Sexually dimorphic sensory gating drives behavioral differences in túngara frogs

Sex differences in social behaviors are widespread. A common behavioral domain in which males and females differ is in the response to conspecific social signals. Even when encountering the same signal, such as a male advertisement call or a territorial pheromone, males and females often react very differently. Olfactory cues that trigger a lordosis response in female rodents elicit no response at all in males (Harlan et al., 1984); similarly, ultrasonic cries from rat pups trigger stereotyped retrieval behavior in mother, but not in father, rats (Allin and Banks, 1972). In response to a male's vocal or visual display, female birds and frogs often approach, produce solicitation displays, or otherwise express affiliative or courtship behaviors. To the same signal, males may increase their own display behavior or become aggressive toward the signaler (Berglund et al., 1996). When both sexes respond to social cues, the range of cues triggering a response can differ in males and females. As a general rule, females are predicted to respond to a more restricted range of potential male signals than do males to potential female signals, or to male signals. This difference in selectivity is predicted by the different costs to the sexes in responding inappropriately to sexual solicitiation signals (higher in females than males) compared with the costs of not responding to a potential sexual signal (higher in males than females) (Searcy and Brenowitz, 1988).

The evolution of sexually dimorphic behaviors in response to the distinct selective pressures on males and females is dependent on the evolution of the neural systems underlying those behaviors. The degree to which the neural mechanisms of social behavior overlap in males and females constrains the opportunity for sex-specific responses to selection and directs the sites at which selection can modify neural systems. Sexually dimorphic behaviors may be generated by either dedicated or multifunctional neural systems. A dedicated neural system is one in which the neurons involved subserve a specific sensory or motor task, such as motor circuits involved in controlling the vocal apparatus or the sensory neurons that respond to pheromones, both of which have specific roles in social communication. A multifunctional neural system is one in which the neurons contribute to a number of sensory or motor tasks, such as the motor systems controlling limb movements (walking, jumping, swimming and turning) or the visual system (social communication, protection from predators and localizing food). We have recently proposed that the sensory–motor relay may be a particularly flexible site for modification when sensory and motor systems subserve multiple functions (Hoke and Wilczynski, 2010). Modifying the sensory responses or motor pathways could have widespread effects on other behaviors utilizing the same neural systems, thus limiting the range of alterations possible for any one behavioral task. The links between the sensory and motor systems are probably restricted to a small number of contexts, so varying those links could influence one behavior while leaving others intact. Sexually dimorphic behaviors are a special case of the distinction between dedicated and multifunctional neural systems. Sexually dimorphic behaviors could depend on sex-specific neural systems dedicated to the male or female behavioral response, or could depend on neural systems that are common to both sexes and potentially regulate a number of behaviors. For behaviors that are not regulated by dedicated neural systems in males or females, the existence of sex-specific sensory motor relays within shared neural systems would offer a flexible site for evolution in response to sex-specific selective pressures. In this paper we ask if sex differences in the influence of sensory relays on motor circuits contribute to sex differences in neural systems underlying social behavior.

To address this, we examined sex differences in neural and behavioral responses to reproductive signals in Physalaemus (=Engystomops) pustulosus, the túngara frog. Túngara frogs differ in both sex-typical motor responses and behavioral selectivity for signals (Baugh and Ryan, 2010; Bernal et al., 2007). Male túngara frogs vocalize both to attract mates and to engage in male–male competition. Sexually receptive females respond to mating calls by phonotaxis, i.e. by approaching an attractive stimulus. Reproductive males typically respond to the same suite of conspecific signals by evoked calling, i.e. vocalizing alternately with the competitor. Laboratory assays of these sex-typical behaviors (phonotaxis assays for females and evoked playback assays for males) demonstrate sex differences in selectivity for signals, with males responding to a broader range of signals than females (Bernal et al., 2007). When not given the opportunity to vocalize, males will perform phonotaxis – a behavioral output in which they show similar selectivity for stimuli as females (Bernal et al., 2009). Despite similar signal selectivity, male and female túngara frogs differ in the motor patterns that comprise a phonotaxis response. Both males and females readily approach speakers playing conspecific signals as adults but only reproductively active females continue to engage in extensive locomotor activity near the source of a conspecific call (Baugh and Ryan, 2010). Male and female túngara frogs therefore demonstrate a number of sex differences in behavioral responses to social signals depending on the experimental and reproductive context, including sex-typical behavioral response, selectivity for signals and pattern of locomotive activity.

