CHC components (e.g. Ferveur & Sureau, 1996;Grillet et al., 2006;Snellings et al., 2018), it is far more common for females to prefer a combination of different male CHCs Steiger et al., 2015). However, the exact combination of male CHCs that females prefer varies across species, with some preferring shorter-chained (and more volatile) male CHCs (e.g. Simmons et al., 2014;Steiger et al., 2013Steiger et al., , 2015, whereas others prefer more specific combinations that appear unrelated to chain length (e.g. Hunt et al., 2012;Rundle et al., 2005;Thomas & Simmons, 2009a).
Despite the independent effects of natural and sexual selection on male CHCs having been well documented, surprisingly few studies have directly examined the interaction between these two modes of selection. A notable exception to this is work on two species of Drosophila (Blows, 2002;Hine et al., 2011;Sharma et al., 2012;Skroblin & Blows, 2006). Blows (2002) used a factorial design to manipulate the intensity of natural and sexual selection in experimental populations of D. serrata to show that the evolutionary response of male CHCs was greater when natural and selection operated together compared to when they operated alone, suggesting that these modes of selection are reinforcing.
However, a subsequent multivariate selection analysis on breeding values for male CHCs in this species found that the direction of natural selection opposed the direction of sexual selection, at least for the subset of male CHCs examined (Skroblin & Blows, 2006). Furthermore, artificial index selection on the vector of male CHCs that are most attractive to females resulted in the rapid evolution of male CHCs for the first seven generations but further evolution beyond this point was halted presumably due to the opposing effects of natural selection (Hine et al., 2011). In a similar factorial design to Blows (2002), natural selection, sexual selection and their interaction were all shown to influence the evolution of male CHCs in D. simulans (Sharma et al., 2012). Importantly, some combinations of male CHCs only evolved in the direction of natural selection when sexual selection was relaxed, suggesting that these modes of selection are opposing (Sharma et al., 2012). Although the weight of evidence from these Drosophila studies suggests that the effects of natural and sexual selection on male CHCs are opposing, this outcome is not universal. One reason for this may be the different ways that natural selection has been applied (or measured) in these studies: in D. serrata, natural selection was measured (Skroblin & Blows, 2006) or manipulated (Blows, 2002) via female productivity, whereas natural selection was manipulated via temperature in D. simulans (Sharma et al., 2012). That is, none of these studies directly measured or manipulated evaporative water loss, which is the main proposed target of natural selection on CHCs. Consequently, how the interaction between natural and sexual selection shapes the evolution of male CHCs very much remains an open empirical question that requires more studies that directly examine evaporative water loss and encompass a broader range of arthropod species.
Field crickets have proved important models for testing sexual selection theory (Zuk & Simmons, 1997). By far, the majority of empirical research on field crickets has focussed on the male acoustic signal (or call), including what call properties females prefer (e.g. Bentsen et al., 2006;Brooks et al., 2005) and how they benefit from this choice (e.g. Ting et al., 2017;Wagner & Harper, 2003), as well as how this sexual signal is countered by natural selection through predation and attack by natural enemies (e.g. Hedrick, 2000;Sakaluk & Belwood, 1984;Wagner, 1996). In contrast, considerably less is known about the role that CHCs play in sexual selection, with our current knowledge limited to two field cricket species: the Australian field cricket (Teleogryllus oceanicus) and the decorated cricket (Gryllodes sigillatus). Females of both species prefer certain combinations of male CHCs to others and this exerts significant nonlinear sexual selection on this male sexual trait (Simmons et al., 2013;Steiger et al., 2015;Thomas & Simmons, 2009b). Male CHCs are heritable in both species (Thomas & Simmons, 2008;Weddle et al., 2012) and female T. oceanicus preferentially mate with males with a more dissimilar CHC profile to their own (and therefore less likely to be related; Thomas & Simmons, 2011) but a similar pattern does not occur in G. sigillatus (Steiger et al., 2015). Males and females physically transfer CHCs to each other during mating and are able to detect these subtle changes to adjust their mating behaviour (Capodeanu-Nägler et al., 2014;Thomas & Simmons, 2009a;Weddle et al., 2013). In G. sigillatus, females are able to recognize their own CHCs transferred to a male during copulation via a system of 'online processing' and use this information to avoid mating with previous mates (Capodeanu-Nägler et al., 2014;Weddle et al., 2013). In T. oceanicus, males are able to detect the CHCs from rival males on a female and adjust the proportion of viable sperm in their ejaculate in accordance with the risk of sperm competition (Thomas & Simmons, 2009a). In contrast, we know far less about the effects of natural selection on male CHCs in these species. A recent study on T. oceanicus showed a negative genetic correlation between the combinations of male CHCs that confer attractiveness and desiccation resistance (Berson et al., 2019).
