K. P. Oh, Department of Neurobiology and Behavior, Cornell University, W319 Seeley G. Mudd Hall, Ithaca, NY 14853, USA. Tel.: +1 520 237 8471; fax: +1 607 254 1303; e-mail: firstname.lastname@example.org
Understanding the genetic architecture of traits involved in premating isolation between recently diverged lineages can provide valuable insight regarding the mode and tempo of speciation. The repeated coevolution of male courtship song and female preference across the species radiation of Laupala crickets presents an unusual opportunity to compare the genetic basis of divergence across independent evolutionary histories. Previous studies of one pair of species revealed a polygenic basis (including a significant X chromosome contribution) to quantitative differences in male song and female acoustic preference. Here, we studied interspecific crosses between two phenotypically less-diverged species that represents a phylogenetically independent occurrence of intersexual signalling evolution. We found patterns consistent with an additive polygenic basis to differentiation in both song and preference (nE= 5.3 and 5.1 genetic factors, respectively), and estimate a moderate contribution of the X chromosome (7.6%) of similar magnitude to that observed for Laupala species with nearly twice the phenotypic divergence. Together, these findings suggest a similar genetic architecture underlying the repeated evolution of sexual characters in this genus and provide a counterexample to prevailing theory predicting an association between early lineage divergence and sex-linked ‘major genes’.
Conspicuous differences in mating signals and associated preferences among recently diverged taxa suggest that the evolution of such traits within lineages plays a causal role in the initial stages of reproductive isolation (Ritchie & Phillips, 1998; Panhuis et al., 2001; Shaw & Parsons, 2002). Elucidating the genetic architecture (e.g. magnitude and direction of phenotypic effects, number and chromosomal distribution of genes) of divergent sexual traits can provide important insights into the mode and tempo of evolutionary changes that accompany speciation (Templeton, 1981; Harrison, 1991; Etges, 2002). For example, if the evolution of sexual isolation is characterized by large and rapid phenotypic changes in sexual signalling systems, one might expect the differences between taxa to be associated with one or a few ‘major genes’ of large phenotypic effect. This follows when large-effect allelic variants enable efficient evolutionary responses that (i) more easily overcome stabilizing selection due to mate choice, (ii) cause the introduction of novel signal components (West-Eberhard 1983) or (iii) more easily establish linkage disequilibrium between sexual signals and preferences (Dieckmann & Doebeli, 1999; Kondrashov & Kondrashov, 1999). Further, these evolutionary scenarios might explain the rapid evolution and disproportionately large effect of genes on sex chromosomes (i.e. ‘Coyne’s Rule’, Ritchie & Phillips, 1998; Presgraves, 2008). In contrast, others point out that, because the evolution of sexual signalling systems requires coordinated changes of both signal and preference, divergence is more likely to be characterized by gradual modification of quantitative phenotypes over time via selection on standing variation—a process enabled by many loci with relatively small additive effects (Type I genetic architecture; Lande, 1981b; Shaw & Parsons, 2002), with no exceptional contribution of sex-linked genes (Ritchie & Phillips, 1998).
Progress on understanding the genetic architecture of premating reproductive isolation has lagged behind that of post-mating isolation, perhaps due to the empirical challenges of measuring quantitative variation in sexual signals and preferences (Shaw & Parsons, 2002). More generally, comparative study of the genetics of speciation must begin with relatively young taxa, in order to avoid confounding the genetic changes involved in initial reproductive isolation with those that have accumulated subsequently. Across the limited number of studies that have focused on phenotypes involved in sexual isolation, generalizations regarding the types of genetic changes that accompany speciation have been difficult to draw because different signalling modalities appear to have distinct genetic architectures and modes of evolutionary divergence (Ritchie & Phillips, 1998).
Species radiations in which reproductively isolating traits have diverged repeatedly among sister taxa due to parallel evolution (Schluter et al., 2004; Shaw & Mullen, 2011) can reveal valuable insights into the mechanisms of speciation. When the same signals and preferences have evolved multiple times independently (e.g. Boughman et al., 2005), we have an uncommon opportunity to investigate common genetic architectures underlying the evolution of sexual isolation. Furthermore, comparisons among species pairs that differ in the degree of phenotypic divergence can provide insight into the genetic changes at progressive stages of differentiation (Harrison, 1991).
Here, we examine the interspecific genetics of acoustic communication evolution between sister species of flightless Hawaiian crickets of the rapidly speciating genus Laupala. Laupala is comprised of 38 morphologically cryptic species that often differ conspicuously in the pulse rate of male calling song used to attract females (Otte, 1994). Each species of Laupala is endemic to a single island, and usually a single volcano within an island. Evidence from both morphological (Otte, 1994) and nuclear DNA sequence divergence (Shaw, 2002) suggests that the current radiation occurred within the last 5 million years beginning on the older island of Kauai. Furthermore, a phylogeny based on AFLP variation supports two independent radiations (resulting in the pacifica and cerasina species groups) that subsequently colonized each younger island establishing common acoustic communities anew on the successively younger volcanoes of these islands (Mendelson & Shaw, 2005). Concomitant with this pattern of radiation, species within each clade appear to have originated in allopatry with some degree of song evolution and female acoustic preference accompanying speciation (Mendelson & Shaw, 2005). In all species, male calling song is structurally simple. Single pulses of sound produced by single wing strokes are delivered in trains lacking in higher order temporal structure. Previous studies have demonstrated repeatedly that species differences in pulse rate and female acoustic preference breed true under laboratory conditions over multiple generations (e.g. Shaw, 1996, 2000; Shaw & Herlihy, 2000; Grace & Shaw, 2011). Critically, species can often produce fertile hybrid offspring of both sexes in reciprocal directions under laboratory conditions, thereby facilitating quantitative analysis of interspecific acoustic differences.