Neural activation at a proposed sensory–motor interface in the midbrain (Walkowiak and Luksch, 1994), the laminar nucleus of the torus semicircularis (homolog of the inferior colliculus), shows sexually dimorphic selectivity for signals using egr-1 mRNA levels as a measure of neural response; a conspecific mating call elicits responses in both males and females whereas a heterospecific call elicits a significantly elevated response only in males (Hoke et al., 2008). The laminar nucleus receives extensive auditory inputs from midbrain and hindbrain auditory nuclei and acts as the primary output region of the torus with widespread connections with forebrain regions (Wilczynski and Endepols, 2007). The laminar nucleus probably acts as a sensory–motor gateway, relaying auditory information both to brainstem motor centers in the tegmentum and to telencephalic areas that modulate motor output via multiple pathways through the thalamus and hypothalamus (Wilczynski and Endepols, 2007). Lesioning the torus semicircularis blocks phonotaxis responses of females, although effects of lesioning specific toral nuclei have not been described (Endepols et al., 2003). The sexual dimorphism in the signal specificity of laminar responses suggests that sex differences in responsiveness at this putative sensory–motor gateway could regulate behavioral selectivity for signals. However, it does not explain the sex differences in which behaviors are typically evoked by the signals, nor in the different motor patterns that comprise behavioral responses in males and females. Two explanations for this aspect of sex differences in social behavior are that (1) the midbrain's relay of information to forebrain areas controlling behavior differs in the sexes, and (2) forebrain areas control behavior differently in the sexes. If the former is true, we would predict that the relationship between activity in the midbrain and forebrain would be different in males and females. If the latter is true, we would predict that the relationship between functional patterns in the forebrain and patterns of behavior would differ in the sexes. The two possibilities are not mutually exclusive.

In this study, we extended our previous analysis of sex differences in the auditory system to the forebrain to test the hypothesis that sex differences in sensory–motor relays underlie sexually dimorphic locomotive responses to social signals. We used the immediate-early gene egr-1 as an indicator of neural activation throughout the brain. We compared both mean egr-1 responses and auditory–forebrain relationships in males and females, and we related activation patterns to sex differences in locomotive responses to signals. Through our analyses, we linked activation of sensory systems to behavior through intervening motor systems, and we identified the most likely point in the system responsible for sex differences in distinct aspects of phonotaxis behavior.

Animal procedures were approved by both University of Texas IACUC and Autoridad Nacional del Ambiente del República de Panamá. The animals included in this experiments are the same individuals whose auditory system we analyzed previously (Hoke et al., 2008). We collected amplexed pairs of male and female túngara frogs (Physalaemus pustulosus, Cope 1864) at mating ponds near Gamboa, Panama, to ensure reproductive condition. Frogs were separated and placed in dark, sound isolation chambers for two hours to decrease baseline levels of egr-1 mRNA. After the waiting period, frogs were assigned to one of three acoustic conditions: no acoustic stimulation (silence), calls from heterospecific Physalaemus petersi (Jiménez de la Espada 1872) males, or calls from conspecific P. pustulosus males (Fig. 1). Sample sizes ranged from 9 to 11 frogs of each sex in each stimulus condition. Acoustic stimuli consisted of a single exemplar of each call repeatedly broadcast every second from speakers on alternate sides of the chamber (SME-AFS, Saul Mineroff Electronics, Elmont, NY, USA). Amplitudes were adjusted to be 82 dB SPL (re. 20 μP) in the center of the chamber, which measured 0.7 m in length.

