Female mate choice of male signals is unlikely to promote ecological adaptation in Enchenopa treehoppers (Hemiptera: Membracidae)

Abstract A key question in speciation research is how ecological and sexual divergence arise and interact. We tested the hypothesis that mate choice causes local adaptation and ecological divergence using the rationale that the performance~signal trait relationship should parallel the attractiveness~signal trait relationship. We used female fecundity as a measure of ecological performance. We used a species in the Enchenopa binotata treehopper complex, wherein speciation involves adaptation to novel environments and divergence in sexual communication. We used a full‐sibling, split‐family rearing design to estimate genetic correlations (r G) between fecundity and signal traits, and compared those relationships against population‐level mate preferences for the signal traits. Animal model estimates for r G between female fecundity and male signal traits overlapped zero—rejecting the hypothesis—but could reflect sample size limitations. The magnitude of r G correlated with the strength of the mate preferences for the corresponding signal traits, especially for signal frequency, which has the strongest mate preference and the most divergence in the complex. However, signal frequencies favored by the population‐level mate preference are not associated with high fecundity. Therefore, mate preferences do not appear to have been selected to favor high‐performance genotypes. Our findings suggest that ecological and sexual divergence may arise separately, but reinforce each other, during speciation.

Mate choice on different environments could then promote differential local adaptation and hence ecological divergence Lorch et al., 2003).
There is mixed support for this hypothesis. Sexual displays only sometimes show the predicted elevated levels of conditiondependence (Cotton, Fowler, & Pomiankowski, 2004;cf. Koch, Josefson, & Hill, 2017). And, artificial selection experiments in which sexual selection is allowed only sometimes result in enhanced adaptation to novel environments (Coyne & Orr, 2004). A potential weakness of some of these tests is not taking into account that only a subset of sexual traits is expected to evolve high levels of conditiondependence-for example, because of variation in their scaling with body size and in their degree of sexual dimorphism (Bondurianky & Rowe, 2005;Bonduriansky, 2007;Eberhard, Rodríguez, & Polihronakis, 2009). An additional problem may lie in the use of body condition as a measure of adaptation to an environment. Conditionthe resources acquired by an individual that can be allocated to any one function or trait (Hunt, Bussière, Jennions, & Brooks, 2004)should covary with performance on local environments. However, even when sexual displays are costly, they may mainly reflect the ability to operate at high levels of performance for relatively brief periods of time, which may be unconnected from the size of the resource pool available (Clark, 2012).
We consider that to relate mate choice of ornaments to local adaptation, the key is to focus on the fecundity of the choosing females and their daughters. This is a version of the hypothesis that male sexual ornaments are selected to indicate the quality of the daughters that males would produce if accepted as mating partners (Trivers, 2002;cf. Miller & Moore, 2007). The key question is therefore whether female fecundity covaries genetically with male ornaments. We do not hold that daughter fecundity will outweigh the attractiveness and/or viability of male offspring or other sources of selection on mate choice (Kokko, Brooks, McNamara, & Houston, 2002). On the contrary, we consider that sexual selection on mate choice may often be predominant (Prum, 2012(Prum, , 2017West-Eberhard, 1983, 2014. However, for the hypothesis that mate choice promotes local adaptation and ecological divergence, it does seem to us to be the most relevant measure. We therefore used female fecundity as a measure of the ecological performance of different genotypes to test the hypothesis that mate choice causes local adaptation. We generated predictions according to the rationale that there should be a relationship between genetic variation in performance and advertisement signals, and that the performance~signal trait function should parallel the attractiveness~signal trait function. We note that mate choice may promote divergent ecological adaptation without preferences for ornaments, if females focus directly on ecological traits (Byers & Waits, 2006;Reinhold, 2004). We focus on the widespread scenario of mate choice of advertisement signals (Andersson, 1994).
We list the predictions that arise from the above rationale in Table 1. Prediction (i) is a prerequisite for testing all the other predictions, rather than a logical part of the hypothesis. Predictions (ii)-(ix) articulate the rationale. Predictions (viii) and (ix) refer to comparative tests with closely related species-we did not test them because of lack of support for the preceding predictions, but we list them here for completeness and to encourage further studies.
We tested the predictions with a species belonging to the Enchenopa binotata complex of treehoppers (Hemiptera: Membracidae). We used a quantitative genetics experiment to examine the genetic relationship between female fecundity and male signal traits. We then used playback experiments to describe population-level female mate preferences for those signal traits.
The E. binotata complex is a clade of plant-feeding insects that is widely distributed across eastern North America, with each treehopper species occurring on its own host plant species Wood, 1993). Enchenopa communicate with plantborne vibrational signals, and signal variation across the complex is mostly associated with differences in host plant species, rather than geographic distance (Cocroft, Rodríguez, & Hunt, 2010). Thus, signal divergence in the complex has occurred through changes in selection on signals associated with the colonization of novel host plants.
Sources of selection on signals that vary across host plants include divergent mate preferences and plant signal-transmission properties (McNett & Cocroft, 2008;Rodríguez, Ramaswamy, & Cocroft, 2006). Of these, mate preferences seem to make the stronger contribution to realized mate choice decisions and, consequently, to selection on signals Sullivan-Beckers & Cocroft, 2010). Pair formation in Enchenopa involves male-female duets (Cocroft & Rodríguez, 2005;Cocroft et al., 2008). Males initiate the duet by producing advertisement signals, and if a female finds a male's signals attractive, she signals back, prompting him to search for her on the plant . Whether a female responds to a male's signals influences the likelihood of her mating with him. Females thus express their mate preferences in selective duetting with males (Rodríguez, Sullivan, & Cocroft, 2004;. The strongest mate preferences in the E. binotata complex are for male dominant signal frequency , and this is the most distinctive signal trait among the members of the complex ; consequently, the strongest genetic correlation (r G ) should be between T A B L E 1 Predictions of the hypothesis that mate choice of male sexual ornaments promotes local adaptation and ecological divergence. Prediction (i) is a prerequisite for testing the hypothesis, more than a logical requirement. Predictions (ii)-(ix) articulate the rationale that there should be a relationship between genetic variation in performance and signal traits, and that this relationship should be parallel to the function relating attractiveness to those signal traits (see text). We state the predictions in general, and we also refine them with information about the mating system and mate preferences of our study species, a member of the Enchenopa binotata complex of treehoppers . We also summarize the results of our experiment indicating the presence or absence of support for the predictions (see Section 3) fecundity and signal frequency (Table 1). Similarly, detailed study of the shape of female mate preferences  results of current study) allows us to specify the shape that the fe-cundity~signal trait functions should have for different signal traits, and where along variation in different signal traits those functions should peak (Table 1).
We also examined corollary factors that could influence the ability to test the predictions. For r G to exist between female fecundity and a male signal trait, there should be some genetic variation in fecundity and the signal trait (prediction i). But, genetic variation may be lower in signal traits with strong mate preferences that exert stronger selection.
We therefore tested for a relationship between the strength of mate preferences and the amount of genetic variation in the corresponding signal traits; and, we also tested for a relationship between the amount of genetic variation in signal traits and the magnitude of r G between female fecundity and the signal traits.

