Complex choice environments shelter unattractive signallers from sexual selection

Abstract

For many animals, options abound when choosing a mate in socially complex environments like a breeding chorus or lek. In such environments, receivers often choose their mate based on individual differences in signal repetition rate. However, signallers also differ in the regularity with which they produce repeated signals. Irregularity in signalling introduces uncertainty in decision-making by masking the among-individual variation in signalling rate that is a target of mate choice. At present, we know little about how the complexity of the choice environment affects selection on rate and regularity, two signalling behaviours that receivers can only compare after sampling series of signals produced by multiple signallers. In this study of female grey treefrogs (Hyla chrysoscelis), we measured multivariate sexual selection on the rate and regularity of male calling behaviour using two-, four- and eight-choice tests. Receivers overwhelmingly chose faster, more regular calling rates in two-choice tests, but did so markedly less often when they chose among four or eight stimuli. Sexual selection imposed by female choice became markedly weaker and differently shaped as the complexity of the social environment increased, suggesting noise and choice overload effects may allow relatively unattractive males to mate.

1. Introduction

Communication mediates animal decision-making with important fitness consequences, including mate choice [1–3]. Sexual advertisement signals function to broadcast the signaller’s identity, quality, and location to receivers, who use the signals to identify and discriminate among potential mates [4–6]. Through mate choice, receivers impose sexual selection on signals and signalling behaviour [5,7,8]. One of the most ubiquitous findings in experimental studies of mate choice across diverse taxa—from widowbirds [1] to tortoises [9] to wolf spiders [10]—is that receivers respond more often and more quickly to repeated sexual advertisement signals produced at higher repetition rates (reviewed in [11]). These studies suggest directional selection for faster signalling rates is common in the context of mate choice, and field studies confirm that males with faster signalling rates can have higher reproductive success [1,12].

A key but seldom considered aspect of signalling rate as a basis for mate choice is that rate is a temporally emergent property of signalling behaviour: comparing individual differences in signalling rate requires receivers to accurately sample multiple, sequentially produced signals from multiple signallers. Hence, mating decisions based on signalling rate necessarily involve perceptual and cognitive processes (e.g. perceptual grouping and working memory) for assigning signals to their correct sources and tracking them through time. Another important but understudied aspect of signalling rate concerns the regularity with which signals are repeated (hereafter, we will refer to regularity in signal repetition rate, specifically, as simply ‘regularity’). More regular signals are those repeated with more consistent timing between signals. Two additional issues arise when considering signalling regularity. First, the challenge that receivers face in accurately assessing among-individual differences in signalling rate is expected to scale inversely with the degree of regularity in signal repetition. Hence, low regularity in signalling (i.e. high within-individual variation) can mask the among-individual variation in signalling rate upon which receivers base their mating decisions [13,14]. Second, in addition to among-individual differences in their average signalling rates, signallers can also differ in regularity. In at least some taxa, receivers treat regularity itself as an important aspect of signalling behaviour, strongly preferring regular over irregular signalling when average signalling rate is held constant [15] (but see [16]). Understanding the fitness landscape of sexual signalling, therefore, necessitates taking a multivariate approach (e.g. [17–20]) that considers temporally emergent properties, such as signalling rate and regularity, to be potentially important components of sexual displays that influence mating decisions.

An additional factor to consider in estimating selection on temporally emergent properties is that receivers often exercise mate choice in relatively complex environments, where multiple signallers compete for receiver attention [21–23]. Leks, in which receivers choose a mate among competing signallers who provide no parental care to their offspring [24], represent a relatively extreme case of a choice among many potential options whose outcome strongly affects individual fitness. Historically, receiver preferences and mate choice have usually been studied under simplified testing conditions, in which experimenters offer subjects either no choice (i.e. one repeated stimulus) or a choice between two alternative stimuli (reviewed in [25]) and in which stimuli are designed to eliminate within-individual variation. Such experiments are ecologically unrealistic because they do not reflect the complexity of decision problems many receivers face in nature [26]. In humans, ‘choice overload’ can occur when the complexity of a decision problem exceeds an individual’s cognitive resources [27–30]. Choice overload has been comparatively understudied in non-humans (reviewed in [31]). Only a handful of studies have explicitly addressed choice overload in non-human animals, and all of these examine invertebrates, including Pacific field crickets [32], acorn ants [33] and mosquitoes [34]. However, in many lek-breeding animals, offering more options to choose from often results in worse choice outcomes and a weakening of preferences based on the properties of an individual signal (e.g. its frequency, amplitude or duration) compared with preferences measured in simpler environments. Such outcomes are consistent with choice overload [27] and are often attributed to the effects of ‘confusion’ (reviewed in [35]).