We videotaped behavioral responses of frogs during the 30-minute stimulus presentation using infrared cameras (PC-6EX-2 IR video camera, Supercircuits, Liberty Hill, TX, USA) mounted on the chamber ceiling and captured images using a ZR60 miniDV digital camcorder (Canon, Lake Success, NY, USA). For behavioral analysis of the videotapes, we recorded the amount of time each frog was in motion using a stopwatch and estimated association with the speaker by noting every 30 s the location of the frog on a grid overlaying the monitor. We calculated the association time by determining the proportion of time the frog was located in the parts of the chamber abutting the speaker rather than the central portion.

We assessed neural responses in the brain using mRNA expression of egr-1, an immediate-early gene with mRNA or protein levels that increase upon activation of neurons. Egr-1 levels are related nonlinearly to a neuron's electrical activity, as egr-1 induction in a neuron requires depolarization but also relies on the complement of co-factors present in the cell (Clayton, 2000; Jarvis, 2004; Knapska and Kaczmarek, 2004). Immediately following the 30-minute acoustic treatment, we decapitated animals and froze their heads embedded in Tissue-Tek O.C.T. compound (Sakura Finetek, Torrance, CA, USA) in liquid nitrogen. We later processed brains for radioactive in situ hybridization for P. pustulosus egr-1 and quantified silver grain density as previously described (Hoke et al., 2004; Hoke et al., 2008). Briefly, our quantification entailed capturing high magnification images sampled at specific sites throughout the brain (3–15 per brain region) and using automated image processing methods to save separate images of Nissl-stained cell bodies and the silver grains over cells. We then calculated silver grain density as the proportion of the area of cells covered in silver grains. We reported egr-1 mRNA levels in the auditory system (Hoke et al., 2008), then extended our analyses of the same brains to encompass 25 forebrain regions in the same individuals using unbiased sampling methods detailed previously (Hoke et al., 2005; Hoke et al., 2007) (supplementary material Fig. S1).

Analysis of the data was done in a set of sequential steps to investigate potential sex differences at several levels. First, we used univariate analyses of covariance (ANCOVAs) to identify parts of the brain that function differently in males and females. We also used analyses of variance (ANOVAs) to test for sex differences in locomotive responses to stimuli. We then used resampling approaches to ask whether the relationships between activity measurements in subsets of brain regions differ in males and females. We complemented resampling approaches with structural equation modeling (SEM) to test hypotheses about sex differences in information flow in the brain.

We examined sex differences in the egr-1 responses of each brain region separately using univariate ANCOVAs with SPSS 11 (SPSS, Inc., Chicago, IL, USA). The dependent variables were the natural logarithms of the mean egr-1 levels in each brain region for each individual. None of the log-transformed variables significantly deviated from a normal distribution, as assessed by the Shapiro–Wilks test. We tested for main effects of sex, stimulus and the sex × stimulus interaction. We included as covariates a variable representing overall brain activation for the individual (Hoke et al., 2007; Hoke et al., 2008) and the proportion of time in motion. Because the relationship between the movement covariate and dependent variable might vary with sex or stimulus, we included three additional interaction terms: stimulus × movement, sex × movement, and sex × stimulus × movement. We retained these three interactions of main effects terms and movement covariate in the analyses if the interaction explained variation in the dependent variable with P<0.2. For those cases in which we retained the three-way interaction (sex × stimulus × movement), we also retained both two-way interactions. If interaction terms were not retained (P>0.2) we then re-ran ANCOVA analyses without the interactions and present the latter results here.

To determine how males and females differed in their behavioral responses to acoustic stimuli, we used ANOVAs with either proportion time in motion or proportion time associated with the speaker as dependent variables and main effects of sex and stimulus as well as the sex × stimulus interaction term. For analyses in which the effect of stimulus condition was significant, we used post-hoc pairwise analyses and corrected for multiple comparisons using Tukey's HSD to isolate which stimuli evoked differences in behavior. We further assessed significant sex × stimulus interactions using planned comparisons to assess sex differences within each stimulus condition.