Prior biometrical analyses revealed a polygenic quantitative genetic basis of pulse rate evolution in the pacifica species group between L. kohalensis, a fast-singing species, and L. paranigra, a slow-singing species (e.g. Shaw, 1996; Ellison et al., 2011). These two species are each endemic to different volcanoes on Hawai’i Island having diverged as a consequence of intraisland speciation (Mendelson & Shaw, 2005). These studies also demonstrated a significant but small contribution of the X chromosome to pulse rate differences between these species with an effect size proportional to expectations of equal effect magnitudes of multiple quantitative trait loci (QTL) distributed across chromosomes (Shaw, 1996; Ritchie & Phillips, 1998). A subsequent QTL mapping study confirmed at least eight QTL underlying pulse rate divergence, each of small or moderate contribution to the ca. three pulses per second (pps) difference between L. kohalensis and L. paranigra (Shaw et al., 2007). Analysis of female acoustic preferences using phonotaxis trials has suggested a polygenic basis to differences between these species as well (Shaw, 2000).
In the present study, we examine the quantitative basis of traits involved in both song and preference divergence in a second, phylogenetically distinct species pair: L. cerasina and L. eukolea. Whereas L. kohalensis and L. paranigra occur in the pacifica, and L. cerasina and L. eukolea in the cerasina species groups, both are closely related and of similar age (ca. 0.2–0.4% nuclear sequence divergence and < 0.43 MY; Shaw, 2002; Mendelson & Shaw, 2005). Unlike the previously studied species pair, L. cerasina (Hawai’i) and L. eukolea (Maui) are endemic to different islands (Otte, 1994), thus presenting an incidence of interisland speciation (Mendelson et al., 2004). Furthermore, L. cerasina and L. eukolea have diverged considerably less in male song (ca. 1.5 pps) compared to L. kohalensis and L. paranigra (ca. 3.0 pps or 25 standard deviations; Shaw, 1996). Thus, although both pairs of species have undergone independent evolution of phenotypic values for song and preference, considerable differences exist in both the geographic context and magnitude of divergence. Utilizing biometrical analyses, we first fit genetic models to distributions of male song and female preference in parental and interspecific hybrid lines to test for additivity and dominance. We then use the Wright–Castle–Lande method (Castle, 1921; Wright, 1968; Lande, 1981a) for estimating the minimum number of genetic factors and test for significant sex chromosome effects contributing to interspecific divergence in both traits. The results are discussed in relation to modes of male–female coevolution and parallel evolution of intersexual signalling systems between closely related species.
Materials and methods
Collection and rearing methods
Wild-caught L. eukolea nymphs were collected from sites at Ginger Camp (20° 41′ 60.0″N, 156° 5′ 18.0″W) and Palikea Peak (20° 40′ 20.640″N, 156° 4′ 5.160″W) in Kipahulu Valley on Maui. L. cerasina individuals were collected from Kalopa State Park (20° 2′ 13.200″N, 155° 26′ 36.960″W) on Hawai’i Island. In the laboratory, animals were fed cricket chow (Fluker Farms, LA) ad libitum and reared in plastic specimen cups with moistened tissues to maintain appropriate humidity and provide a substrate for oviposition. To ensure virginity, all males and females were housed separately prior to reaching sexual maturity. Phenotyping of males and females was carried out at age 2–10 weeks post final moult, which corresponds to a period of heightened sexual receptivity in Laupala (Mendelson & Shaw, 2006; Grace & Shaw, 2011). Eggs from mated pairs were collected at 5- to 7-day intervals and allowed to develop in specimen cups.
Interspecific hybrid crosses
All parental phenotypes were measured on wild-caught individuals reared to maturity in the laboratory. Pure parental individuals involved in hybrid crosses were derived from both wild-caught and inbred generations (laboratory generation, 2–4).
Hybrid crosses in reciprocal directions (L. cerasina dam × L. eukolea sire; L. eukolea dam × L. cerasina sire) were established once individuals had aged at least 14 days after the final moult. Males and females were paired assortatively with respect to age, and once mated, partners were housed together for the duration of their lifetime (ca. 5 months from final moult, K. L. Shaw, personal observations). Upon maturing, F1 progeny were sib-mated to generate F2 individuals, following the breeding design utilized by Shaw (1996). First- and second-generation hybrid males were derived from nine distinct parental pairings. First-generation hybrid females were derived from single parental pairings in each reciprocal direction. Second-generation hybrid females were also generated from two distinct pairings, but because we observed no differences among F1 reciprocal crosses (see Results), females were pooled into a single F2 population. For subsequent analysis, mean number of F2 individuals phenotyped per line was 16.1 (± 11.3 SD).