Here, we investigated how females exert multivariate selection through mate choice on two temporally emergent properties of signalling behaviour—calling rate and regularity—in choice environments that varied from relatively simple (two options) to more complex (eight options). An alternative way to frame this is that we examined the fitness of male signalling phenotypes that varied in calling rate and regularity across three choice environments, ranging from simple (two signallers) to more complex (four and eight signallers). We posited three alternative hypotheses about receiver performance that make different predictions about how selection should change across environments. The choice overload hypothesis holds that a receiver’s performance in selecting preferred signallers deteriorates as the number of signallers increases. This hypothesis predicts sexual selection on calling rate, and regularity should weaken in more complex environments that are characterized by more available options. Alternatively, the more is better hypothesis holds that a receiver’s performance in selecting preferred signallers based on temporally emergent features improves as the number of signallers increases [36,37]. According to this hypothesis, sexual selection on calling rate and regularity should become stronger as the number of available options increases. Such an outcome might be expected, for example, if more complex signalling environments facilitate direct comparisons of two or more signallers, uncover call features relevant primarily for tracking signallers in noise or increase the likelihood of encountering signallers with more attractive trait combinations. Finally, a receiver’s ability to behaviourally discriminate among signallers may not change as the number of signallers varies over some biologically relevant range. This null hypothesis predicts sexual selection on calling rate and regularity should remain unchanged as environments become more complex. We term this null hypothesis the performance maintenance hypothesis, because under this hypothesis, while receivers would not experience any of the changes in decision-making indicative of choice overload, they would also show none of the increased selectivity predicted by the more is better hypothesis.

We tested these hypotheses experimentally in the diploid grey treefrog, Hyla chrysoscelis, a lek-breeding North American species that relies entirely on its advertisement calls for mate discrimination and localization [4,38,39]. In simple choice environments (single- or two-stimulus tests), females of this species prefer advertisement calls produced at faster and more regular rates [14,15,20]. The natural environment in which female grey treefrogs choose a mate is a noisy breeding chorus where many potential mates can be sampled simultaneously. Grey treefrog females choose a mate by approaching and entering amplexus with a stationary calling male in the chorus. In the laboratory, they reliably exhibit phonotaxis towards speakers broadcasting synthetic (computer-generated) advertisement calls, allowing us to generate simulated male phenotypes. By drawing trait values at random from normal distributions, we generated many unique simulated males differing in calling rate and regularity. We calculated relative fitness scores (fitness of a given organism relative to other organisms in the same population) for each simulated phenotype based its ability to attract a gravid female in playback tests. We performed our experiment with two, four and eight stimuli to estimate linear, quadratic and correlational selection gradients comparing the strength and shape of multivariate phenotypic selection on calling rate and regularity across increasingly complex choice environments.

2. Methods

(a). Subjects

We tested 125 gravid females that were captured in amplexus at night (2200−0100 h) between 15 May and 7 June 2023. All subjects were captured at Carver Park Reserve (44°52′49.0″ N, 93°43′03.1″ W; Carver County, MN) or Crow-Hassan Park Reserve (45°11'16.5″ N, −93°39' 13.4″ W; Hennepin County, MN) in east-central Minnesota. Each amplectant pair of frogs was housed in a clean plastic container and taken to the laboratory where they were given aged tap water and cooled to 4°C to delay egg deposition prior to testing [40]. Animals caught in amplexus are no less discriminating than females caught prior to mate choice [41]. Before testing, pairs were given room-temperature water and transferred to an incubator, where they were kept for at least 30 min to reach a body temperature of 20°C ± 1°C. Following testing, animals were released at their points of capture within 48 h after collection.

(b). Experimental design, stimuli and playback

Our overall goal was to assess how choice set size affects the performance of females choosing mates based on temporally emergent properties of signalling behaviour, and hence the strength of sexual selection imposed on those traits. To this end, we measured the relative fitness of simulated calling males with signalling phenotypes differing in calling rate and regularity across choice contexts that differed in the number of stimuli (2, 4 or 8). We used synthetic stimuli designed to break up the natural positive correlation between rate and regularity (i.e. males with faster call rates also tend to call more regularly [14]). Stimulus values for calling rate and regularity were derived from previously published analyses of 1000 calls recorded from 50 calling males (20 calls per male) in our local Minnesota population after standardizing call properties to 20°C [42]. In these previous analyses, each call from a recorded male was assigned an ‘instantaneous calling rate’, computed as the inverse call period of the given call. This approach allowed for calculating estimates of regularity in calling rate for each individual, which was determined as the coefficient of variation within males (CV_w = standard deviation in instantaneous calling rate divided by the mean instantaneous calling rate or $\frac{\sigma }{\stackrel{-}{x}}$). The grand mean calling rate in our local population is 11.39 ± 2.91 calls per minute (cpm) and the grand mean CV_w is 0.306 ± 0.95 [42].