We used resampling approaches in R in an extension of our previously described methods (Hoke et al., 2007) to ask where sex differences in the brain arise, i.e. whether auditory–forebrain relationships, within forebrain networks, and/or forebrain–behavior associations differed between males and females. Resampling is a flexible method of comparing group differences in user-defined measurements using random permutation of the dataset to create the null distribution against which to test statistical significance of a comparison. We calculated an index of association, R_SqDiff, to summarize the sex differences in the individual pairwise correlation coefficients comprising each comparison as detailed below. We then compared the actual association index with the range of indices found by random permutation of the sex of each individual in the data set 2000 times without replacement, reassigning each individual as either male or female to reflect the overall sex ratio in the original data set on every round. We determined how often the measured sex differences would occur by chance, with the P-value calculated as the proportion of group differences in the original data set plus permutations that were as large as the calculated sex difference.

We first tested the hypothesis that the set of midbrain auditory–forebrain correlations differed in the sexes, specifically focusing on the laminar nucleus as the major output area of the midbrain auditory system (Wilczynski and Endepols, 2007). To do so, we calculated R_SqDiff as the sum over all forebrain regions of the squared difference between the Pearson correlation coefficient of laminar egr-1 levels with egr-1 levels in that forebrain region in females and the corresponding coefficient in males.

We next addressed whether there were sex differences in the functional associations within forebrain networks to test the hypothesis that forebrain systems operate differently in males and females. We tested the hypothesis that correlations among all pairs of forebrain regions differed significantly in males and females. We calculated the full Pearson correlation matrix describing relationships among egr-1 mRNA levels in each pair of forebrain regions in males and females. We then calculated R_SqDiff as the sum over all off-diagonal elements in the correlation matrix of the squared difference between the corresponding correlation coefficients in males and females.

We finally asked whether the relationship between forebrain activation and behavior was sexually dimorphic. We tested the hypothesis that males and females differed significantly in how egr-1 mRNA levels in the group of forebrain regions correlated with time in motion. We calculated R_SqDiff as the sum over all brain regions of the squared difference in the Pearson correlation coefficient between egr-1 levels in that forebrain region and the proportion of time in motion for males and females.

The resampling methods detailed above allowed us to test for sex differences in networks of interacting brain regions, but did not enable us to consider simultaneously co-activation patterns across the brain and isolate specific pathways that functioned differently in males and females. To determine the sexually dimorphic functional pathways responsible for the sex differences in behavior, we compared connectivity of neural systems using SEM. SEM is an extension of linear regression in which the researcher tests and compares different models of the relationships among variables. Models proposed relationships between variables based on anatomical connections, including both direct causal effects of one variable on another and covariation between variables that were not detailed explicitly in the model (correlated error terms). Direct relationships are termed paths, and the strengths of those relationships, akin to regression coefficients, are termed path coefficients.

SEM enables simultaneous tests of sex differences in auditory–forebrain, forebrain or forebrain–behavior relationships but precludes modeling each brain region individually given our sample sizes. To reduce the number of variables, we summarized egr-1 levels for each forebrain division using principal component analysis. We calculated principal components representing activation in the three forebrain divisions separately (thalamus, hypothalamus and telencephalon) using SPSS 11 (SPSS, Inc.). We prioritized by including all subjects over including all brain regions, and thus excluded the following brain regions that had several missing values: rostral striatum (rST), dorsal pallium (DP), dorsal hypothalamus (DH) and central thalamus (Cthal). We ran three separate principal components analyses for the three forebrain divisions, each based on the logarithm of the mean egr-1 levels in the brain regions in that division. A single component for each division had an eigenvalue greater than 1, thus we represented each brain division by a single component. We calculated components by regression with cases excluded pairwise. Principal components calculated with and without the above excluded variables were highly correlated (correlation coefficient 0.980–0.998), thus we concluded that omitting these variables did not alter the generality of these components. A single component in each division explained more than 65% of the variance in egr-1 levels in the telencephalon, hypothalamus and thalamus. We used those three components in SEM analyses along with the egr-1 levels in the laminar nucleus and the percentage of time each frog moved.

We tested two types of SEM models – structured and unstructured. Both model types assumed a basic flow of information from the laminar nucleus to the forebrain areas to regulate locomotive behavior. The unstructured model did not postulate any particular relationship among forebrain areas whereas the structured model further assumed that the thalamus and hypothalamus received laminar inputs and relayed them to the telencephalon, which directly influenced locomotion. We based the structured model on the general view of the sensory–motor pathways underlying acoustic communication in frogs (Wilczynski and Endepols, 2007), and included the unstructured model due to the many anatomical connections among brain regions that are not detailed in the structured model. Due to mathematical limitations that require path diagrams to be non-recursive, we did not include any of the feedback pathways from the forebrain to the auditory system.

