Sexual selection is a powerful evolutionary force shaping mate choice phenotypes, initiating phenotypic shifts resulting in (or reinforcing) population divergence and speciation when such shifts reduce mating probabilities among divergent populations. In the Hawaiian cricket genus Laupala, pulse rate of male calling song, a conspicuous mating signal, differs among species, potentially behaving as a speciation phenotype. Populations of the widespread species Laupala cerasina show variation in pulse rate. We document the degree of population differentiation in three features of calling song: pulse rate, pulse duration, and carrier frequency. All show significant population differentiation, with pulse rate showing the greatest heterogeneity. A Mantel test found no relationship between geographic distance and pulse rate divergence, indicating that a simple model of greater divergence with increasing distance cannot explain the observed pattern of differentiation. We demonstrate that female preference functions for pulse rate are unimodal, and that preference means show significant differentiation among populations. Furthermore, estimates of pulse rate preference correlate significantly with mean pulse rates across populations, indicating song and preference coevolve in a stepwise manner. This correlated divergence between signal and preference suggests that sexual selection facilitates the establishment of sexual isolation, reduced gene flow, and population differentiation, prerequisites for speciation.

Many closely related species differ conspicuously in secondary sexual signals and preferences of the mate recognition system (Darwin 1871; West-Eberhard 1983; Ritchie 2007). The processes by which these patterns arise are of intense interest because of their direct bearing on the origins of species. Although the processes causing the evolution of such phenotypes occur within populations, the consequential divergence among populations can have significant repercussions for communication between populations (Lande 1981; Endler 1992). Evolved mating signals within a population may result in sexual isolation from other populations whose mating signals have remained the same or evolved differently (Kirkpatrick 1982; Lande 1982). If signals and preferences covary among populations, assortative mating could result in reduced mating probabilities and depressed gene flow between populations (Panhuis et al. 2001; Uyeda et al. 2009).

Although interspecific differences in mate recognition phenotypes appear to make critical contributions to reproductive isolation among species, determining the origins and evolutionary dynamics of these differences must involve the study of intraspecific variation in signals and preferences (Härdling and Karlsson 2009; Uy et al. 2009). The functional interdependence of signal and preference suggests that a preference function (often expressed by females) can describe the selective pressure on signals (often expressed by males) due to female choice (Wagner 1998; Brooks et al. 2005; Gerhardt and Brooks 2009; Sullivan-Beckers and Cocroft 2010). The degree of overlap between signal and preference variation is an important factor in predicting the evolutionary potential and trajectory of signal-preference evolution (Ritchie 1996). For example, it is often hypothesized that a mate recognition system with strong stabilizing selection will tend toward extensive trait-preference matching within and covariance across populations (Butlin and Ritchie 1989; Rodriguez et al. 2006). In addition, the genetic architecture of these traits can constrain, or promote, the tempo and precision of trait and preference coevolution that ultimately leads to interspecific differences. For example, theory suggests that details of the genetic architecture of diverging sexual traits can influence the likelihood and rate of speciation (Dieckmann and Doebeli 1999; Kondrashov and Kondrashov 1999; Verzijden et al. 2005; Hall and Kirkpatrick 2006; Hayashi et al. 2007).

Elucidation of the potential pathways by which interspecific differences in mate recognition evolve requires systems that exhibit intraspecific variation in these same traits. Although numerous studies are beginning to examine geographic variation in mating signals and preferences (Simmons et al. 2001; Simmons 2004; Prohl et al. 2006, 2007; Reynolds and Fitzpatrick 2007; Ryan et al. 2007), particularly valuable would be studies that simultaneously document (1) variation in both signals and preferences for those signals among populations, (2) a genetic basis to covariation between trait and preference, and (3) interspecific variation in these same traits, enabling mechanistic and evolutionary studies of different stages of the speciation process (Panhuis et al. 2001; Boake 2002). Intraspecific variation in such systems can yield temporally fine-grained observations of trait differentiation, promising insights into how interspecific differences accumulate from intraspecific differentiation, and the relationship between female preference functions and the degree to which signal and preference coevolve or become decoupled during divergence.

The Hawaiian cricket genus Laupala has undergone rapid speciation (Mendelson and Shaw 2005), resulting in 38 closely related species that differ conspicuously in acoustic aspects of sexual communication, but that are otherwise cryptic (Otte 1994). Males of the different species of Laupala produce structurally similar songs that vary primarily in pulse rate, and species occurring in sympatry can always be distinguished by pulse rate (Shaw 1999, 2002; Parsons and Shaw 2001; Mendelson and Shaw 2002, 2005). Furthermore, females preferentially respond to the pulse rate produced by males from their own species (Shaw 2000; Shaw and Herlihy 2000; Mendelson and Shaw 2002). Genetic studies of species differences in Laupala show that both pulse rate and preference variation is due to relatively small, additive allelic effects (Shaw et al. 2007; Shaw and Lesnick 2009). Furthermore, loci underlying both song and preference differences between species show genomic linkage (Shaw and Lesnick 2009; Wiley and Shaw 2010). These observations between species suggest the hypothesis that divergent evolution of song and preference within species occurs through small, coordinated changes in signal and preference.