We generated 64 ‘stimulus sets’ of 8 unique acoustic stimuli to simulate a total of 512 calling males; no more than two subjects were assigned to the same set of 8 stimuli in choice tests. To begin, we generated normal distributions representing the distributions of calling rate and calling regularity in the population. We truncated both distributions at the limits of natural variation because we wished to examine selection on extant variation in the population. We, thus, avoided generating simulated males with mean values of calling rate and CV_w that exceeded the range of natural variation in our population (mean calling rate range: 5−20 cpm; CV_w range: 0.105−0.584). Next, to create each simulated male, we drew his mean calling rate at random from the normal distribution of calling rates (figure 1A) and his CV_w from the normal distribution of CV_w values (figure 1B). For each simulated male, we back-calculated the standard deviation in calling rate that resulted from his drawn mean calling rate and CV_w. These values were used to generate a unique distribution of instantaneous calling rates for each simulated male after truncating to exclude supernormal draws (instantaneous calling rate range: 0.31−31.19 cpm). In this manner, we established for each of the 512 unique simulated males a distribution of possible instantaneous calling rates with a mean calling rate and a standard deviation that resulted from its CV_w (figure 1C). Finally, to generate an audio track for the playback experiment, we concatenated a series of identical synthetic calls (figure 2A) using a series of instantaneous calling rates drawn randomly from each simulated male’s distribution until the audio track was longer than 11 min (figures 1D and 2B).

Figure 1.

Methods for generating simulated male phenotypes and measuring their relative fitness. The figure follows a representative set of eight unique stimuli used in two-, four- and eight-choice tests. (A) The distribution of calling rates (mean 11.39, s.d. 2.91) in the Minnesota population (black curve) and the two random draws used to create the simulated males (colours). (B) The distribution of regularity (coefficients of variation within males (CV_w); mean 0.306, s.d. 0.095) in the population (black curve) and two random draws used to create the stimuli (colours). (C) The distributions representing the two simulated males were created using the randomly drawn mean calling rates and coefficients of variation from (A) and (B). (D) An overhead diagram of the testing arena (2 m diameter) with a female treefrog (not to scale) at the release point in the centre and the response zones in front of each speaker indicated by a dashed line. The two hypothetical phenotypes in (C) were used to generate audio tracks represented by the waveforms arrayed around the edges of the arena (minimum angular separation 45°). All waveforms show two minutes of signalling. Each of the colours corresponds to a single hypothetical male phenotype in stimulus set 8. Panels (E–H) show the same process illustrated in (A–D) for a four-choice test, while (I–L) illustrate an eight-choice test. Each two-choice test was performed once; each four-choice test was performed twice; each eight-choice test was performed four times.

Figure 2.

Waveforms (amplitude over time) depicting the synthetic treefrog signals used in phonotaxis tests. (A) A single synthetic treefrog call consisting of a series of 29 sound pulses. (B–I) Each waveform shows 2 min of a single-simulated male phenotype from stimulus set 8. The single call depicted in (A) was concatenated with the randomly drawn intercall intervals to generate audio tracks used in phonotaxis tests (e.g. B–I).

Calls (figure 2) were synthesized de novo using custom scripts in MATLAB v. 2022b (MathWorks, Natick, MA) at a sampling rate of 44.1 kHz and a bit-depth of 16. Call traits not under manipulation were held constant such that the calls themselves were identical, and differing rates were achieved through changes in the intercall intervals. To approximate the temperature-corrected (20°C) mean trait values of the frogs in our population [20,42], calls consisted of 29 pulses delivered at 48.8 pulses per second with a constant pulse duty cycle of 50%. Each pulse was composed of two harmonically related and phase-locked sinusoids. The dominant frequency (second harmonic) was 2500 Hz, and the fundamental frequency (first harmonic) of 1250 Hz had an amplitude of −11 dB relative to the dominant frequency. Each pulse was 10.2 ms in duration. Pulse amplitude envelopes were shaped to have a 3.7 ms rise time and a 6.5 ms fall time. Pulses reached 50% of their maximum amplitude 1.5 ms after pulse onset and fell to 50% of their maximum amplitude 3.17 ms after the amplitude began to fall.

We performed playback experiments under infrared light in a custom-built, temperature-controlled hemi-anechoic chamber (2.8 m by 2.3 m by 2.1 m, length by width by height; IAC Acoustics, North Aurora, IL). The floor of the chamber was covered in grey, low-pile carpet. On the floor, we built an acoustically transparent but visually opaque testing arena from hardware cloth covered in black fabric. The arena measured 2 m in diameter. Eight speakers (Orb Audio Mod1 speakers, New York, NY) were arrayed on the floor on the outside of the arena wall pointing towards the centre of the arena to generate a 45° angular separation between adjacent speakers (figure 1D). In the centre of the testing arena, we placed a circular release cage made of acoustically transparent plastic mesh (9 cm by 2 cm, diameter by height), the lid of which could be lifted to let the subject out by means of a rope and pulley system operated by an experimenter on the outside of the anechoic chamber.