All SEM model estimation was run with Mplus using the MLM estimator (Muthén & Muthén, Los Angeles, CA, USA). After estimation of all of the model parameters, we tested each model for model fit based on how well the implied covariance matrix of that model matched the actual matrix, thus allowing us to reject models in which the proposed relationships among variables did not match the measured relationships. The overall model fit was tested by a χ² statistic as well as by close fit indices. We report the Chi-square test, with P-values less than 0.05 indicating poor model fit. Two close fit indices were assessed: Bentler's Comparative Fit Index (CFI), which compares model fit with baseline model positing zero relationship among variables, and the standardized Root Mean Square Residual (SRMR), a measure of the total amount of covariance explained. Commonly used criteria for good model fit include CFI>0.96 and SRMR<0.1 (Hu and Bentler, 1999). In addition, we report the Akaike Information Criteria (AIC) for comparisons of non-nested models. We tested unconstrained models in which all paths were estimated separately in males and females, and compared those unconstrained models with nested ones in which some or all paths were constrained to be equal in the sexes. If a constrained model had poor fit while the unconstrained model did not, then we concluded that one or more of the constrained paths were significantly different in males and females.

We found sex differences in neural activation in a number of forebrain regions that we selected based upon their proposed roles in motivation or behavior. In a subset of forebrain regions (supplementary material Table S1) egr-1 mRNA levels consistently covaried with the locomotive responses of the individual and the acoustic stimulus to which it was subjected. Seven brain regions had significant main effects of sex [lateral pallium (LP), caudal striatum (cST), medial septum (MS), nucleus accumbens (NA), medial amygdala (MA), anterior preoptic area (aPOA), suprachiasmatic nucleus (SC); Fig. 2]. The lateral septum (LS), central septum (CS) and MA had significant interactions of sex and stimulus in predicting egr-1 levels (Fig. 2). In addition, the relationships between egr-1 level and locomotive behaviors differed in males and females in five brain regions (LP, rST, CS, MA, aPOA; sex × movement or sex × stimulus × movement interactions; not shown). Males and females did not consistently differ in the range or variability in egr-1 levels across the brain (supplementary material Fig. S2). Together, these results demonstrate widespread sex differences in egr-1 induction in forebrain regions, either on overall levels, or on the patterns of stimulus- or motor-related activation.

We found evidence that males and females differ in the relaying of auditory information to the forebrain. We examined the correlations between forebrain egr-1 levels and the egr-1 levels in the major auditory input to the forebrain, the laminar nucleus of the torus semicircularis (Fig. 3). Resampling demonstrated that the measured index of sex differences, R_SqDiff, was significantly higher than expected from the null distribution based on randomly assigning individuals to male and female groups (P=0.017); thus, the associations between egr-1 levels in the laminar nucleus and each forebrain region were significantly higher in females than in males. This laminar–forebrain association was unique. If we selected any of the 25 forebrain regions and compared its correlations with all of the other forebrain regions, none showed significant sex differences (all P>0.05). Thus, there was a sex difference in the correlation between egr-1 levels in a midbrain auditory relay, the laminar nucleus and the responses in the forebrain.

Sex differences in auditory–forebrain relationships do not coincide with sex differences in the pairwise associations between different forebrain regions. Using resampling to compare all pairwise correlations among forebrain areas, we found that intra-forebrain correlations did not differ consistently in males and females (R_SqDiff=21.08, P=0.516; Fig. 4). The higher auditory–forebrain covariation in females was therefore not indicative of widespread differences in correlations across forebrain networks. Instead, once activated, the functional networks within the forebrain operate similarly in males and females.