We examine this hypothesis in a study of Laupala cerasina, a widely distributed species endemic to the Big Island of Hawaii. Field recordings have suggested extensive pulse rate differentiation across the range of L. cerasina (Otte 1994), indicating that L. cerasina populations have diverged in the same character that varies conspicuously among species. The current study provides statistical support for the extensive pulse rate variation in L. cerasina and tests the hypothesis that male sexual signals and female preferences for those signals covary across the range of the species. The hypothesis that differences among populations in male song and female preference have a genetic basis is tested using a common garden experiment. The results are discussed with respect to the mode of change in both male and female sides of the acoustic communication system and in the context of geographic structure among populations.



Subadult L. cerasina crickets were collected from 13 locations (Pololu Valley, Muliwai Plateau, Waimea Reservoir, Kalopa Park, Laupahoehoe, Akaka Falls, Eucalyptus Toe, Kaiwiki, Ola’a Flume, Glenwood, Wright Road, Naulu Trail, and Kaiholena), hereafter referred to as populations, from the Big Island of Hawaii in July 2004, 2005, and 2006 (Fig. 1). The distribution of L. cerasina is made up of spatially disjunct populations. Sample sites were chosen to cover the geographic range and anticipated pulse rate variation of the species. In the laboratory at the University of Maryland, individuals were reared to maturity under a 12:12 light: dark cycle in a temperature-controlled environment maintained at 20°C, reflecting field conditions. Because the census size of most populations is quite large, and individuals used in this study were wild-caught and spanned different age classes, it is unlikely that any individuals sampled are related. Two to three wild-caught individuals of the same sex were reared together in medical specimen cups, with a moistened kimwipe for water and humidity. Fresh cricket chow (Fluker's Farms, LA) was provided twice per week (2004, part of 2005) or cricket chow treated with the anti-fungal agent methyl paraben (Tegosept, Thermo Fisher Scientific, Waltham, MA) once per week (part of 2005, 2006). First laboratory generation crickets (G1) were collected as eggs from naturally inseminated females caught in the wild or paired with a wild-caught male in the laboratory. Second laboratory generation crickets (G2) were generated through random pairing of nonsibling males and females from the same population.

Figure 1.

Topographic map of the Big Island of Hawaii, showing the approximate collecting locations of 13 populations of Laupala cerasina, along with estimated coordinates. Scale bar is approximate.


Single mature males were recorded with a Sony (Tokyo, Japan) Professional Walkman (model WM-D6) and Sony condenser microphone from a clear plastic cup fitted with a screen top in a temperature-controlled room maintained at 20°C. A female from the same population was placed with males that failed to sing spontaneously (males produce only one type of song in Laupala). Songs were digitized and analyzed using Raven version 1.2.1 (Bioacoustics Research Program, Cornell University Lab of Ornithology). From each song, five nonconsecutive measurements were taken of pulse period, pulse duration, and carrier frequency, with a total duration of about 5 s of song being sampled. Pulse period, measured as the time between the beginning of one pulse and the beginning of the next pulse, was transformed to calculate the pulse rate (the inverse of the pulse period). Pulse duration is the time elapsed between the beginning and end of a single pulse. Finally, carrier frequency is calculated as the dominant frequency of the pulse.

Male song data were analyzed using SAS version 8.2 for Windows (SAS Institute, Cary, NC). The five measurements of pulse rate, pulse duration, and carrier frequency were averaged for each male to get a single measurement per individual. The coefficient of variation (CV) was calculated for each male by dividing the standard deviation by the mean and multiplying by 100 to get a percent. A low CV suggests that using mean values from individuals is appropriate for subsequent analysis. An analysis of variance was performed to test the hypothesis that song variables show significant heterogeneity among populations. Additionally, a correlation analysis was performed using pulse rate, pulse duration, carrier frequency, and temperature to test for independence of song characters. Each song value was adjusted to account for population differences by subtracting the value for each trait from the population mean for each individual, and using those values (the residuals) in the analysis.

To determine whether geographic variation in song has a genetic component, individuals from the 2005 collections at Glenwood Road, Kalopa Park, Laupahoehoe, Muliwai Plateau, Naulu Trail, and Waimea Reservoir were subjected to a common-garden study in the laboratory for two generations. Then the songs of the resulting progeny of outbred pairs were recorded under the same experimental conditions as the wild-caught generation. Measurements of pulse rate, pulse duration, and carrier frequency were conducted in the G2 males using the same procedures as in the wild-caught males, and the results were compared across generations. The mean song values measured from the G2 progeny were used in a correlation analysis with the means from the parental (wild-caught) generation.

To test for a relationship between geographic distance and pulse rate divergence, a Mantel test was performed using XLSTAT in Microsoft Excel. Geographic distances were calculated using the ruler tool in Google Earth, with collecting locations estimated within 1 s of the recorded coordinates listed in Figure 1. Song divergence was measured as the absolute value of the difference between the mean male pulse rates for each population pair.