While this suggests that the sexually and naturally selected functions of CHCs are opposing in this species, this study only included the seven most abundant CHCs (of the 22 possible CHCs for this species, Thomas & Simmons, 2009a) and because these were measured after attractiveness and desiccation resistance assays, the potential exists for these assays to directly influence male CHCs (Berson et al., 2019). Clearly, more work is still needed to understand how the interaction between natural and sexual selection has shaped male CHCs in these species, as well as in field crickets more generally.
Here, we examine the role that CHCs play in restricting evaporative water loss (EWL) and enhancing mating success in male black field crickets, Teleogryllus commodus. While sexual selection has been well studied in this species (e.g. Bentsen et al., 2006;Brooks et al., 2005;Bussière et al., 2006;Hall et al., 2008Hall et al., , 2013Hunt et al., 2004), we currently do not know if male CHCs play a role in this process or the extent to which CHCs are also shaped by natural selection. Using a custom-built desiccation chamber, we directly measured the EWL of a random sample of males. In a second random sample of males, we measured mating success using 'nochoice' mating trials. We used solid-phase microextraction (SPME) to sample the CHC profile of each male prior to these measurements to ensure that any possible contaminants or the physical transfer of CHCs did not influence our results. We conducted multivariate selection analysis (Lande & Arnold, 1983) on these data to characterize the strength and form of natural selection (acting via EWL) and sexual selection (acting via mating success) operating on male CHCs. We then formally compare these modes of selection to determine if natural and sexual selection on male CHCs are opposing or reinforcing in this population. We discuss how the interaction between natural and sexual selection is likely to shape the evolution of male CHCs in T. commodus, as well as the more general diversification of CHCs across populations and species.

| Animals and husbandry
The T. commodus used in this study were collected from the wild in March 2009 from Smith's Lake, New South Wales, Australia (32.3871° S, 152.4109° E), and used to establish a large laboratory culture. Approximately 400 gravid females were collected and placed into a single 90-L plastic container with cardboard egg carton for shelter, water in 50-mL test tubes plugged with cotton wool, cat biscuits for food (Purina Go Cat Senior©) and a total of eight eggs pads for oviposition. Each egg pad consists of moist cotton wool provided in a Petri dish (90 mm diameter). Each week we removed the egg pads, express couriered them to our insect facility at the University of Exeter (Cornwall Campus) and replaced them with fresh egg pads.
This process was repeated for three consecutive weeks.
Nymphs were collected from egg pads on the day they were hatched and distributed at random between four large 110-L plastic culture containers. Each container was provided with an abundance of cardboard egg carton for shelter, water ab libitum in 50-mL test tubes plugged with cotton wool and a 50% mixture of cat biscuits (Purina Go Cat Senior©) and rat pellets (SDS Diets).
Culture containers were housed in a constant temperature room set to 28 ± 2°C and a 13 h light: 11 h dark cycle. Culture containers were cleaned and fresh food and water bottles were provided weekly. When newly eclosed were observed in each culture container, eight egg pads were added for oviposition. Once sufficient nymphs were collected to establish four new culture containers (~2000 nymphs per container), adults were killed by freezing at −20°C to prevent overlapping generations. To preserve genetic variation in our culture, nymphs were distributed at random between culture containers each generation to enforce gene flow, and the number of breeding adults in each culture container was always kept high (~500 crickets). At the time of our experiment, crickets had been maintained according to this protocol for a total of 12 generations.