| MATERIALS AND METHODS
Most of the species in the E. binotata complex have not been formally described (Hamilton & Cocroft, 2009), but can be recognized by the host plants they use and the signals of the adult males. We worked with the species that lives on Ptelea trifoliata host plants (Rutaceae).
We kept voucher specimens in 75% ethanol in the Rodríguez laboratory collection.

| Experiment 1: Quantitative genetics of female fecundity and male signals
To test for a correlation between the signals and fecundity of siblings-between brothers' signals and sisters' fecundity-we used a full-sibling, split-family rearing design (Lynch & Walsh, 1998 We recorded the advertisement signals of the males when they reached sexual maturity, ca. 2-3 weeks after the adult molt (vibrational recording procedure below). And, we used vibrational playbacks to describe female mate preference functions ca. 2 weeks later, when they reached maturity (playback procedure below).
We then paired the females with randomly chosen, unrelated, field-collected males, placing each pair on a potted host plant. (As Enchenopa females mate only once, the fecundity of experimental females could not be assessed across more than one male; Sullivan- Wood, 1993.) The plants were covered with a screen cage to prevent the treehoppers from flying away while allowing the male and the female to interact freely. Note that this design confounds variation in the fecundity of each treehopper female with variation due to the male with which she was paired and with variation due to the host plant on which she was placed. However, the key parameter of fecundity for each full-sib family was assessed with replication across host plant individuals. We allowed the females to lay eggs until they died in the fall with the first frost. Finally, we counted the eggs laid by each female as an estimate of her fecundity and ecological performance. We note that aspects of fitness other than fecundity (such as growth rates and survivorship to the adult stage) may F I G U R E 1 Spectrogram (top) and waveform of the signal bout produced by an Enchenopa male. Note that the signal consists of a pure tone that sweeps slightly downwards in frequency, followed by pulses. The spectrogram is for illustrative purposes; we took all measurements from the waveforms. We took the following measurements: the number of signals in the bout; the length of the signal; the number of pulses at the end of the signal; and the dominant frequency of the signal, calculated from the length of 10 cycles at the point of highest amplitude in the signal