Phonotaxis methods were identical to those in Shaw and Herlihy (2000). Briefly, a circular test arena (47 cm radius) covered in a fiberglass screen was housed in a temperature-controlled acoustic chamber (Acoustic Systems) for phonotaxis trials. Two 8.5 cm speakers (Radio Shack model 40–1218) placed 180° apart just outside the arena broadcast synthesized songs offered in pairs from a PC computer using custom software. From each speaker, a pulsed, sinusoidal tone was played back through a 16-bit digital/analogue converter (Tucker-Davis Technologies, Gainesville, FL). The pulse amplitude envelope had a rise time of 10 ms and a fall time of 30 ms and the acoustic output was filtered at 10 kHz using a Krohn-Hite filter (model 3322) to avoid aliasing. Sound pressure levels were equalized to 90 dB on a 4.0-pulses per second (pps) tone monitored with a Bruel and Kjaer sound pressure level meter (type 2230, fast-root-mean-square setting).

To conduct a trial, a female was placed beneath a plastic cup in the center of the arena. Pulse duration and carrier frequency were held constant at 40 ms and 5.0 kHz, respectively, while pulse rate varied depending on the trial. These values were selected a priori based on previous work on a different population of L. cerasina (now extinct) that showed carrier frequency preferences ranging from 4.7 to 5.2 kHz and open-ended pulse duration preferences that exceeded the natural variation of the population (Shaw and Herlihy 2000). We chose a carrier frequency (5.0 kHz) from within the preferred range and a pulse duration (40 ms) near the population mean of 41.7 ms as our standard values. Each female was used in a series of six different trials (2.0 vs. 2.5 pulses s−1, 2.2 vs. 2.7 pulses s−1, 2.4 vs. 2.9 pulses s−1, 2.6 vs. 3.1 pulses s−1, 2.8 vs. 3.3 pulses s−1, and 3.0 vs. 3.5 pulses s−1) in randomized order, two per day, over the course of three days. These pulse rate values more than span the natural range of variation observed in male calling song across L. cerasina populations (Otte 1994). Paired songs were broadcast for 5 min before the cup was remotely raised, allowing the female access to the arena. If the female approached within 10 cm of one of the speakers within the trial period, the pulse rate choice was recorded and the trial ended. If the female did not approach either speaker after 5 min, the trial was ended and no choice was recorded. Only sexually mature, virgin females 20 days past the final molt were used in trials. In each trial, the speakers were randomly assigned a pulse rate.

In general, females responded to the faster pulse rate at the slow end of the trial series and to the slower pulse rate at the fast end of the trial series. We conducted an analysis to estimate the pulse rate corresponding to the inflection point, or point of indifference, where the proportion of females responding to the faster (or slower) pulse rate equaled 0.5 in a given population. Responses from a given population were analyzed simultaneously using logistic regression with the statistical software package R. To determine a standard error for these estimated mean preferences, the preference datasets (preference functions) of females within a given population were sampled with replacement to generate 1000 bootstrap replicates per population, maintaining the original sample size for each replicate. For each bootstrap replicate, the pulse rate corresponding to the inflection point was calculated, providing a pulse rate preference estimate for that replicate sample. The mean preference estimate and its standard deviation (i.e., standard error) were calculated for each of the 1000 bootstrapped replicates.

To determine whether there is a genetic component to geographic variation in female acoustic preference, individuals from the 2005 collections at Glenwood Road and Naulu Trail were reared in a common garden study for two generations (as with males; see above). The resulting G2 progeny were used in the same experimental design as the wild-caught individuals. Logistic regressions of phonotaxis response were compared across generations within populations (to test whether population preferences change across generations), and between populations within generations (to test whether differences persist across generations between populations) using Wald chi-square tests (Grace and Shaw 2004).


To test for population-level correlations between song and preference, a Spearman rank correlation test was performed using the mean pulse rate and preference estimates for each population for each year separately. We also performed a global correlation by combining all of the data for each population and performing a single correlational analysis across all 13 populations.

For the two populations (Glenwood Road and Naulu Trail) where song and preference data were available for wild-caught and G2 generations, we tested the hypothesis of no difference between pulse rate and pulse rate preference between populations in each generation.



Wild-caught generation

Temperature varied little throughout the course of the experiment (average recording temperature = 20.09 ± 0.17), and as expected, there was no relationship between temperature and any of the measured song characters (Pearson r=−0.01 to −0.05, P= 0.29–0.84). Analyses were conducted separately for each collection year, and results were compiled separately. In total, songs were recorded from 328 males and individual measurements were deposited at Dryad (doi:10.5061/dryad.8645). Pulse rates were significantly heterogeneous among populations of L. cerasina in all years sampled (2004: F(5,103)= 24.07, P < 0.0001; 2005: F(6,101)= 42.97, P < 0.0001; 2006: F(6,104)= 32.13, P < 0.0001). In 2004, seven of 15 pairwise comparisons were significantly different (P < 0.05) following Sidak correction for multiple comparisons; in 2005, 16 of 21 pairwise comparisons were significantly different; and in 2006, 11 of 21 were significantly different (Table 1). Carrier frequency was also significantly heterogeneous among populations in all years (2004: F(5,103)= 7.02, P < 0.0001; 2005: F(6,101)= 3.62, P= 0.0027; 2006: F(6,104)= 4.27, P= 0.0007), but with fewer significant pairwise comparisons (five out of 15 in 2004, one out of 21 in 2005, and two out of 21 in 2006; Table 1). Pulse duration was significantly heterogeneous among populations in 2004 (F(5,103)= 6.12, P < 0.0001), but not in 2005 or 2006 (2005: F(6,101)= 1.84, P= 0.099; 2006: F(6,104)= 0.37, P= 0.89). In 2004, four of 15 pairwise comparisons were significantly different (Table 1). The pulse rates for Glenwood Road males sampled in 2004 were significantly slower than males sampled in 2005, as was true for Wright Road in these same years (Table 1). Wright Road males also had a significantly lower carrier frequency in 2004 than 2005 (Table 1). No other year effects were found within populations. A correlation analysis of pulse rate, pulse duration, carrier frequency, and recording temperature found one significant negative correlation between carrier frequency and pulse duration, with all other relationships nonsignificant (Table 2). There was no support for a relationship between geographic distance and pulse rate divergence across populations (Mantel test: 50,000 permutations; Pearson r=−0.168, P= 0.346).