| Experimental procedure
A total of 2000 nymphs were taken at random from our culture on the day they hatched from eggs and established in individual containers (5 × 5 × 5 cm) provided with a single piece of cardboard egg carton for shelter, a small 5 mL tube plugged with cotton wool for water and ground cat biscuit provided in the lid of a 1.5 mL Eppendorf for food.
Each container was cleaned and fresh food and water were provided weekly. After 3 weeks, we replaced the ground found with two cat biscuits per cricket. When crickets reached fourth instar, containers were checked daily for eclosion to adulthood. These crickets were maintained in the same constant temperature room, and therefore the same temperature and light conditions, as our cultures.
On each day of eclosion, half of the males were randomly allocated to measure mating success and the remaining half to measure EWL. For both mating success and EWL, adult males were measured at 8 days of age and because they were reared in individual containers, were all virgin and had not interacted physically with other crickets at the time of measurement. A random subset of the adult females that were reared were used to assess male mating success and these females were also 8-day-old virgins and socially naive when used. In total, we measured the EWL of 300 adult males and the mating success of an additional 300 adult males (total n = 600 males).

| Measuring male evaporative water loss
We used a custom-built device that enabled us to measure the EWL of eight males simultaneously. This device consisted of laboratorygrade compressed zero air (21%O 2 and 79%N 2 mix; BOC) from a cylinder passed through two glass columns (7 cm diameter, 29 cm tall; Drierite®), using Tygon® S3™ laboratory-grade tubing (E-3603). The first column contained indicating drierite (Drierite®) to remove any water and the second contained activated charcoal (Finest-Filters®) to remove any volatile organic compounds. The outlet of the second column was connected to a stainless steel eight-way airline splitter (One Stop Grow Shop), with each outlet connected by tubing to an independently calibrated flow meter (MR300 Series, 2-30 L/ min, Brooks® Instrument). In turn, each flow meter was connected by tubing to a plastic holding vial (85 mm long, 28 mm diameter). A 9.5 mm hole was drilled in the bottom of the vial to serve as an inlet, with the tubing secured in place with silicon sealant. Half of the internal diameter of the screwcap lid was removed and replaced with wire mesh (size 12 mesh, 1.6 mm opening) that was also secured in place with silicon sealant. Each holding vial was housed in an incubator (Sanyo MIR 553) set to 28 ± 2°C with reduced florescent lighting to minimize movement.
On the day of testing, we ensured that the airflow to each holding vial was set to 10 L/min. We then sampled the CHCs of each male using SPME and weighed them to the nearest milligram on an electronic microbalance (UMX2; Mettler Toledo). Each male was then introduced to one of the holding vials at random and kept there for 2 h to measure EWL. We used this time period as our pilot data showed that males exhibited the greatest rate of EWL in the first 2 h of measurement ( Figure 1a). Importantly, we also found that males with the highest EWL in the first 2 h were significantly more likely to die when returned to the holding vial for a further 6 h (Logistic regression: χ 2 (1) = 44.65, p = 0.0001; Nagelkerke R 2 = 0.83; 90% of cases correctly classified; Figure 1b). After 2 h, each male was removed from the holding vial and reweighed. We used the reduction between the initial and final weight as our measure of EWL for each male. To account for the variation in male size, we expressed this weight change as a percentage of the initial weight of the cricket for analysis.

| Measuring male mating success
We measured the mating success of each adult male in noncompetitive ('no-choice' mating trial) male-female pairs (see Hall et al., 2008;Shackleton et al., 2005). We sampled the CHCs of each male using SPME and then allocated a female at random to the male. The male from each pair was then introduced into a plastic container (20 × 10 × 10 cm) with the bottom lined with paper towel and given 5 min to acclimate. After acclimation, we introduced the female into the container. When the pair had made antennal contact, we commenced timing the observation, and the pair was given 2 h to mate.