| Male signals
Mate-searching Enchenopa males move from one plant to another, signaling on each plant until they receive a response from a female . They also tend to signal when placed on a plant stem in the lab, and we took advantage of this behavior to record them. We placed males individually on a potted host plant in the lab.
If a male did not start singing within 2 min, we played back a primer to help induce signaling. This primer consists of a male-female duet, and does not change the signaling behavior, other than encouraging a reluctant male to start signaling. We measured temperature near the recording plant to the nearest 0.5°C.
For our analysis, we measured four signal traits that are associated with mate preferences of varying strength in our study species . For the hypothesis, the most relevant signal trait is dominant frequency (Figure 1): It has the strongest mate preference and is the most distinctive signal trait among species in the complex . The other signal traits that we measured were as follows, in decreasing order of the strength of the mate preferences: signal length, number of pulses, and number of signals ( Figure 1).
Enchenopa males produce signals in bouts, along which signal features vary slightly . We standardized our measurements with a landmark position on a signal bout: the third signal of the second bout produced by a male, or the closest to this signal as possible (e.g., the second signal if a male produced only two signals in his second bout, or the third signal of the first bout if a male produced only one bout). We took all measurements from the signal waveforms in AUDACITY (Figure 1).
Signal dominant frequency was influenced by temperature; we therefore standardized all measurements to 23.5°C before the analyses described below (using the slope of the regression on temperature). The other signal traits were not influenced by temperature (p ≥ .25).

| Experiment 2: Population-level female mate preferences
We described univariate mate preferences for each of the above signal traits to compare each one with how the signal traits relate to variation in female fecundity. Differences in the shape and strength of the preferences give additional leverage to test hypotheses about their contribution to signal evolution and speciation .
Mate preferences are representations of the attractiveness of signals according to their features (Kilmer et al., 2017). Analyzing mate preferences requires assessing attractiveness across a relevant range of signal trait values, that is, they are function-valued traits (Kilmer et al., 2017). We used vibrational playback with synthetic stimuli resembling male advertisement signals. We placed each female on the stem of a potted playback plant and allowed her to acclimate for 2 min.
We presented the stimuli through a piezoelectric stack connected to the stem of the plant, driven by a piezoelectric controller (Thorlabs, Newton, NJ, USA). We delivered stimuli at an amplitude of 0.15 mm/s. playbacks (in random sequence) to describe her preference for signal frequency and for signal length (in random order, with 10 min separating the playbacks for each preference). We recorded the playbacks and the females' responses with the laser vibrometer and AUDACITY as per above. We were only able to conduct playbacks as above to describe the mate preferences for signal frequency and signal length. We did not have time to run playbacks for the other signal traits. We therefore used the data from a prior study with females from the same population  for the preferences for signals/bout and pulse number.
Our assay of preference is based on the natural duetting behavior of Enchenopa females (see above). Females duet with artificial stimuli just as they do with live males, allowing for fine-scale analysis of their mate preferences (e.g., Fowler-Finn & Rodríguez, 2012aRodríguez, , 2012bRodríguez et al., , 2013. We noted the number of responses that females produced in response to the stimuli. We then averaged across replicates (split-families) and families for each stimulus to generate the population-level response data. For the preference for signals/bout, rather than the number of responses, we used the percentage of the females tested that responded to the stimuli (because an increase in the number of responses with the number of signals/stimulus bout might simply reflect the opportunity to respond, rather than a preference for more signals in a bout).
We used the program PFunc (Kilmer et al., 2017) to generate preference functions. This program fits nonparametric cubic splines to the female response~stimulus feature data, and avoids any assumptions about the shape of the functions other than some level of smoothness (Kilmer et al., 2017;Schluter, 1988). To compare the preferences for different signal traits, we scaled the preferences produced by PFunc to the same maximum value (=1), by dividing by the maximum value for each preference. We then compared the mate preferences against the plots of the relationship between genetic variation in female fecundity and the corresponding signal traits (see below).
We estimated the strength of the mate preferences (the degree to which attractiveness changes over signal trait values) as the ratio of the standard deviation of female response across the preference function and the mean female response (Kilmer et al., 2017).
T A B L E 3 Animal model estimates of broad-sense heritability (with 95% confidence interval) and CV genetic in Enchenopa female fecundity and male signal traits (Figure 1)