Table 1.  Geographic variation in male song. Mean male pulse rate (PR), pulse duration (PD), and carrier frequency (CF) with one standard error (SE) for all 13 populations, for each year sampled. Populations are ordered from north to south. Coefficient of variation (CV) shown for each measurement is averaged across all individuals.
PopulationYearNPR (SE)PD (SE)CF (SE)
Pololu Valley2005 52.58 (0.037)39.1 (2.65)4780 (47)
Muliwai Plateau2004112.55 (0.026)40.6 (1.65)4840 (62)
Muliwai Plateau2005202.62 (0.019)42.7 (0.90)4840 (40)
Muliwai Plateau2006282.59 (0.020)41.0 (1.38)4980 (40)
Waimea Reservoir2004232.49 (0.017)44.5 (0.88)4530 (44)
Kalopa Park2004242.33 (0.015)46.0 (0.68)4600 (33)
Laupahoehoe2005182.74 (0.026)44.3 (1.64)4730 (36)
Laupahoehoe2006162.65 (0.023)41.3 (1.78)4730 (61)
Akaka Falls2006212.95 (0.024)39.2 (1.16)4910 (40)
Eucalyptus Toe2005192.91 (0.020)40.3 (1.17)4900 (47)
Eucalyptus Toe2006122.91 (0.025)39.6 (1.51)4870 (57)
Kaiwiki2006 52.80 (0.046)40.4 (1.54)4700 (112)
Ola’a Flume2004 92.61 (0.035)40.9 (1.65)4690 (102)
Ola’a Flume2006142.70 (0.022)38.9 (1.37)4750 (64)
Glenwood Road2004252.53 (0.024)40.0 (0.97)4590 (28)
Glenwood Road2005182.63 (0.020)39.7 (1.16)4740 (48)
Naulu Trail2005132.44 (0.023)40.2 (1.28)4880 (36)
Wright Road2004172.61 (0.022)40.0 (1.60)4420 (45)
Wright Road2005152.75 (0.019)40.3 (1.69)4680 (44)
Kaiholena2006152.73 (0.033)40.3 (1.71)4780 (32)
CV (all individuals)  0.69%4.35%0.55%
Table 2.  Correlations among song characters in wild-caught males. The correlation coefficients from a Pearson correlation analysis using pulse rate, pulse duration, carrier frequency, and recording temperature are shown below, adjusted to account for population differences. Following a Bonferroni correction for multiple comparisons, only one correlation, between pulse duration and carrier frequency, showed a significant (negative) relationship (in bold), with a P<0.0001. All other relationships were nonsignificant (P>0.05).
 Pulse rate Pulse durationCarrier frequency
Pulse rate  
Pulse duration0.053  
Carrier frequency0.1190.213

Laboratory generation

Songs were analyzed from 284 males in the G2 generation, and individual measurements were deposited at Dryad (doi:10.5061/dryad.8645). The six populations bred for two generations in the laboratory (G2 males) were significantly heterogeneous for pulse rate (F(5,278)= 67.84, P < 0.0001) and carrier frequency (F(5,278)= 3.67, P= 0.0031), but not pulse duration F(5,278)= 1.69, P= 0.136). Fewer pairwise comparisons were significant for pulse rate in G2 versus wild-caught generations (9 out of 15), but all of the comparisons that were significant in the G2 generation were also significant in the parental generation. For carrier frequency, two of 15 pairwise comparisons were significant, including the pair that was different in the parental generation. Population means derived from wild-caught and G2 males were significantly correlated across the six populations in pulse rate (Pearson r= 0.972, P= 0.0012) and carrier frequency (Pearson r= 0.899, P= 0.0149), but not pulse duration (Pearson r= 0.699, P= 0.51).

The pulse rate means of wild-caught males from Glenwood Road and Naulu Trail collected in 2005 were among those that differed significantly (t(29)= 6.19, P < 0.0001). Although we found no significant differences in pulse rate between the parental and G2 generations within each population (Glenwood t(56)=−0.93, P= 0.36; Naulu t(51)=−0.66, P= 0.51; Fig. 2), the pulse rates of G2 songs did differ between the two populations (t(78)= 10.71, P < 0.0001; Fig. 2). From the analysis of wild-caught males, the two populations also differed in carrier frequency (t(29)=−2.12, P= 0.043), but not pulse duration (t(29)=−0.30, P= 0.77). However, the difference in carrier frequency between these populations did not persist across generations (G2 carrier frequency analysis: t(78)=−1.30, P= 0.20).