If the male performing the full courtship repertoire (i.e. producing a courtship call while positioning his rear end to the female) was mounted by the female and transferred a spermatophore to the female, the mating was considered successful and the male was assigned a score of 1. However, if the male courted and was mounted by the female but he did not transfer a spermatophore, the mating was considered unsuccessful and the male assigned a score of 0. If the male did not court the female in the 2 h provided, he was tested the following day with a different female. If this male did not court three consecutive females, he was excluded from the experiment and replaced with another male. This occurrence was rare, however, with only 9 of 300 males (3%) needing to be replaced in this manner.
In T. commodus, less than 5% of males that fail to mate in 2 h successfully mate if given a further 2 h (J. Hunt, personal observations), indicating that our observation period is sufficient time to accurately assess male mating success.
All mating behaviour was observed under red lighting in a constant temperature room set to 28 ± 1°C during the dark phase of the light cycle.

| Analysis of male cuticular hydrocarbons
Immediately prior to measuring EWL and mating success, we sampled the CHCs of each male using SPME. Crickets were sampled by lightly rubbing a 7 μm polydimethylsiloxane (Supelco) fibre across the dorsal surface of the pronotum and fore wings continuously for a 1-min period.
Each SPME fibre was manually injected into an Agilent 7890A GC coupled to an Agilent 5975B mass spectrometer equipped with an HP-5 MS capillary column (30 m × 0.25 mm ID × 0.25 μm; Agilent J&W). Fibres were injected into a split/splitless inlet and held at 250°C in splitless mode for 1 min. The helium carrier gas flow was 1 mL/min. The initial oven temperature was held at 50°C for 1 min, then ramped at a rate of 20°C/min to 250°C followed by a 4°C/min ramp to 320°C and a 5 min hold at this temperature. Ionization was achieved by electron ionization (EI) at 70 eV. The quadrupole mass spectrometer was set to 3.2 scans/s, ranging from m/z 40 to 500.
The abundance of each CHC peak in chromatograms was estimated using MSD ChemStation software (version E.02.00.493; Agilent Technologies) by measuring the area under the peak, using ion 57 as the target ion ( Figure 2). CHCs were identified using NIST library matches provided in MSD ChemStation.
Prior to analysis, we divided the abundance of each CHC peak by the abundance of peak 1 (a methyl alkane, Figure 2), and the resulting value was log 10 -transformed (to produce a log contrast for each CHC peak) to achieve a normal distribution. This meant that although we identified 45 unique CHC peaks for T. commodus ( Figure 2, Table 1), only 44 of these (peaks 2-45) were available for further analysis.

| Statistical analysis
Due to the large number of CHCs examined for T. commodus, we used principal component (PC) Analysis to reduce the dimensionality of this data set. PCs were extracted from males and used to measure F I G U R E 1 Pilot data showing that (a) males had the greatest rate of evaporative water loss in the first 2 h of measurement. Different letters represent statistically significant differences (at p < 0.05) across sampling intervals. (b) Males with the highest percentage of water loss in the first 2 h of measurement were significantly more likely to die if returned to the desiccation device for a further 6 h. Individual data points for males are given by grey circles. The solid line represents a thin-plate spline through the data and the dashed lines represent the 95% confidence interval for this spline.
EWL and mating success together to ensure that PCs were directly comparable. PCs were extracted using the correlation matrix and we retained PCs with eigenvalues exceeding 1 for further analysis (Tabachnick & Fidell, 1989). We interpret factors loading that exceeds |0.25| as biologically important (Tabachnick & Fidell, 1989).