| Testing predictions (i) and (ii): Heritabilities and genetic correlations
To estimate heritabilities, we only used families with data for at least two individuals/replicate ( families. This small sample size was a function of the difficulty of obtaining the full data for enough individuals in each replicate and family, which required keeping alive not only the males through signal recording but also the females through mating and the end of egg-laying well into the Fall. We used the animal model, implemented in R using the MCMCglmm package (Hadfield, 2010;Wilson et al., 2010). To represent the full-sib, split-family design in terms of the animal model, we coded pedigrees with one sire and one dam per family, with no relatedness among sires and dams. To examine whether the priors influenced the outcome of the model, we ran the analysis with three different ratios: with even priors (divided equally among individual, replicate, and residual effects); with priors biased toward the animal term (90% animal, 5% replicate, 5% residual); and with priors biased heavily toward the residual term (5% animal, 5% replicate, 90% residual). In all cases, we used low belief in the priors. We used chain lengths of 1,000,000 iterations, sampling every 1,000, with a burn-in of 500,000. All autocorrelation Note that our full-sib, split-family rearing design yields estimates of broad-sense heritability as they may include nonadditive components such as dominance variance and maternal effects. We also report the amount of genetic variation as a coefficient of variation, following Houle (1992), but using the notation CV genetic rather than Houle's CV A to indicate that our estimates of genetic variance may include nonadditive components. We calculated CV genetic thus: CV genetic = 100 √(variance estimate)/mean.
We also estimated r G with another method. We calculated the Pearson's correlation between family median values for female fecundity and male signal traits in JMP (v. 7.0.1; SAS Institute, Cary, NC, USA). To obtain these values, we first calculated the median for each split family and then obtained the median of those values for each family.

| Testing predictions (iii)-(vii): Comparing the fecundity~signal trait relationship with mate preferences
We used Pearson's correlations to assess the relationship between the strength of mate preferences and the absolute value of r G between female fecundity and the signal traits. With n = 4 signal traits, this correlation would have to be of very large magnitude to be statistically significant. We therefore focused on its effect size.
We overlaid scatterplots showing family median values and the mate preference function splines. The linear or curvilinear shape of T A B L E 4 Animal model estimates of the genetic correlation (with 95% confidence interval) between Enchenopa female fecundity and male signal traits

Residualbased priors
Correlation with: these preferences was established in prior work  and confirmed here. We tested for curvilinearity in the fecun-dity~signal trait relationships by fitting quadratic regressions.

| Corollary analyses
We used Pearson's correlations to assess the relationships between the strength of mate preferences and the amount of genetic variation in the corresponding signal traits. We also used Pearson's correlations to determine the relationship between the amount of genetic variation in a signal trait and the magnitude of r G between female fecundity and the signal trait. As above, we focused on the effect size of these correlations.

| RESULTS
We found heritability of small effect size in fecundity, and of smallto-large effect size across signal traits (Table 3). The different priors did not substantially influence the heritability estimates for female fecundity and most male signal traits, except for dominant frequency and signal length (Table 3; Figure 3). The key signal trait (dominant frequency) had heritability of low-to-medium effect size. This met the requirement outlined in prediction (i), and we therefore proceeded to test the other predictions (Table 1).
All r G estimates by the animal model had 95% CIs that overlapped zero (Table 4). This result offers a rejection of the hypothesis that mate choice causes local adaptation (Table 1): Without support for prediction (ii), none of the other predictions can be supported. However, this could be due to the small sample of full-sib families that our criteria for within-family sample sizes allowed (n = 16 families; see above) (cf. Sharma, Wilson, & Hosken, 2016). The mode of the posterior probability distributions of the animal model estimates was consistently at or near zero for only one signal trait (signals/bout) (Figure 4). For the other signal traits, the mode was consistently either positive (dominant frequency, number of pulses) or negative (signal length) across the different priors (Figure 4), suggesting that these correlations may be nonzero but that our analysis may be weakened by the small sample of families, which could represent support for prediction (ii). We therefore complemented our study with the below analyses, to seek a more robust test of the hypothesis.
We found correlations of mostly large effect size between the strength of mate preferences and the magnitude of |r G | between female fecundity and the signal traits ( Figure 5). This lends support to prediction (iii) ( Table 1). Additionally, we found a significant r G estimate of large effect size between female fecundity and male signal frequency when we used family median values ( Figure 6). This would support predictions (ii)-(iv) (Table 1), because signal frequency is the signal trait for which female mate preferences are strongest, and the signal trait that is most distinctive among the species in the E. binotata complex . Nevertheless, the shape of the fecundity~signal frequency relationship did not match the shape of the population mate preference, which is curvilinear and favors signal frequencies that are not associated with the highest fecundity values ( Figure 6). This therefore fails to support predictions (v)-(vii) ( Table 1). The estimates for r G between family median values for female fecundity and the other signal traits were small to medium in effect size and nonsignificant, and would also not match the shape of the corresponding population mate preferences ( Figure 6).
In the corollary analyses, we found mixed results for the correlations between the strength of mate preferences and the amount of genetic variation in signal traits. Correlations with signal trait H 2 were very weak and of variable sign (Figure 7a). But, correlations with signal trait CV genetic were of mostly large effect size and consistently negative (Figure 7a). These results thus leave some possibility that selection due to the mate preferences has eroded genetic variation in signal traits in such a way that it might limit the potential for the presence of fecundity~signal genetic correlations.
Results for the correlations between the amount of genetic variation in signal traits and the magnitude of r G between female fecundity and the signal traits were also mixed. Correlations between signal trait H 2 and r G were mostly weak, although consistently positive (Figure 7b). And, correlations between signal trait CV genetic and r G were mostly strong but of varying sign (Figure 7b).
Thus, there is some suggestion that the amount of genetic variation in signal traits may limit the potential for fecundity~signal genetic correlations.
F I G U R E 5 Relationship between the strength of mate preferences and the magnitude of the genetic correlation (r G ) between female fecundity and the corresponding male signal trait in Enchenopa in our rearing experiment. We used the absolute value of r G to focus on its magnitude. We show correlations for the r G estimates obtained with the different priors used in the animal model