Figure 2.

Comparison of male pulse rate and female preference for pulse rate across generations. The histograms on the left show male pulse rate from second-generation (G2) laboratory-reared individuals (bottom) and their wild-caught ancestors (top) from Naulu and Glenwood. The figures on the right show female phonotaxis results from G2 individuals (bottom) and their wild-caught ancestors (top) from Naulu and Glenwood. The y-axis shows the proportion of females responding to the faster pulse rate in a two-choice trial, along with the 95% confidence intervals. The dashed line at 0.5 indicates no preference for a given trial; populations whose confidence intervals exclude 0.5 show significant preference for one of the two pulse rates in the trial. Trials in which the confidence intervals of one population exclude the mean preference of the other population have significantly different pulse rate preferences for that trial.


Wild-caught generation

In total, 269 wild-caught females were run in preference trials, and individual responses were deposited at Dryad (doi:10.5061/dryad.8645). Typically, females responded to the faster song of the pair at the slow end of the trial range, and to the slower song at the fast end of the trial range. Population means and standard errors within populations were estimated using the bootstrap analysis described above. After applying a logistic regression to the bootstrap replicates, it became apparent that estimates of pulse rate preference scaled with pulse rate means from their respective populations (see below). However, preference estimates from slower singing populations were associated with substantially larger 95% confidence intervals when based on all six acoustic choice trials. This likely occurred because the trials at the fast end of the range were beyond the normal discrimination range of females from slower populations. To improve estimates, the means and standard errors were recalculated using the first four trials only, the first five trials only, and all six trials, for those populations where male pulse rates were less than 2.5 pulses per second (pps), between 2.5 and 2.8 pps, and over 2.8 pps, respectively. Of the 17 groups that had recalculated estimates, 14 showed a reduced standard error. With this systematic adjustment, standard errors dropped by 35% and were similar across populations (Table 3).

Table 3.  Geographic variation in female preference for pulse rate. Mean female preference plus one standard error (SE) for all 13 populations, for each year sampled. The trials used in calculating the estimate are also shown for each population and year (see text). Populations are ordered from north to south. Average male pulse rates (PR) are reported here again for reference.
PopulationYearNPR preference (SE)Trials usedMale PR
Pololu Valley2005102.48 (0.040)1–52.58
Muliwai Plateau2004162.56 (0.050)1–52.55
Muliwai Plateau2005 62.52 (0.070)1–52.62
Muliwai Plateau2006202.58 (0.047)1–52.59
Waimea Reservoir2004192.56 (0.063)1–42.49
Kalopa Park2004182.51 (0.033)1–42.33
Laupahoehoe2005 82.77 (0.081)1–52.74
Laupahoehoe2006252.73 (0.044)1–52.65
Akaka Falls2006112.88 (0.069)1–62.95
Eucalyptus Toe2005 92.91 (0.105)1–62.91
Eucalyptus Toe2006142.90 (0.060)1–62.91
Kaiwiki2006 92.85 (0.040)1–52.80
Ola’a Flume2004102.61 (0.044)1–52.61
Ola’a Flume2006 82.69 (0.035)1–52.70
Glenwood Road2004212.63 (0.029)1–52.53
Glenwood Road2005 72.64 (0.030)1–52.63
Naulu Trail2005112.40 (0.063)1–42.44
Wright Road2004172.75 (0.029)1–52.61
Wright Road2005122.77 (0.157)1–52.75
Kaiholena2006182.78 (0.035)1–52.73

Population preference means were considered significantly different when the mean of one population was excluded from the 95% confidence interval (=1.96SE) of a second population. Note that this allows for three categories of relationship: when the means of both populations are within the other's confidence intervals (NS), when one population's mean is excluded from the other's confidence interval, but the other mean is included within the confidence interval of the first (one-way), and when both means are reciprocally excluded from each other's confidence intervals (two-way).

Populations were analyzed separately by year to mirror the analysis of song variation, although no two-way significant year effects were detected in female preference estimates (see below). As with song, statistically significant differences in pulse rate preference among populations were detected in all three years of the study. In 2004, nine of 15 pairwise comparisons were significantly different, with two comparisons showing one-way significance and seven comparisons showing two-way significance. In 2005, 15 of 21 pairwise comparisons were significantly different, with five comparisons showing one-way significance, and 10 comparisons showing two-way significance. In 2006, 16 of 21 pairwise comparisons were significantly different, with two comparisons showing one-way significance and 14 comparisons showing two-way significance. There was one significant year effect on preference: at Ola’a Flume, our analysis revealed a one-way significant difference between 2004 and 2006, with the 2004 estimate (2.61) falling just outside the 95% confidence interval of the 2006 estimate (2.62–2.76).