Males and females do not fundamentally differ in the relationship between forebrain egr-1 mRNA levels and behavioral responses. We found no significant difference between males and females in the correlations between forebrain activation patterns and the proportion of time the frog spent moving using resampling (R_SqDiff=2.47, P=0.097; Fig. 5). Despite several forebrain regions with significant interactions between sex and movement in an ANCOVA, we did not find evidence for a widespread difference in males and females in how forebrain activation is translated into motor output.

The differential stimulus specificity of the laminar nucleus in males and females (Hoke et al., 2008), along with the sexually dimorphic associations between the laminar nucleus and forebrain egr-1 levels, is reflected in sex differences in locomotive responses to advertisement calls. We found a significant difference in the amount males and females moved in response to the three stimulus conditions using ANOVA (Fig. 6A, sex × stimulus interaction: F_2,52=6.605, P=0.003; main effect of sex: F_1,52=7.491, P=0.008; main effect of stimulus: F_2,52=0.595, P=0.555). Planned comparisons of estimated marginal means determined that the sexes differed specifically in their locomotive responses to conspecific signals (F_1,52=18.946, P<0.001 for sex differences in behavioral responses to conspecific calls; F_1,52=0.618, P=0.435 for locomotion without acoustic stimulation; F_1,52=1.525, P=0.222 for locomotion in response to heterospecific calls). By contrast, we did not find sex differences in the time frogs spent near the speaker using ANOVA (Fig. 6B, main effect of sex: F_1,52=0.525, P=0.472; sex × stimulus interaction: F_2,52=0.285, P=0.753). Males and females spent more time in proximity to the speakers in response to the conspecific calls than in either the heterospecific or silence conditions (main effect of stimulus: F_2,52=5.815, P=0.005; post-hoc pairwise comparisons: silence vs heterospecific P=0.974; silence vs conspecific P=0.011; heterospecific vs conspecific P=0.021). Thus, we found sex differences in locomotive responses to conspecific signals (but not other signals) but not in association time with the sound source.

We used SEM to relate auditory system activation to behavioral output through forebrain networks as a complement to the resampling approach. Both structured and unstructured models fit the data well (Table 1), suggesting that these simplified models (e.g. lacking feedback pathways) are able to approximate well the measured covariance matrix and thus are amenable to testing the hypothesis about sex differences in particular pathways in the brain.

To localize sex differences in SEM models, we tested a series of models in which different subsets of paths were constrained to be equal across sexes. Models in which all paths are constrained equally were rejected based on poor fit (U2 and S2, Table 1). Constraining any one of the laminar–forebrain relationships to be equal in males and females also resulted in poor model fit (U4, U5, U6, S4 and S5; Table 1). By contrast, models U3 and S3, in which direct paths between the laminar nucleus and forebrain were freely estimated in males and females while all other paths were constrained to be equal, adequately described relationships among variables (Table 1). Taken together, these results suggest that, if the models accurately describe information processing, males and females differ in their auditory–forebrain relationships but not in forebrain–forebrain networks or in forebrain influences on locomotive behaviors (Fig. 7B,C). This is consistent with the resampling results detailed above (summarized in Fig. 7A).

Our results support the hypothesis that sex differences in behavioral responses to social stimuli stemmed from sex differences in the relay of auditory information from the midbrain to forebrain circuits. There was a linear relationship between egr-1 levels in the laminar nucleus and in its forebrain targets in females but not in males. These functional networks within the forebrain otherwise operated similarly in males and females as assessed by correlated activation patterns, despite sex differences in mean egr-1 levels in many forebrain regions. Moreover, we did not find evidence for sex differences in the relationship between egr-1 induction in these forebrain networks and locomotive behavior, although the sexes differed in locomotive responses to signals. We propose that sex differences both in the selectivity of a midbrain sensory–motor gateway (Hoke et al., 2008) and in the influence of this gateway on forebrain targets together produce sex differences in neural activation within forebrain regions that are associated with sexually dimorphic behavioral responses to social signals.