We used standard multivariate selectin analysis (Lande & Arnold, 1983) to evaluate the strength and form of linear and nonlinear selection acting on male CHCs through EWL and mating success. As increased EWL reduces survival, we refer to the selection acting through this mechanism as natural selection. Conversely, as increased mating success is likely to improve reproductive fitness, we refer to the selection acting through this mechanism as sexual selection. Each male was assigned an absolute fitness score: for natural selection, we used the percentage water loss and for sexual selection we used mating success (1 = mated, 0 = not mated). Following Lande and Arnold (1983), we transformed absolute fitness into relative fitness by dividing by the mean absolute fitness of the population. As high values of EWL and mating success are likely to have opposite effects on fitness, we reversed the sign of relative fitness for EWL (larger values mean lower EWL) to facilitate the direct comparison of natural and sexual selection on male CHCs.
To estimate the standardized linear selection gradients for natural and sexual selection (ß), we fit a first-order linear multiple regression that used the three PCs that described the variation in male CHCs as the predictor variables and relative fitness as the response variable (Lande & Arnold, 1983). We then used a second-order quadratic multiple regression model that included all the linear, quadratic and cross-product terms to estimate the matrix of nonlinear selection gradients (γ) for natural and sexual selection. As standard multiple regression analysis underestimates the quadratic regression coefficients by 0.5, we doubled the standardized quadratic selection gradients from this model (Stinchcombe et al., 2008).
Relative fitness did not conform to a normal distribution and although this does not influence the sign and magnitude of the resulting selection gradients (Lande & Arnold, 1983), it can impact the significance testing of these gradients (Mitchell-Olds & Shaw, 1987).
We, therefore, tested the significance of all standardized selection gradients using a resampling procedure where we randomly shuffled relative fitness scores across males in our dataset to obtain a null distribution for each selection gradient where there is no relationship between our PCs describing the variation in male CHCs and relative fitness. We used a Monte Carlo simulation to determine the proportion (p) of times (of 10 000 iterations) that each gradient pseudoestimate was equal to or less than the original estimated gradient, and this was used to calculate a two-tailed probability value (as 2p if P < 0.5 or as 2(1-p) if P > 0.5) for each selection gradient in the model (Manly, 1997). We conducted separate randomization tests for the linear and full quadratic model for natural and sexual selection following the procedure outlined above. We used univariate splines to visualize the linear and nonlinear natural and sexual selection acting on each of the three PCs using the 'SPLINES' package of R (version 4.1.1, www.r-proje ct.org).
As the strength of nonlinear selection can be underestimated by interpreting the size and significance of γ (Blows & Brooks, 2003), we examined the extent of nonlinear selection acting on the PCs describing the variation in male CHCs by conducting a canonical rotation of the γ to locate the major eigenvectors of the fitness surface for natural and sexual selection (Phillips & Arnold, 1989).
For both fitness surfaces, we used the permutation procedure outlined in Reynolds et al. (2010) to locate and determine the strength F I G U R E 2 A chromatograph of a typical cuticular hydrocarbon (CHC) profile of male Teleogryllus commodus. All peaks were present in each individual male but in different relative amounts. The numbers above peak correspond to the peak number provided in Table 1. Peak 1 (a methyl alkane) was used as a divisor for all other peaks to generate logcontrasts for analysis and therefore is not present in Table 1. and significance of nonlinear selection operating along the eigenvectors of γ for natural and sexual selection. We adapted the permutation procedure of Reynolds et al. (2010) to also estimate the strength and significance of linear selection operating along the eigenvectors of γ for natural and sexual selection (R code provided in Text S1). The strength of linear selection along each eigenvector (m i ) is given by theta (θ i ), whereas the strength of nonlinear selection is given by their eigenvalue (λ i ) (Phillips & Arnold, 1989).
We used thin-plate splines (Green & Silverman, 1994) to visualize the major eigenvectors of the fitness surface for natural and sexual selection. We used the Tps' function in the 'FIELDS' package of R to fit the thin-plate splines and visualized the splines in both the perspective and contour views. In each instance, we visualized the thin-plate splines using the smoothing parameter (λ) that minimized the generalized cross-validation score (Green & Silverman, 1993).