Laboratory generation

From the G2 generation (derived from the 2005 wild-caught generation), a total of 16 females from Glenwood Road and 16 females from Naulu Trail were run in preference trials, and individual responses were deposited at Dryad (doi:10.5061/dryad.8645). Consistent with the hypothesis that pulse rate preferences breed true and thus have a genetic basis, a Wald chi-square test for logistic regression showed no significant difference between the parental and G2 generations within the Naulu Trail or Glenwood Road populations, regardless of whether trials 1–4, 1–5, or all trials 1–6 were considered (Trials 1–4: Glenwood: Wald χ2(1)= 1.63, P= 0.20; Naulu: Wald χ2(1)= 0.10, P= 0.75). Analyses using 1–4 trial data revealed significant differences between Glenwood and Naulu populations in both the wild-caught and G2 generations (wild-caught: Wald χ2(1)= 5.38, P= 0.0204; G2: Wald χ2(1)= 5.79, P= 0.0161; Fig. 2). Consistent with our observation that the fifth and sixth phonotaxis trials inflated the standard error estimates for populations with slow songs and preferences (like Naulu), we found no significant differences between these two populations when data from the fifth trial or all six trials were included.

The bootstrap method for estimating preference and 95% confidence intervals (above) revealed a two-way significant difference between Naulu and Glenwood in the wild-caught generation (Table 4). When applied to the G2 generation, the analysis resulted in a one-way significant difference, with the Naulu estimated preference mean falling outside the 95% confidence interval of Glenwood, but the Glenwood estimated preference mean falling just inside the Naulu 95% confidence region (Table 4). This was true when data from trials 1–4 and 1–5 were used for Naulu and Glenwood, respectively (consistent with our inclusion criteria outlined above) or when only trials 1–4 were used for both populations.

Table 4.  Wild-caught and G2 comparisons. Female mean pulse rate (PR) preference estimates and male mean pulse rates, both measured in pulses per second (pps) are compared across generations for two populations of L. cerasina.
PopulationGenerationPR preference (pps) mean (SE, N)95% CI (pps)Pulse rate mean (SE, N)
Naulu Trail2005 (wild-caught)2.40 (0.063, 11)2.28–2.522.44 (0.023, 13)
Glenwood Rd2005 (wild-caught)2.64 (0.030, 7)2.58–2.702.63 (0.020, 18)
Naulu TrailG2 (laboratory-reared)2.36 (0.112, 16)2.14–2.582.43 (0.011, 40)
Glenwood RdG2 (laboratory-reared)2.56 (0.054, 16)2.45–2.672.61 (0.013, 40)

The 95% binomial confidence intervals were also calculated for wild-caught and G2 generations for each pulse rate trial separately. Females from Naulu and Glenwood in both generations responded significantly differently only in trial 2 (trial midpoint pulse rate = 2.45 pps), corresponding to the pulse rate range (2.2 vs. 2.7) that distinguishes males of these two populations (Fig. 2). Collectively, these results suggest that the observed differences in phonotaxis preference among populations persist across generations, demonstrating a genetic basis to population variation.


For 2005 and 2006, pulse rate and preference were significantly correlated (2005: n= 7 populations, Spearman r= 0.964, P= 0.0005; 2006: n= 7 populations, Spearman r= 0.929, P= 0.0025; Fig. 3). Although there was a qualitatively similar trend, there was no significant relationship between song and preference in 2004 (n= 6, Spearman r= 0.148, P= 0.67; Fig. 3). A global analysis combining data across years for each population found a significant correlation between the mean estimated female preference and the mean male pulse rate across all 13 populations (n= 13, Spearman r= 0.934, P < 0.0001; Fig. 4). Due to significant year effects, only 2004 data (yielding the largest sample) were used from Glenwood Road and Wright Road populations in the global analysis.

Figure 3.

Correlation between female preference and male song, by year. Female pulse rate preference means were plotted against male pulse rate means for all populations of L. cerasina, across all years sampled. Lines show the slope of the linear regression (trend line) for each year.

Figure 4.

Covariation between female preference for pulse rate and male pulse rate across all populations of L. cerasina. Preference and pulse rate means were averaged across all years for which there were data. The error bars shown are the 95% confidence intervals for each estimate.


Recent, rapid radiations present ideal opportunities to study species that have diverged in relatively few characters, facilitating focus on evolutionary processes relevant to speciation. If such radiations harbor species that exhibit intraspecific differentiation as well, this provides the opportunity to examine the multiple temporal stages of differentiation that Mayr (1963) viewed as essential to understanding the mechanisms of speciation. When the divergent characters in question are involved in sexual communication and mate recognition, the populations are predisposed toward the evolution of sexual isolation, likely the earliest form of reproductive isolation to evolve in many taxa (Lande 1981; West-Eberhard 1983). In the genus Laupala, the pulse rate of the male calling song is a conspicuous character that varies among otherwise cryptic species, suggesting that this trait may be a speciation phenotype, that is, a trait whose divergence reduces gene flow among incipient species (Shaw 2001; Shaw et al. 2007; Shaw and Mullen, 2011). If this is true, then populations of L. cerasina that have diverged in pulse rate may be incipient species, provided that pulse rate preference has evolved in a correlated fashion with pulse rate.