Male courtship song
Recording and quantification of male song variation for each generation were carried out using methods detailed previously for this genus (Shaw, 1996). A single recording for each individual (L. cerasina, n = 24; L. eukolea, n = 16; F1, n = 33; F2, n =172) was made between 1000 and 1600 h in a temperature-controlled (20 °C) anechoic chamber with a Marantz PMD-430 cassette recorder and Telex microphone from virgin adult males placed individually in open plastic chambers with screen covers. Unfiltered songs were digitized using soundscope/16 software digitizing technology (GWI Instruments, Cambridge, MA, USA) at 44.1 kHz to generate oscillogram plots that display trains of pulses comprising the male song bout. Five measurements were made of the pulse period (the beginning of one pulse to the beginning of the following pulse) from single singing bouts. Pulse period measurements were accurate to 10 ms. Data were transformed to pulse rates (the inverse of pulse period); means and variances were calculated for parents and hybrid generations and tested for deviations from normality.
Female acoustic preference
Estimates of female acoustic preference for pulse rate in parental and hybrid generations were carried out using two-stimulus playback experiments with synthesized songs. Single virgin adult females were placed beneath a plastic specimen cup in the centre of a 47-cm-diameter circular arena in a temperature-controlled (20 °C) anechoic chamber. Only females that appeared healthy and vigorous were phenotyped. Two digitally synthesized songs with differing pulse rates were simulated on a personal computer using custom software and played simultaneously through identical speakers (3.5″, Radio Shack model no. 40-1218) placed at opposite ends of the arena. A pulsed, sinusoidal tone was generated via a 16-bit digital/analog converter (Tucker-Davis Technologies, Gainesville, FL, USA), and the synthesized song was filtered at 10 kHz to prevent aliasing (Krohn-Hite filter model 3322). For all trials, pulse duration and carrier frequency were held constant at 40 ms and 5 kHz, and pulses had an amplitude envelope with rise and fall times of 10 ms each. Prior to trials, sound pressure levels from both speakers were measured using a Bruel and Kjaer SPL Meter (Type 4155) and equilibrated on a 4.0 pulses s−1 pulsed tone using Tucker-Davis digital attenuators.
After a 5-minute acclimatization period during which stimulus playback was audible, the specimen cup was remotely retracted and female phonotactic behaviour was observed over a five-minute period. Positive phonotactic response (movement to within a 10-cm zone in front of each speaker) was scored as zero when the female responded to the speaker broadcasting the slower rate or one when responding to the faster rate, at which point the trial was immediately ended as females rarely exhibit responses to both speakers (Shaw, 2000). Female responsiveness to acoustic stimuli across parental and hybrid lines was high (proportion responding: L. cerasina, 16/18; L. eukolea, 11/11; F1 hybrids, 21/21; F2 hybrids, 54/72), although F2 hybrids responded at lower rates compared to L. eukolea and F1 females (comparison of all proportions: table probability = 1.5 × 10−4, n = 122, P = 0.01, Fisher’s exact test; Tukey-type post hoc comparison: P < 0.05 for F2 vs. L. eukolea and F2 vs. F1, P >0.05 for all other comparisons).
Pulse rate preference design
Across Laupala, females have repeatedly exhibited unimodal preference functions in relation to pulse rate (e.g. Shaw, 2000; Grace & Shaw, 2011), indicating a single peak of acoustic preference. Moreover, previous work suggests that females prefer conspecific versus heterospecific song types (J. L. Grace and K. L. Shaw, unpublished data). To test this assumption, we first conducted trials in which parental females (n =6 of each species) were presented synthetic song stimuli with both conspecific and heterospecific typical pulse rates (L. eukolea: 3.99 pps, L. cerasina: 2.31 pps). Additionally, F1 hybrid females (n =19) were subjected to two trials in which an F1-type pulse rate (3.10 pps; i.e. intermediate relative to parental songs) was paired with either parental-type pulse rate (i.e. 2.31 or 3.99 pps).