We used a sequential model building approach to determine whether the linear and nonlinear selection targeting male CHCs differed for natural and sexual selection (Draper & John, 1988).   Formal comparison showed that the linear and quadratic gradients differed significantly between natural and sexual selection, but the correlational selection gradients did not (Table 4). The significant difference in the standardized linear selection gradients was due to the fact that the linear gradient for PC1 was positive for natural selection and negative for sexual selection (Figure 3a,d) and because the linear gradient for PC3 was negative for natural selection and positive for sexual selection (Figure 3c,f; Table 3). The significant difference in the standardized quadratic gradients is due to the gradient for PC1 being more negative for sexual selection (indicating stronger stabilizing selection) than for natural selection (Figure 3a,d; Table 3). Therefore, although natural and sexual selection both impose significant linear and stabilizing selection on male CHCs, they appear to be targeting very different combinations of these traits.

| RE SULTS
Indeed, the angle (θ) between the linear vectors (β) of selection for natural and sexual selection was 126.90° (95% credible interval: 103.70°, 156.90°), demonstrating that these modes of selection on male CHCs are strongly opposing in this population (Table 4).

| DISCUSS ION
Although the interaction between natural and sexual selection features prominently in most models of sexual selection (Mead & Arnold, 2004), surprisingly few direct empirical tests of this interaction exist. Empirically testing this interaction has proven difficult because it is not always easy to quantify how each mode of selection targets a given phenotypic trait. Cuticular hydrocarbons (CHCs) are widespread in terrestrial arthropods and represent a 'dual trait' that has clear functions in preventing evaporative water loss (EWL) and also as a chemical cue that operates in many different social contexts, including sexual interactions that directly influence male mating success (Chung & Carroll, 2015). In this study, we characterize the strength and form of natural selection (acting via EWL) and sexual selection (acting via mating success) operating on male CHCs in the black field cricket (Teleogryllus commodus).
We show that EWL was reduced when there was an increase in the total abundance of CHCs, especially those with a longer chain length. In contrast, mating success was highest at a low total abundance of CHCs, with the exception of a few specific CHCs (six peaks in total) that increased mating success. Importantly, natural and sexual selection acting on male CHCs was strongly opposing, with a large angle between the linear vectors of selection (126.90°). Our findings therefore suggest that balance between natural and sexual selection is likely to play an important role in the evolution of male CHCs in T. commodus and that this interaction may help explain why CHCs are so divergent across populations and insect species.
By far, the majority of studies examining the relationship between CHCs and EWL in insects have been indirect. That is, most studies have altered EWL by manipulating temperature (e.g. Sharma et al., 2012;Wagner et al., 2001;Woodrow et al., 2000) or humidity (e.g. Sprenger et al., 2018;Woodrow et al., 2000), shown an increase in the total abundance and/or the proportion of longer-chained CHCs in response to warmer temperature, lower humidity and with desiccation resistance, the possibility that these manipulations are influencing CHCs beyond EWL cannot be ruled out. Reassuringly, studies that have directly measured EWL across the cuticle have largely confirmed these patterns: an increase in longer-chained CHCs reduces EWL through the cuticle (e.g. Toolson, 1982;Gibbs et al., 1997). Our finding that EWL was reduced at high values of PC1 (more total CHCs) and low F I G U R E 4 Thin-plate spline visualizations provide a perspective (a and c) and contour (b and d) view of the two major axes of nonlinear natural (a and b) and sexual (c and d) selection (m 2 and m 3 ). On each surface, white colouration represents regions of highest fitness, whereas red colouration represents regions of lowest fitness. Individual data points are provided as black circles on the contour views.
values of PC2 (more long-chained and less short-chained CHCs) is therefore consistent with these earlier studies. Importantly, our work builds on these earlier studies by providing the first quantitative estimates of linear and nonlinear natural selection acting on male CHCs through EWL. Our estimates of linear natural selection acting on PC1 and PC2 were markedly lower than the median (|ß| = 0.16) reported for natural populations, whereas our estimate of quadratic natural selection acting on PC2 was similar (|γ| = 0.10) and generally considered weak .