The experimental design we used offered a unique opportunity to relate variation in sensory responses to variation in behavioral output through a complex network of brain regions using two types of multivariate analyses. Resampling statistics considered the activation measure of each forebrain region individually, and few assumptions were made as to the specific roles of each region. This was followed by an SEM approach that allowed a global analysis that spanned activation in sensory regions, forebrain network patterns and consequent behavioral outputs to isolate components of the functional network with sex differences (after summarizing average activity across each brain division due to sample size requirements). Despite the distinctly different statistical requirements of each approach, we found entirely consistent results; network covariation patterns did not differ significantly in males and females within the forebrain and between forebrain areas and behavior whereas the sexes differed in auditory–forebrain relationships.

We have focused on sex differences in correlations as a means to argue that neural systems function differently in the two sexes, and to propose specific locations and pathways that warrant more detailed physiological characterization. Network covariation patterns, or connectivity, are typically estimated using physiological measures such as electrophysiology or functional magnetic resonance imaging in which the researcher measures responses to multiple stimuli within each individual and determines how tightly correlated activity levels are in two different cells or different brain regions across stimulus presentations. Given that in our experiments each animal is subject to only one stimulus condition, our measure of connectivity is intrinsically different, as we compare correlations across individuals. Patterns of correlated egr-1 expression could arise because of direct synaptic coupling between brain regions, because of shared common modulatory inputs or because of some generalized mechanisms that influence egr-1 induction across the brain (e.g. hormone levels or neuromodulators). Moreover, given the long time scale of egr-1 induction, we cannot demonstrate that auditory activation preceded forebrain activation, thus we cannot determine whether ascending synaptic connections underlie this pattern of covariation. Due to the nonlinearities in egr-1 expression, we additionally cannot rule out the possibility that the lack of auditory–forebrain covariation in males could reflect differences in cofactors regulating egr-1 induction in some brain region rather than differences in correlated electrical activities, as electrical activation does not always induce egr-1 expression (Clayton, 2000; Jarvis, 2004; Mello and Riberio, 1998). The lack of widespread differences in the overall means or variances of egr-1 levels in males and females (supplementary material Fig. S2) does suggest that a lack of covariation in males is not the trivial consequence of a lack of egr-1 induction. We suggest that follow-up anatomical and physiological studies focus on laminar–forebrain pathways to detail cellular mechanisms for the sex differences we found in this study.

How do the sex differences we find in auditory–forebrain relationships correspond to the sex differences in social behavior? Sex differences in social behavior encompass (1) differences in the propensity to perform a particular behavioral response; (2) differences in the signal specificity in eliciting a response; and (3) differences in the motor patterns that comprise a given behavioral response. The difference in sex-typical behavioral responses in frogs is well established; in túngara frogs, as in most frog species, the males are the only ones that produce advertisement calls in response to calls, and females typically respond by phonotaxis (Ryan, 1985). The sexes also differ in signal specificity in túngara frogs, with males less selective in vocal responses than are females in phonotaxis responses (Bernal et al., 2007). Note, however, that this specificity difference depends on behavioral output. Bernal et al. (Bernal et al., 2009) found that male and female túngara frogs display similar signal specificity when performing phonotaxis responses, suggesting that the greater selectivity of females in their previous study (Bernal et al., 2007) is, in fact, due to a task difference (vocalization vs phonotaxis) rather than a sex difference per se. In their work demonstrating the lack of sex difference in phonotaxis responses, Bernal et al. used the dichotomous outcome of phonotaxis as an indicator of behavioral response (Bernal et al., 2009). A choice was defined as approaching within 10 cm of a speaker. A subsequent paper found sex differences in a quantitative measure of phonotaxis behavior, locomotive perseverance or the amount of locomotor activity performed by males and females in the region adjacent to the speakers as measured by path length, despite similar overall percentages of individuals that approached speakers playing conspecific signals (Baugh and Ryan, 2010). We found concordant sex differences in the motor responses to a conspecific signal; males approached the speaker and ceased moving whereas females approached the speaker and engaged in searching behavior.