Quantification of female preference phenotypes followed a ‘sliding window’ design used previously for Laupala species (e.g. Shaw, 2000). In the parental generation, each female was subjected to two trials per day between 1000 and 1600 h, over three successive days, for a total of six trials that covered a range of pulse rates centred approximately on the species typical values. Paired songs differed in pulse rate by 0.5 pps, which previous studies (Shaw, 2000; Shaw & Herlihy, 2000) have demonstrated to be well within the discrimination ability of females in this genus. Range of pulse rates presented varied by 0.2 pps increments, resulting in the following presentations: L. eukolea females, 3.2 vs. 3.7, 3.4 vs. 3.9, 3.6 vs. 4.1, 3.8 vs. 4.3, 4.0 vs. 4.5 and 4.2 vs. 4.7; L. cerasina females, 2.0 vs. 2.5, 2.2 vs. 2.7, 2.4 vs. 2.9, 2.6 vs. 3.1, 2.8 vs. 3.3 and 3.0 vs. 3.5. For F1 hybrid females, the range of six trials was centred on a value (3.15 pps) intermediate to the two parental phenotypes. As F1 individuals are assumed to be uniformly heterozygous, and thus exhibit reduced phenotypic variation, trials varied in 0.1 pps increments, resulting in the following presentations: 2.6 vs. 3.1, 2.7 vs. 3.2, 2.8 vs. 3.3, 2.9 vs. 3.4, 3.0 vs. 3.5 and 3.1 vs. 3.6. Likewise, to accommodate the expected increase in variation in the first segregating generation, a two-phase design was employed in which F2 females were initially presented with four trials that spanned the range of parental values, varying by increments of 0.3 pps: 2.5 vs. 3.0, 2.8 vs. 3.3, 3.1 vs. 3.6 and 3.3 vs. 3.9. These responses were subsequently used to determine the range of values presented to each female in a second, more fine-scaled assay consisting of an additional six trials within a range of 2.1–4.3 pps and varying by 0.1 pps increments. Peak acoustic preference was estimated for individual females as the midpoint value at which the phonotaxis response switched from the faster to the slower pulse rate stimulus. In all generations, the sequence in which trials were presented to each female was randomized, as was the speaker through which the faster song was played.
Tests for additivity and dominance
The relative contributions of additive and dominance composite effects to differentiation in male song and female preference were assessed through joint-scaling tests (Cavalli, 1952; Hayman, 1960b). This method uses weighted least-squares regression (details of regression model provided in Appendix A) to generate predicted means and variances for parental and hybrid populations under particular genetic models, which are then assessed for adequacy of fit to observed values (Lynch & Walsh, 1998). For analysis of means, a model with additive effects only was evaluated first, followed by a model that included dominance effects. The improvement in fit from the inclusion of dominance terms can be assessed as
where and are fit statistics for the additive and additive-dominance models, respectively. This expression is equivalent to a likelihood-ratio test statistic, which is asymptotically χ2 distributed, thus permitting a test for significance of the improvement in fit from the inclusion of dominance effects. Although it is possible to test for higher order genetic effects (e.g. epistasis, Demuth & Wade, 2005) using this approach, the additional degrees of freedom (number of lines) required exceed that of our experimental design, and thus, such analyses are not pursued here.
For joint-scaling tests of variances, only models with additive effects were assessed, as the large standard errors of population variance estimates generally preclude meaningful assessment of nonadditive composite effects (Lynch & Walsh, 1998).
Minimum number of genetic factors
Biometrical analysis of song pulse rate and acoustic preference was carried out using the Castle–Wright estimator as generalized by Lande (1981a) for genetically heterogenous natural populations to calculate the minimum number of genetic factors contributing to the differences between L. cerasina and L. eukolea. Using this method, the segregation variance () and associated sampling variance are first estimated from the phenotypic variances observed in the parental, F1 and F2 generations as follows:
where Ni and are the sample size and observed phenotypic variance, respectively, in the ith generation. The minimum number of genetic factors (nE) and associated variance can then be estimated using the mean parental values (μP) as
where is the sampling variance of the mean for the ith parental line.
X chromosome effect
Given the XO/XX (male/female) sex-determination system in crickets, the relative contribution of the X chromosome to interspecific differences in male phenotypes (IX) can be estimated from the difference between mean F1 phenotypes from reciprocal crosses (which inherit different X chromosomes but are assumed to be uniformly heterozygous at autosomal loci):
where C represents the phenotypic means of the heterogametic sex from the parental and reciprocal hybrid crosses (Reinhold, 1998). Similar contrasts are not possible for phenotypes of the homogametic sex (female).
X chromosome effects estimated in this manner may be confounded with cytoplasmic inheritance and maternal effects on F1 progeny from reciprocal crosses. To evaluate the potential contribution of maternal cytoplasmic effects, we compared mean pulse rates of F2 males from reciprocal parental crosses, which, on average, should differ only with respect to inherited extranuclear components.
All statistical analyses were carried out using sas 9.2 (SAS Institute, Cary, NC, USA). Mean phenotypic values among parental and hybrid populations were compared using anova (PROC GLM), followed by Tukey–Kramer tests for multiple comparisons. Tests for equality of variances among lines were carried out using Levene’s test (Schultz, 1985). Phenotypic differences between reciprocal F1 crosses were assessed by fitting data to mixed linear models (PROC MIXED) with cross direction as a fixed effect and family (nested within cross direction) as a random effect. Covariance parameters were estimated using restricted maximum likelihood, and significance was evaluated with likelihood-ratio tests. Significance of cross direction was assessed using type 3 test of fixed effects, with denominator degrees of freedom estimated by the containment method.