Collectively, this demonstrates that the natural selection we document on these two vectors is relatively weak. We also show significant (albeit weak) negative linear natural selection on PC3 which represents a trade-off between specific CHCs, independent of carbon chain length. Understanding how PC3 influences EWL is more speculative, but it is interesting that nine of the ten CHCs that weigh heavily on this vector contain either methyl (peaks 4, 13, 14 and 16) or dimethyl (peaks 3, 7, 8, 15 and 29) groups. The presence of methyl branches is known to lower the melting temperature (and therefore increase cuticular permeability and EWL) of CHCs because molecular packing is less tight Menzel et al., 2019). Furthermore, the position of methyl branches is also important with CHCs containing methyl groups located more centrally melting earlier than those with methyl groups located more distally (Gibbs & Pomonis, 1995). More work is clearly needed on the chemical structure of the CHC components contributing to PC3 before we understand how this vector influences EWL in T.

commodus.
Our work shows that male CHCs in T. commodus are also targeted by sexual selection imposed by female mate choice. The role of CHCs in mate choice is widespread in insects (e.g. Chung & Carroll, 2015;Steiger & Stokl, 2014), and for most species, females tend to prefer certain combinations of CHCs over others rather than exhibiting a preference for one or a few specific CHCs (but see Ferveur & Sureau, 1996;Grillet et al., 2006;Snellings et al., 2018).
However, exactly what combination of CHCs females prefer is highly variable across species. In some species, females prefer combinations of shorter-chained CHCs that are more volatile (e.g. Simmons et al., 2014;Steiger et al., 2013Steiger et al., , 2015,  (Bussière et al., 2006;Shackleton et al., 2005). Inspection of the individual data points along the major axis of nonlinear sexual selection (m 3ss , Figure 3d), which is most heavily weighted by PC1, shows that male mating success only decreases at the very highest PC1 scores (i.e. where males are largest). Given that males were paired with a female at random in our study, it is also possible that this reduction in mating success with an increase in PC1 occurs due to a mismatch in size between the sexes (e.g. Han et al., 2010). It is more difficult to interpret the effects of PC3 on mating success and clearly work is needed, possibly using electroantennography or single sensillum recordings (e.g. Jacob, 2018) to determine how individual male CHC components stimulate the female olfactory system.
A key finding of our work is that natural selection and sexual selection acting on male CHCs are strongly opposed in T. commodus.
This was confirmed by the significant differences in our sequential model and the large angle (126.90°) between the linear vectors of natural and sexual selection and indicates that males cannot have a CHC profile that is optimal for both mating success and evaporative water loss. This finding is therefore broadly consistent with previous studies on CHCs showing that natural selection opposes sexual selection (Hine et al., 2011;Sharma et al., 2012;Skroblin & Blows, 2006), as well as a number of iconic studies in sexual selection including the opposing effects of predation on the evolution TA B L E 4 Sequential model building approach used to statistically compare the sign and strength of standardized linear, quadratic and correlational for the natural and sexual selection acting on male CHCs in T. commodus. When an overall significance was detected, univariate interaction terms are provided (below of male colour patterns in guppies (Endler, 1980) and calling in the Túngara frog (Ryan et al., 1992). Our results do, however, contrast the findings of Blows (2002) that showed the effects of natural and sexual selection on the evolution of male CHCs were reinforcing in D. serrata. It is important to note that with the exception of Skroblin and Blows (2006), all of these previous studies have examined the evolutionary response of CHCs to different regimes of natural and sexual selection (Blows, 2002;Sharma et al., 2012) or artificial selection (Hine et al., 2011) rather than directly quantifying the strength and form of each mode of selection targeting CHCs.
Moreover, Skroblin and Blows (2006) did not directly measure natural selection acting on male CHCs through EWL (but rather indirectly through male productivity) and they did not formally estimate the degree of divergence between these two modes of selection. Our work is therefore novel by directly quantifying both natural and sexual selection targeting male CHCs, as well as the degree to which these modes of selection are opposing for this trait in T. commodus.