The egr-1 expression patterns we described reflect these different aspects of sex differences in social behavior. Egr-1 induction within the laminar nucleus (Hoke et al., 2008) parallels the sex difference in behavioral selectivity when the animals are performing sex-typical responses (Bernal et al., 2007); egr-1 levels are high only in females that heard conspecific signals whereas both conspecific and heterospecific P. petersi calls induce egr-1 mRNA in males. Laminar nucleus expression does not predict the sex difference in motor output during phonotaxis – this difference emerged within laminar targets in the forebrain. Our data suggest that in females but not in males, the forebrain regions regulating phonotaxis depend on laminar inputs to forebrain motor areas in a straightforward fashion. The sex difference in signal selectivity within the laminar nucleus would be transformed into a distinct sex difference in the patterns of egr-1 responses in forebrain targets based on the suggested sexually dimorphic influence of laminar inputs on forebrain regions. Because the pattern of activation in forebrain areas is strongly related to motor output in both males and females, the sex differences in forebrain regions mirror the sex differences in behavior. We propose that the difference in signal selectivity in the laminar nucleus combined with differences in laminar–forebrain associations may fully account for sex differences in the motor output performed.

Our results extend our understanding of how sex differences in social behaviors can arise by differential activation of motor control circuits that are present in both sexes. The well-established mechanisms of sex differences in vocalizations of fish, frogs and birds depend on sex differences in the anatomy and function of motor circuits controlling vocalization (reviewed in Ball and MacDougall-Shackleton, 2001; Forlano et al., 2007; Zornik and Yamaguchi, 2008). In each of these cases, however, the motor neurons involved in the behavioral output are dedicated to the production of vocalizations. By contrast, sexually dimorphic gating of common motor circuits has been proposed to distinguish male-typical and female-typical behaviors in three systems in which the motor neurons generating the social behavior also contribute to other behavioral outputs: mounting behavior in mice, courtship behaviors in flies, and phonotactic locomotion in frogs. In mice, eliminating a pheromonal input in females results in the expression of male-typical mounting behavior (Kimchi et al., 2007), suggesting that motor circuits regulating male-typical behaviors are normally inhibited in female mice, although these experimental findings were not replicated in a subsequent study (Martel and Baum, 2009). In flies, males and females differ in their behavioral responses to a sex pheromone (Kurtovic et al., 2007) as a consequence of differences in the projection pattern of the olfactory neurons (Datta et al., 2008). Experimentally activating song motor commands in females produces song by patterning wing vibrations (Cline and Miesenbock, 2008), suggesting that female flies lack stimulatory inputs to drive the functional motor circuit that regulates male-typical singing behavior. Our results expand on the previous work in other systems by examining the functional activation of the neural systems that regulate female-typical behavioral responses. We found evidence that sensory-motor gating differed in males and females in evoking female-typical responses to acoustic signals. We detailed how the activation of this sensory–motor gateway differed both in its overall selectivity for conspecific signals (Hoke et al., 2008), and in the relationship between sensory inputs and forebrain motor-control circuits (this study). We speculate that sex differences in social behaviors that originate via sexually dimorphic sensory gating rather than dimorphic motor systems may be the consequence of multifunctionality of motor centers; perhaps the sensory–motor connections are dedicated to the particular behavioral response despite linking multifunctional auditory and locomotive circuits, and thus this sensory–motor gateway may be a malleable site for modulation by sex, reproductive condition, learning or evolution (Hoke and Wilczynski, 2010).

Telencephalon (Telenceph):

 
• AA

  anterior amygdala

 
• CS

  central septum

 
• cST

  caudal striatum

 
• DP

  dorsal pallium

 
• DS

  dorsal septum

 
• LA

  lateral amygdala

 
• LP

  lateral pallium

 
• LS

  lateral septum

 
• MA

  medial amygdala

 
• MP

  medial pallium

 
• MS

  medial septum

 
• NA

  nucleus accumbens

 
• rST

  rostral striatum

Hypothalamus (Hypothal):

 
• aPOA

  anterior preoptic area

 
• DH

  dorsal hypothalamus

 
• NP

  periventricular nucleus

 
• pPOA

  posterior preoptic area

 
• PT

  posterior tuberculum

 
• SC

  suprachiasmatic nucleus

 
• VH

  ventral hypothalamus

Thalamus (Thal):

 
• Athal

  anterior thalamus

 
• Cthal

  central thalamus

 
• Pthal

  posterior thalamus

 
• VMthal

  ventromedial thalamus