Pulse rate means and variances for parental and hybrid males are given in Table 1. Hybrid males sang at intermediate pulse rates compared to parental males (Fig. 1a,b). On average, F1 hybrid males with L. cerasina mothers sang significantly slower than those with L. eukolea mothers (type 3 test of fixed effect, F7,24 = 12.36, P = 0.010); though we observed no significant heterogeneity among F1 families within each cross direction (log-likelihood-ratio test, χ2(1 d.f.) = 0.2, P =0.65). This difference suggests an X chromosome effect equal to 7.59% of the phenotypic difference between L. cerasina and L. eukolea. Therefore, in all subsequent analyses, the mean and variance for F1 pulse rate were calculated as the average of the values calculated within reciprocal crosses, weighted by the number of individuals in each cross direction. Unweighted F1 mean and variance (i.e. arithmetic mean of within cross mean and variance) were nearly identical to weighted values and were therefore not included in subsequent analyses. Pulse rates for L. cerasina and L. eukolea males differed significantly from one another and in relation to both hybrid generations (F3,70 = 171.2, P <0.001; P < 0.05 for pairwise comparisons, Tukey–Kramer test), whereas F1 and F2 males sang at similar pulse rates (P > 0.05, Tukey–Kramer). Variance was greater among F2 compared to F1 males (F1,203 = 12.68, P <0.001, Levene’s test), which is consistent with expectations for a polygenic trait segregating in the second hybrid generation. However, there was no significant difference in pulse rate between F2 males descended from reciprocal parental crosses (F1,143 = 0.07, P =0.79), suggesting that the observed X chromosome effect is unlikely to be confounded by maternal cytoplasmic inheritance.
Table 1. Means and variances of male song pulse rate and female acoustic preference for parental, F1 and F2 hybrid crosses between L. cerasina and L. eukolea. Values for ‘F1 (cross means)’ were calculated as the average of means from reciprocal cross directions (N = 2) and were used in all subsequent biometrical analyses.
Female acoustic preference
Mean pulse rate (pps)
Mean Pulse rate (pps)
*Statistic weighted by the number of individuals in reciprocal crosses.
P (L. cerasina)
0.54 × 10−2
1.33 × 10−2
P (L. eukolea)
1.47 × 10−2
3.69 × 10−2
F1 (L. cerasina dam)
0.95 × 10−2
0.46 × 10−2
F1 (L. eukolea dam)
0.68 × 10−2
1.69 × 10−2
0.79 × 10−2*
0.98 × 10−2
F1 (cross means)
0.81 × 10−2
7.44 × 10−2
6.29 × 10−2
Female acoustic preference
Parental females exhibited completely assortative acoustic preferences with all individuals preferring song stimuli with conspecific (vs. heterospecific) pulse rates (Fig. 2a,c; χ2(1, n = 12) = 12.0, P = 0.001, Fisher’s exact test). Additionally, F1 hybrid females preferred songs with intermediate pulse rates compared to parental values (Fig. 2b; L. cerasina vs. F1: Z(19)= 4.36, P < 0.001; L. eukolea vs. F1: Z(19)= 3.90, P < 0.001).
Overall, average response functions from parental and hybrid female populations were consistent with a unimodal acoustic preference (Fig. 2, Shaw & Herlihy, 2000). Accordingly, each individual female’s preference was estimated as the midpoint of the pulse rate choice between the trials in which the female’s phonotactic response switched from the faster to the slower pulse rate stimulus. Across all populations, the majority of females (74%) were completely consistent in their phonotactic preferences, as evidenced by response profiles with single inflection points. For the proportion that showed some degree of inconsistent response, acoustic preference was estimated from the first (in considering responses from slower to faster pulse rate) identifiable inflection point, as this was consistent with the overall population-level pattern as well as those females that were perfectly consistent in their responses (Fig. 3). Preference data for L. cerasina were published previously (Grace & Shaw, 2011) but reanalysed here to conform with this methodology.
Analysis of phonotactic responses across the study indicated significant differences in mean preference between parental species and in comparison with hybrids (Table 1, F3,83 = 47.68, P <0.001; P < 0.05, Tukey–Kramer test), although preference did not differ between F1 (Fig. 1c) and F2 (Fig. 1d) females (Tukey–Kramer, P >0.05). As with male song, variance was greater in the F2 compared to F1 population, although the difference only approached statistical significance (F1,63 = 3.79, P =0.056; Levene’s test). As expected for hybrids with homogametic sex chromosomes, F1 females from reciprocal crosses had similar acoustic preferences (F1,17 = 0.03, P =0.87) and were thus combined for subsequent analysis.
Joint-scaling tests for means and variances
Parameter estimates obtained from weighted least-squares regression on means and variances of male song and female acoustic preference are shown in Table 2. Fitting of simple additive models revealed significant additive genetic parameter estimates and no departure between expected and observed means for parental and hybrid generations (Fig. 4a,c; song pulse rate: χ2(2, n = 245)= 0.357, P = 0.83; acoustic preference: χ2(2, n = 100)= 3.26, P = 0.20). The inclusion of dominance terms did not significantly improve model fit for either male song pulse rate (Λ = 0.042, P = 0.84) or female acoustic preference (Λ = 2.70, P = 0.10).
Table 2. Parameter estimates and model fit statistics from joint-scaling test for (a) means and (b) variances of male song and female acoustic preference. Tests for statistical significance of parameter estimates were carried out for analysis of line means only.
Male song pulse rate
Female acoustic preference
Var(S), segregational variance.
Parameter estimates: μ0– modelled mean, αc– composite additive effects, δc– composite dominance effects. Statistical significance of parameter estimate (t-test) and model fit (χ2): NS, P > 0.05; *P <0.05; †P <0.01.