We found that the mean pulse rates across the range of L. cerasina are heterogeneous to a statistically significant degree. Whereas Otte (1994) presented qualitatively similar results from field-recorded L. cerasina, the variation demonstrated here is observed among populations of wild-caught crickets brought to the laboratory where temperature conditions were monitored and controlled. Because temporal components of song in orthopterans are notoriously sensitive to temperature variation (Gerhardt and Huber 2002; Grace and Shaw 2004), any demonstration that pulse rates are statistically different among populations requires measurement at constant temperatures. Two populations, Glenwood Road and Wright Road, showed a significant year effect on pulse rate between 2004 and 2005. Because pulse rate can be measured very precisely, it is possible that these differences are due to subtle temperature differences at the field sites prior to their collection. The environmental temperature experienced during development can have a lasting effect on adult song phenotypes (Grace and Shaw 2004), so any temperature differences experienced at those locations between 2004 and 2005 could have contributed to the differences in pulse rate observed under laboratory conditions.

Following two generations of laboratory rearing (G2 generation), we showed that the pulse rate heterogeneity among populations persisted, and that the pulse rate means of wild-caught and G2 laboratory-reared males were highly correlated. The mean pulse rate difference between wild-caught males from Naulu Trail and Glenwood Road, the two populations examined for preference variation in the G2 generation, were among those that remained significantly different in the G2 generation. These second laboratory generation observations allow us to rule out the possibility that maternal by environment effects account for the difference between populations. Instead, these findings implicate a genetic contribution to the pulse rate variation among natural populations of L. cerasina.

To a lesser degree, significant variation was observed in carrier frequency and pulse duration among wild-caught L. cerasina populations, although only carrier frequency differences remained significantly heterogeneous in the G2 generation. However, fewer pairwise significant differences in carrier frequency were found among populations, and the significant difference in carrier frequency between Naulu and Glenwood wild-caught males did not breed true in the G2 generation. Thus, male condition, natal environment, and/or maternal effects may play a larger role in both pulse duration and carrier frequency variation. In a previous study, Shaw and Herlihy (2000) found that pulse rate and carrier frequency showed relatively low levels of intramale variability, whereas pulse duration exhibited relatively high intramale variability. Correspondingly, females express unimodal preference functions for pulse rate and carrier frequency but open-ended preferences for pulse duration (Shaw and Herlihy 2000). These patterns match those found in frogs (Gerhardt 1991; Gerhardt and Huber 2002) and are consistent with the suggestion that signal variation associated with open-ended preference functions is substantially influenced by environment or male condition. Both low carrier frequency and long pulse duration have been associated with large body size in other crickets (Brown et al. 1996; Simmons and Ritchie 1996), but this has not been investigated in Laupala.


The phonotaxis data indicate that females across the range of L. cerasina express unimodal preference functions for pulse rate, but females differ by population as to which pulse rates they are most attracted. Unlike the pulse rate estimates for male song, there was no effect of year on estimates of female pulse rate preference (with the partial exception of Ola’a Flume). In general, variability in the standard errors associated with mean pulse rate preference estimates were higher than for mean pulse rates, likely due to either the nature of the measurement or expression of preference. Nonetheless, we found statistically significant differences in pulse rate preference among populations of wild-caught females as measured under controlled temperature conditions in the laboratory. Furthermore, the significant difference in mean pulse rate preference between wild-caught Naulu and Glenwood females persisted in the G2 generation. As with pulse rate, these findings demonstrate a genetic contribution to the variation in preference for pulse rate among natural populations of L. cerasina.


Pulse rate variation was strongly predicted by pulse rate preference across the range of L. cerasina (Figs. 3 and 4). This finding provides support for the hypothesis that the divergent evolution of song and preference among populations is characterized by small, parallel changes to both trait and preference characters. Additionally, we found a high degree of matching between trait and preference means across populations (Fig. 4). Phenotypic correlations such as these could be explained by genetic correlations (through linkage disequilibrium), pleiotropy (where the same genes similarly affect both song and preference), or environmentally based divergence, including genetic × environment interactions. The persistence of song-preference covariation across generations in a common-garden environment indicates that environmental differences among populations are not sufficient to explain variation, and suggests that there is a genetic basis to divergence. Although the genetic architecture of song and preference variation has not been examined in L. cerasina, these findings are consistent with a genetic correlation between trait and preference within populations underlying the correlated phenotypic divergence of these characters among populations. In a closely related pair of congeners, L. kohalensis and L. paranigra, song and preference variation colocalized to a common QTL (Shaw and Lesnick 2009), and multiple QTL that were associated with song variation also predicted preference variation (Wiley and Shaw 2010), suggesting that linkage disequilibrium and/or pleiotropy may keep song and preference characters matched during population divergence.

There was one deviation from this pattern of correlated divergence. At Kalopa Park, females appear to prefer songs that are faster than the population average. This may be a statistical anomaly produced from the experimental design. At the faster end of the range of preference trials, the choices of pulse rates presented to females (2.8 vs. 3.3 pps; 3.0 vs. 3.5 pps) lie far outside the range of male song variation (2.19–2.47 pps). Females were less responsive, and exhibited no significant preference for either song in these trials. When the final trial was removed, the mean preference decreased from 2.61 to 2.55, and the standard error was reduced by half. When the final two trials were removed, the mean preference estimate dropped to 2.51 and the standard error decreased another 25%. This estimate is still faster than the average male song at Kalopa Park. The pulse rate estimate of males at Kalopa Park was slower than preliminary research had suggested, and a better estimate for these females might be obtained by offering slower song choices. Despite the coarseness of the method for estimating female preference across a wide spectrum of variation, the results are remarkably consistent overall, suggesting that the striking pattern of song and preference covariation is a real biological phenomenon, and strong enough to overcome the limitations of the methods for measuring it.