| DISCUSSION
We tested the hypothesis that mate choice causes local adaptation such that, across environments, it can promote specialization and speciation Lorch et al., 2003 F I G U R E 6 Comparison of the female fecundity~signal trait relationship (genotypic values obtained as family medians) and the populationlevel female mate preference function for the signal trait. In each panel, the axes show a range corresponding to the mean ± 2 standard deviations. Symbols in black indicate family median values, and the error bars in black correspond to the 40th-60th percentiles. The curves in blue indicate the population-level mate preference functions. (a) r G between fecundity and signal frequency estimated with family median values was significant (r = .51, p = .042, n = 16). There was no indication of curvilinearity (quadratic fit on signal frequency: F 2,13 = 2.33, p = .14) and thus no match with the mate preference. (b) r G between fecundity and signal length estimated with family median values was not significant (r = −.02, p = .95, n = 16). There was also no indication of curvilinearity (quadratic fit: F 2,13 = 0.28, p = .76), so that the relationship would not have matched the mate preference. (c) r G between fecundity and the number of signals/bout estimated with family median values was not significant (r = −.30, p = .25, n = 16). The test for curvilinearity was marginally significant (quadratic fit: F 2,13 = 3.22, p = .07) but would not in any case result in genotypes associated with high fecundity being favored by the mate preference. (d) r G between fecundity and then number of pulses was not significant (r = .26, p = .36, n = 16), and there was no indication of curvilinearity (quadratic fit: F 2,13 = 1.20, p = .33). Additionally, the range of genotypic values for pulse number was so narrow that it would not allow the mate preference to favor genotypes associated with high fecundity Natural and sexual selection are involved in the process of speciation in the E. binotata complex, but divergent sexual selection due to mate choice is a main cause of signal evolution in the complex Sullivan-Beckers & Cocroft, 2010 Cocroft et al., 2010). These two results suggest that mate preferences could be selected to favor signal genotypes associated with high ecological performance. Nevertheless, this does not seem to have been the case: The signal frequencies that are favored by the population-level mate preference are not associated with high fecundity.
Our rejection of the hypothesis that mate choice promotes local adaptation and ecological divergence is tentative because of the small F I G U R E 7 Analysis of potentially confounding factors in our test for genetic correlations between female fecundity and male signal traits. (a) Relationship between the strength of mate preferences and the amount of genetic variation in signal traits, measured as H 2 or CV genetic (see text). (b) Relationship between the amount of genetic variation in signal traits (measured as H 2 or CV genetic ) and the magnitude of r G (absolute value) between female fecundity and the corresponding signal trait sample size of full-sib families that we were able to obtain, and we encourage further tests with other species. However, our findings suggest that signal-preference divergence and ecological divergence may occur in parallel but separately during speciation. Even in such a case, there are various possible synergistic interactions between natural and sexual selection (Maan & Seehausen, 2011;Safran et al., 2013).
For example, rapidly diverging signals and preferences may offer early prezygotic reproductive isolation, complemented eventually by postzygotic incompatibilities due to ecological specialization (Coyne & Orr, 2004). Nevertheless, our results suggest that the challenge for theory may lie in explaining how speciation can result from the joint but separate action of natural and sexual selection, rather than from a single process. Data will be archived in DRYAD upon acceptance for publication.