‡Negative variance component equated to zero.
(a) Line means
μ0 ± SE
2.997 ± 0.016†
αc ± SE
0.569 ± 0.028*
0.357 (2) NS
3.26 (2) NS
Additive ± dominance model
μ0 ± SE
3.028 ± 0.017†
αc ± SE
0.659 ± 0.031*
δc ± SE
0.315 (1) NS
0.556 (1) NS
(b) Line variances
Var(L. eukolea) ± SE
0.0127 ± 0.003
0.0209 ± 0.008
Var(L. cerasina) ± SE
0.0052 ± 0.001
0.0114 ± 0.004
Var(S) ± SE
0.0655 ± 0.009
0.0468 ± 0.012
0.473 (1) NS
Observed variances of parental and hybrid male song pulse rate were consistent with predicted values under a model of additive gene action (Fig. 4b, χ2(1, n = 245)= 0.473, P =0.49). Variances in female preference, however, showed significant departure from expectations of additivity (Fig. 4d, χ2(1, n = 100)= 5.19, P =0.02).
Minimum number of genetic factors
Biometrical analysis of variation among parental and hybrid males resulted in an estimate of nE = 5.28 (± 0.70 SE) genetic factors contributing to differences in pulse rate between L. cerasina and L. eukolea. Analysis of female acoustic preference yielded a similar value of nE = 5.01 (± 1.51 SE). In our experimental design, females in the parental generation were subjected to trials with pulse rates that varied incrementally by 0.2 pps, while hybrid females were tested with trials that varied by 0.1 pps. To assess how this difference affected our estimate of nE, we recalculated preference values for hybrid females after removing every other trial from the data set, thereby effectively producing a series of phonotaxis trials that varied by 0.2 pps, equivalent to those presented to females in the parental generation. This resulted in a value of nE = 4.11 (± 1.15 SE), reducing the estimate by 0.87 (ca. 18%). Because this does not qualitatively change our interpretation, we report the value using the full data set.
The evolutionary divergence of intersexual signalling traits can be a potent mechanism for curtailing gene flow between lineages (Panhuis et al., 2001) and consequently has figured prominently in speciation theory (Coyne & Orr, 2004). Understanding the genetic basis of such divergence thus presents an opportunity to test contrasting models regarding the mode and tempo at which reproductive isolation evolves (Templeton, 1981; Harrison, 1991; Etges, 2002). On the one hand, if the evolution of premating barriers is characterized by large and rapid changes in sexual traits within lineages, we might expect differentiation to be associated with ‘major genes’ of large effect, along with a disproportionately large contribution of sex-linked factors (Ritchie & Phillips, 1998). On the other hand, if isolation results from the gradual and coordinated modification of quantitative male and female phenotypes over time, differentiation in sexual traits is predicted to involve many loci with small phenotypic effects, and no ‘special’ contribution of loci on sex chromosomes (Lande, 1981b; Shaw & Parsons, 2002). The results of our study on interspecific divergence in male song and female preferences in Laupala are most consistent with the latter model, suggesting that rapid evolution is sometimes coupled with coordinated, gradual evolution of sexual signalling phenotypes.
Several lines of evidence support our conclusions regarding the mode of evolutionary divergence in acoustic behaviour between L. cerasina and L. eukolea. First, in the parental generation, female acoustic preferences in each species corresponded to the song pulse rates of conspecific males (Table 1), suggesting that both traits have diverged in a coordinated fashion. Second, following two generations of interspecific hybridization, we found that mean F1 and F2 phenotypes were intermediate to parental values for both male song pulse rate and female preference, whereas variances increased between successive hybrid generations, patterns that are consistent with expectations for polygenic quantitative traits (Fig. 1). Furthermore, joint-scaling tests for means and variances of both traits produced estimates that are largely consistent with expectations under additive genetic models (Fig. 4), and biometrical analysis revealed similar genetic architectures, with a minimum of 5.28 (± 0.70 SE) and 5.07 (± 1.51 SE) segregating genetic factors contributing to divergence among species in song and preference, respectively. Differences in pulse rate between F1 males (the hemizygous sex) from reciprocal crosses indicate a significant yet proportional (7.6%) contribution of the X chromosome to divergence between L. cerasina and L. eukolea. Analysis of reciprocal F2 males suggests that this X chromosome effect is unlikely to be confounded by maternal cytoplasmic inheritance, although parental effects cannot be excluded entirely.
Further evidence that premating isolation between these species has evolved via the accumulation of relatively small quantitative changes over time is provided by recent work on intraspecific variation in mating phenotypes within L. cerasina that found incremental and correlated differentiation in pulse rate and preference among populations (Grace & Shaw, 2011). Additionally, biometrical studies of song and preference divergence in other Laupala species (L. kohalensis and L. paranigra) reported similar results with respect to the polygenic architecture of both traits (Table 3, Shaw, 1996, 2000). These findings are corroborated by a mapping study that identified multiple QTL with small to moderate unidirectional phenotypic effects influencing both song pulse rate and acoustic preference (Shaw et al., 2007). Taken together, these studies suggest that independent divergence events across Laupala may have shared a common genetic architecture.