However, understanding how the opposing natural and sexual selection we document shapes the overall pattern of selection on male CHCs requires more information on the relative contribution of EWL and mating success to total male fitness (e.g. lifetime reproductive success; Hunt et al., 2009). Given that unit changes in EWL and mating success are unlikely to have equivalent effects on total fitness, empirically quantifying these effects will be an important first step in understanding the broader implications of our findings to evolution of male CHCs in T. commodus.
CHCs are some of the most highly divergent traits across insect populations and species (e.g. Kather & Martin, 2012;Menzel et al., 2019;Otte et al., 2018) and our findings suggest that the balance between natural and sexual selection may play an important role in explaining some of this diversity. Whenever natural and sexual selection targets the same sexual trait but acts in opposing directions, the trait optimum will be determined by the balance between these two modes of selection (Svensson & Gosden, 2007). In this case, the most obvious effect of sexual selection will be to push a population away from the mean sexual trait optimum determined by natural selection (Kirkpatrick, 1982;Lande, 1981). Although this will temporarily reduce local adaptation in a single population, theory suggests that there are several possible ways that this can promote divergence between allopatric populations and potentially drive reproductive isolation (Servedio & Boughman, 2017). First, it is possible that as sexual selection pushes the mean sexual trait away from one naturally selected peak, it moves into a broad zone of instability between alternate peaks. On entering this unstable region, the combined action of natural and sexual selection can drive the rapid evolution of the mean sexual trait across this region to a new naturally selected peak, resulting in ecological divergence (Bonduriansky, 2011;Lande & Kirkpatrick, 1988;Miller, 1994). Second, the interaction of natural and sexual selection with genetic drift can promote the rapid evolution of preference and sexual traits in geographically separated populations (Lande, 1981), resulting in reproductive isolation when preferences are either neutral or costly (Uyeda et al., 2009). Third, if preference landscapes are rugged (i.e. have multiple peaks), it is possible that populations may evolve to different sexually selected peaks as novel sexual traits emerge, even when these populations initially experience similar natural and sexual selection (Mendelson et al., 2014). Although the occurrence of gene flow poses more challenges (by potentially bringing maladapted migrants into the population), theory suggests that sexual trait divergence can still evolve across sympatric populations if preference is relative (to the population mean sexual trait), open ended (Lande, 1982) or based on a condition-dependent sexual trait that indicates locally adapted males (Proulx, 2001). However, sexual trait divergence across sympatric populations is most likely to occur when preferences are under direct selection, as occurs when preferences become locally adapted through sensory drive (Endler, 1992), are based on context-dependent benefits (Cornwallis & Uller, 2010) or are directed towards a trait that is also possessed by the female (i.e. phenotyping matching; Kirkpatrick, 2000;Servedio, 2011). Despite the many theoretical conditions that can promote the diversification of sexual traits under opposing regimes of natural and sexual selection, relatively few empirical tests currently exist (Svensson & Gosden, 2007). Consequently, there is a clear need for more empirical studies and the dual function of CHCs in reducing EWL and enhancing mating success, making this trait an excellent model for future work. We know that male CHC profiles in T. commodus are genetically divergent across populations in southern Australia but we do not know the role (if any) that the balance between natural and sexual selection plays in shaping this divergence (C. Mitchell & J. Hunt, unpublished data). An obvious first step is therefore to formally quantify natural and sexual selection targeting CHCs in these populations and determine if any differences in these modes of selection are related to CHC divergence across populations. A similar approach including other Australian field cricket species within a phylogenetic context could be used to understand if changes in the balance between natural and sexual selection can drive speciation, although this is likely to prove more challenging given how rapidly individual CHCs components appear to evolve in arthropods (e.g. .

AUTH O R CO NTR I B UTI O N S
CM and JH conceptualized the work. CM, JR and JH conducted the experimental work and data collection. JH and EDC conducted the formal analyses. ZW, CMH and JH wrote the original draft. All authors contributed to the final version of the manuscript.