Although this study was not designed to address the causes of divergence, leading candidates include interspecies interactions and sexual selection (Otte 1989, 1994; Mendelson and Shaw 2005). If species interactions influences population divergence in song, there should be evidence that males have shifted their pulse rates to maximize the difference from sympatric congeners (Hoskin and Higgie 2010), resulting in reproductive character displacement. The congeners that live sympatrically with L. cerasina are part of a distinct morphological and genetic lineage of the Laupala radiation (the pacifica group) and are not immediate relatives of L. cerasina (Otte 1994; Shaw 2002), despite extensive sharing of mitochondrial DNA haplotypes in zones of sympatry (Shaw 2002). However, reproductive character displacement is not readily apparent. For example, L. cerasina at Pololu Valley co-occur with a fast-singing species (L. kohalensis, pulse rate ∼3.7 pps) (Shaw 1996), but the pulse rate of this population is intermediate with respect to other L. cerasina populations (∼2.6 pps); the nearby Waimea Reservoir population, where apparently no congeners occur, has a somewhat slower pulse rate mean (∼2.5 pps). In addition, although much of the southern portion of the range of L. cerasina is shared with L. pruna (∼2.0 pps) (Otte 1994; Shaw 1999), this includes not only the populations where the fastest L. cerasina songs are documented (Eucalyptus Toe, Akaka Falls, Kaiwiki), but also populations where songs are intermediate (Ola’a Flume, Laupahoehoe, Glenwood Road) and slow (Naulu). However, the pulse rates of L. pruna are also geographically heterogeneous (Otte 1994), and may shift in concert with L. cerasina across their collective ranges. The question of whether reproductive character displacement influences diversification within L. cerasina is complex and deserves further attention.

The finding of a significant correlation between trait and preference suggests that sexual selection is involved in maintaining population differences, but it is difficult to disentangle the ultimate cause of the initial divergence from the consequential effect of that divergence among isolated populations. Sexual selection could influence population divergence in song through either genetic correlations between trait and preference (Lande 1981), song's role as an indicator of material or genetic benefits, or sensory drive involving pre-existing biases in female perception (Boughman 2002). If sexual selection influences trait divergence through genetic correlations with female choice, or its correlation with benefits to females, then variation in pulse rate should be correlated with variation in female preference for pulse rate, and this correlation should be found at every stage of the speciation process (Panhuis et al. 2001). In contrast, evolution due to sensory drive may not consistently result in correlations between trait and preference across disparate populations, due to the inherent mismatch between trait and preference that drives evolution by this process. This study found that song and preference are significantly correlated across 13 geographically disparate populations of L. cerasina, representing nearly the entire range of the species, weighing against sensory drive. Genetic correlations have been implicated in other species of Laupala (Wiley and Shaw 2010) but material benefits are exchanged from male to female during courtship (Shaw and Khine 2004). Further work is needed to understand the forces acting to cause song and preference differentiation.


In L. cerasina, we find a general pattern of preference functions suggesting stabilizing selection on the pulse rate of the male calling song. Statistically significant, but relatively small, incremental variation characterizes populations in male acoustic traits and female acoustic preferences, and we demonstrate a genetic contribution to this variation. We also find support for the hypothesis that acoustic preference covaries with song across the range of L. cerasina. Taken together, these results indicate that song and preference coevolve in a stepwise manner, revealing an overall pattern of signal-preference matching during early divergence. This pattern of intraspecific song differentiation is reflective of the pattern of divergence among species of Laupala, although to a lesser degree, which makes an investigation of the consequences of population divergence ideal for understanding the mechanisms that influence speciation in this group. The correlated divergence of song and preference among populations of L. cerasina has the potential to reduce gene flow across the species and could promote speciation, if the differences among populations in song and preference are sufficient to cause sexual isolation through assortative mating.

Associate Editor: U. Candolin


This work would not have been possible without the assistance of numerous friends and colleagues. P. Danley, T. DeCarvalho, D. Fergus, S. Lesnick, and S. Mullen collected many of the crickets that were used in these experiments. In addition, they have provided inspiration and helpful feedback throughout the project. S. Lesnick also created the basic topographic image of Hawaii used in Figure 1. J. Booth provided helpful advice and scripts for running bootstrap analysis in R, and F. Siewerdt provided additional statistics advice. J. Jadin, G. Conte, T. Mendelson, E. Turnell, C. Wiley, C. Ellison, K. Oh, S. Reynolds, J-F Savard, J. Keagy, L. Cendes, J. Hebert, J. West, M. Murray, and G. Chen provided support and assistance at various times throughout the completion of the project and during the preparation of the manuscript. Comments from U. Candolin and two anonymous reviewers improved the manuscript. Portions of the project were funded through NSF grants to KLS. The Center for Comparative Evolutionary Biology of Hearing (C-CEBH) graduate fellowship, the Department of Biology at the University of Maryland, the Chesapeake Bay Fund, and the Ann G. Wylie Dissertation Fellowship provided both research and fellowship support to JLG.