Table 3. Geographic context, phenotypic differences and genetic architectures of phylogenetically independent divergences of male song pulse rate and female acoustic preference in two pairs of Laupala species.
L. kohalensis vs. L. paranigra
L. cerasina vs. L. eukolea
pps, pulses per second, nE, minimum number of genetic factors.
With respect to the polygenic basis of interspecific differences in male sexual traits, although our results are consistent with a number of prior investigations (e.g. nE = 4–7, Barson et al., 2007; Shaw et al., 2007), they contrast with a number of examples in which divergence in signalling traits is characterized by one or a few genes of large effect (e.g. nE = 1.5–2, Butlin, 1996; nE = 1–2, Henry et al., 2002). Of the relatively few studies that have examined the genetic architecture of reproductively isolating female mate preferences, the majority have concluded that interspecific differences are due to the influence of one or a few ‘major’ genes of large effect (e.g. nE = 1–4, Haesler & Seehausen, 2005; Velthuis et al., 2005), whereas only a handful (Ritchie, 2000; Shaw, 2000) have suggested a polygenic basis (for detailed review see Ritchie & Phillips, 1998). Although this likely reflects a certain degree of bias arising from the empirical challenges of studying complex quantitative traits, these contrasting results may also be explained by differences in the types of sexual traits involved. Ritchie & Phillips (1998) have suggested that different signalling modalities are associated with distinct genetic architectures. For example, with regard to sexual signal production, insect chemosensory signals may be expected to exhibit a high degree of functional and developmental modularity and thus likely to accommodate the effects of ‘major genes’ that influence single stages along a biosynthetic pathway, thereby leading to the production of novel compounds (e.g. positional isomers of cuticular hydrocarbons, Takahashi et al., 2001) or chemical compositions (e.g. phereomone blends, Roelofs et al., 2002). In contrast, acoustic or behavioural displays, which arguably require more tightly coordinated interaction of morphology and neurophysiology, are expected to evolve in a manner characterized by smaller, incremental modifications to different components via numerous small genetic changes. Alternatively, certain types of genetic architectures may predispose traits to divergent evolution under particular modes of speciation (Templeton, 1981), suggesting that the observed differences in genetic architecture of premating isolation traits may reflect the particular geographic or demographic context of speciation.
Clades in which the same intersexual communication system has repeatedly diverged among sister species provide important opportunities to examine the relationship between genetic architecture and the evolution of mating barriers through a comparative approach. In the following, we contrast the results of this study for L. cerasina and L. eukolea with those from previous studies of L. kohalensis and L. paranigra (Table 3, Shaw, 1996, 2000). Morphological (Otte, 1994) and molecular phylogenetic evidence (Shaw, 2002; Mendelson & Shaw, 2005) strongly supports the placement of these two pairs of species in distinct and acoustically diverse species group, thus supporting the hypothesis that each represents an evolutionarily independent occurrence of pulse rate and preference coevolution. In relation to geographic modes of speciation, the most conspicuous difference between these two pairs of species is that L. kohalensis and L. paranigra diverged via intraisland speciation, whereas L. cerasina and L. eukolea have resulted from interisland speciation (Mendelson et al., 2004). One possible implication of this difference is that repeated divergence of sexual signals and preferences might have occurred under distinct demographic regimes, with a L.eukolea-like ancestor giving rise to L. cerasina following a founding event. Founder-effect models of speciation (Carson & Templeton, 1984), which highlight the role of extreme population bottlenecks (e.g. interisland colonization event) in the evolution of reproductive isolation, propose that traits with type II genetic architectures are fulcrums enabling a genetic ‘revolution’. While understanding the role of such processes in the diversification of Laupala will require explicit tests that model distinct demographic histories, our results of a type I genetic architecture provide preliminary evidence that founder-effect models may not explain the evolution of acoustic communication, an important speciation phenotype (Shaw & Mullen, 2011), in sexual isolation of this species pair.
Another intriguing contrast among species pairs concerns the relationship between genetic architecture and degree of divergence as, all else being equal, one might predict that the number of genetic factors involved will increase with greater phenotypic differences. Interestingly, although phenotypic divergence of male song pulse rate and female preference between L. kohalensis and L. paranigra is approximately double that of L. cerasina and L. eukolea, the number of loci contributing to divergence in these traits among each species pair was similar. Thus, at least on the scales considered here, our results do not suggest a clear relationship between number of loci and magnitude of phenotypic differentiation.
Patterns of interspecific differentiation in sexual signalling traits in Laupala add to a growing number of examples of parallel evolution of sexually isolating traits (e.g. Boughman et al., 2005). Yet numerous questions regarding the genetic architecture associated with such macroevolutionary patterns remain unresolved (Schluter et al., 2004; Gompel & Prud’homme, 2009), such as whether the same loci with equivalent phenotypic effects are involved across independent events of sexual trait divergence. Taken together with previous work (Shaw, 1996; Shaw et al., 2007; Ellison et al., 2011), our results suggest that a polygenic architecture underlies two evolutionarily independent divergences in acoustic behaviour in Laupala. Elucidating whether the identity and location of the genes involved are similarly conserved awaits further study, although our findings do indicate a relatively small contribution of the X chromosome (∼7.6% of species difference) to divergence in male song for both pairs of species. Remarkably, despite a two-fold difference in the magnitude of divergence between species pairs, the effect of the X chromosome was similarly small for both L. kohalensis × L. paranigra (Shaw, 1996; Ellison et al., 2011) and L. cerasina × L. eukolea (Table 3). However, contrary to empirical examples and theoretical expectations of large effects for sex-linked genes (Ritchie & Phillips, 1998; Presgraves, 2008), the X chromosome contribution in our results is consistent with expectations if genetic effects influencing song were distributed evenly among chromosomes. Indeed, previous authors have argued that large effect by sex-linked genetic factors should be particularly relevant in systems where males are hemizygous, as this should facilitate rapid evolution of male traits (Ritchie, 2000), especially when populations are subject to founder effects. Overall, we suggest that future additional studies utilizing fine-scale QTL mapping in L. cerasina and L. eukolea, as well as other species pairs at various stages of divergence, could provide critical insight into understanding the distribution and magnitudes of allelic effects for the genetic factors implicated here.
Estimates of the number of loci contributing to divergence between L. cerasina and L. eukolea in both male song pulse rate and female preference are most consistent with a type I genetic architecture (i.e. many genes of small effect, Templeton, 1981). However, the influence of ‘major genes’ cannot be excluded entirely as the biometrical approach we employ makes several assumptions including additive genetic effects from unlinked loci, homology of genes affecting trait expression between parental species, random mating of individuals used in crosses, as well as equal and unidirectional allelic effects from loci within each lineage (Lande, 1981a; Zeng et al., 1990). With respect to the assumption of additivity, the only indication of a departure from expectations of additive genetic models observed in our study was in the joint-scaling test for variances of female preferences (Table 2). However, as discussed by Lynch & Walsh (1998), scaling tests for variances are generally considered less powerful than tests for means because of the large errors associated with population variance estimates. In our results, although we are unable to rule out entirely a contribution of dominance effects, observed variances in acoustic preference were all within approximately two standard errors of predicted values (Fig. 4d), and thus, we conclude there is little justification at this point to reject the simple additive model. Violations of the final assumption of the Castle–Wright estimator (e.g. multiple loci with opposing, redundant or unequal effects on phenotypes) will lead to an underestimation of the number of loci contributing to interspecific differences, and thus, we emphasize that the values reported here should be interpreted as minima. However, our findings suggest that, at the very least, this assumption is not violated with respect to the effect of the X chromosome (i.e. X chromosome effect was in the same direction as species differences). At the same time, our estimates of the number of genetic factors are unlikely to be limited by the recombination index, which is approximately twice the haploid chromosome number (n = 8 for Laupala, Shaw, 1996). Interestingly, recent studies in this genus have shown that QTL underlying variation in male song pulse rate co-localize with QTL for female preferences (Shaw & Lesnick, 2009; Wiley & Shaw, 2010), which is consistent with our findings that show similar genetic parameters for male and female traits, and also suggest a mechanism for the coevolution of signal and preference through genetic linkage or pleiotropy, as has been posited by several notable sexual selection models (for reviews see Butlin & Ritchie, 1989; Shaw et al., 2011).
In conclusion, here we have presented evidence consistent with a polygenic basis to intersexual signalling traits that have repeatedly played a central role in the evolution of reproductive isolation in one of the most rapidly diversifying metazoan clades observed to date (Mendelson & Shaw, 2005). Thus, not only does our study stand as a counterexample to a growing trend of studies that highlight the simple genetic structure of traits contributing to interspecific divergence, but it also suggests that rapid speciation need not always require the action of ‘major genes’ with large phenotypic effects.
We would like to thank the members of the Shaw laboratory for helpful discussions that improved the manuscript, R. Yu for assistance in the laboratory and R. Nagata for assistance in the field. This work was supported by USA National Science Foundation grant IOS-0843528 to K.L.S.
Joint-scaling test for line means
Joint-scaling tests for means were carried out as detailed by Lynch & Walsh (1998), utilizing the weighted least-squares multiple regression
where is the mean across all lines, is the mean trait value, is the sampling error, and and are coefficients for the composite additive genetic () and dominance () effects for the ith line, respectively. Presented in matrix form, the coefficients for additive genetic effects in all lines occur in the right column of the matrix
where as the matrix of coefficients for the model including body additive and dominance () effects is
In matrix form, the full linear model is then
where is the vector of observed phenotypic means, a is the vector of genetic effects and e is the vector of errors.
Joint-scaling test for line variances
Joint-scaling tests for variances were also carried out utilizing a weighted least-squares regression following Lynch & Walsh (1998). In matrix form, the linear model is
where v is the vector of observed line variances, e is vector of errors, a is the vector of variance components (, and ) and MA is the matrix of coefficients for these variance components across all four lines:
Final variance components estimates were obtained using the maximum-likelihood procedure proposed by Hayman (1960a) where the elements of the sampling covariance matrix (V) were iteratively replaced with values from the least-squares regression until variance components stabilized (see pg. 229, Lynch & Walsh, 1998).