Subjects were tested 12 times each in a within-subjects design (n_{within subjects} = 12; n_subjects = 125; n_{total tests} = 1500). The tests were performed in randomized order between subjects. Females can easily complete 15−20 tests in the laboratory and do not show carry-over effects between tests [43,44]. At the start of each trial, the subject was separated from her mate and enclosed in the release cage. Subjects were placed into the release cage with no particular orientation to speaker locations, and they were able to freely move about and reorient within the release cage at any time. The lights of the sound chamber were extinguished, and its door was closed. A 1 min acclimation period was allowed to elapse before the playback of stimuli began. To prevent the initial calls in the audio tracks from all playing simultaneously, each stimulus began with its first randomly drawn intercall interval to generate random variation in the relative timing of the first calls. After playback began, the relative timing of different stimuli that were playing in a given trial was allowed to vary according to their specified values of calling rate and regularity. Hence, calls produced by different simulated males were allowed to overlap, as they often do in natural breeding choruses [45]. Subjects were kept in the release cage for a further 30 s following the beginning of playback after which they were remotely released and allowed to freely move within the test arena.

(c). Linking receiver performance to signaller fitness

Female performance in phonotaxis tests was used to estimate the strength of selection on the signalling phenotypes of simulated males. We observed phonotaxis trials in real time from outside the anechoic chamber using an overhead infrared-sensitive camera. Trials ended when one of the following conditions was met: (i) a subject indicated a choice by entering a choice bin described as a 10 cm radius semi-circle in front of the speaker [40] or (ii) after 10 min (600 s) elapsed, in which case a female was given a score of ‘no response’. Under natural conditions, females listen for only about 2 min before making a decision [46]. Across our study, the mean latency to respond was 109.1 s (median 85.0 s; range 16−594 s), and 90% of trials ended with a female making a choice in 205 s or less. If a female scored ‘no response’ for two trials in a row, we added a ‘reference’ test into the sequence to ensure that she was still responsive to acoustic stimuli before proceeding with the rest of the tests. In the reference test, a subject was presented with a single ‘standard’ conspecific call (average in all call properties) in a procedure that was otherwise identical to our experimental procedure [47–49]. Any subject that responded positively to playback in the reference test was deemed responsive, her previous ‘no response’ was considered valid and she continued to complete the rest of the trial series. Only one subject failed to respond in the reference test; she was removed from the experiment and the data from the trials she completed prior to the reference test were excluded from this analysis. Most (100 of 125) subjects responded to a stimulus in all 12 tests: 12 subjects responded in 11 tests; 8 responded in 10 tests; 2 responded in 9 tests; and 3 responded in 8 tests. Given the nature of our hypotheses, we considered not responding to a stimulus in a given test to be a valid outcome. For all trials, we recorded the speaker chosen and the values of calling rate and regularity for the male it simulated.

A simulated male’s signals were considered more attractive if females chose it more often. To estimate the strength of selection, we generated relative fitness scores for simulated males that could be directly compared across manipulated choice contexts. ‘Two-alternative choice tests’ (hereafter ‘two-choice tests’), in which animals are offered two stimuli between which to choose, are standard in phonotaxis studies of frog mate choice. We chose a design that could scale relative fitness scores in four- and eight-choice tests to a maximum score of 0.5, which is the fitness of the chosen male in a standard two-choice test if a subject were to choose at random (probability of being chosen = ½). Each subject was tested with her eight unique stimuli representing different simulated males in a series of 12 two-, four- and eight-choice tests, as follows. We randomly assigned each of the 8 stimuli in the set to speaker positions in the arena (figure 1) and then divided them into 4 subsets of 2 (speakers 1−2, 3−4, 5−6 and 7−8) and 2 subsets of 4 (speakers 1−4 and 5−8). We gave each subject four two-choice tests in which she chose directly between each of the four pairs of two adjacent speakers, such that each of the eight simulated males was present in exactly one two-choice test. We gave each subject two four-choice tests in which she chose directly among each of the sets of four adjacent speakers, such that each simulated male was present in exactly two replicate trials of a four-choice test. Finally, we gave each subject four eight-choice tests in which she chose directly among eight speakers, such that each simulated male was present in four replicate eight-choice tests. We scored each hypothetical male phenotype’s relative fitness as the number of times he was chosen in a given choice context divided by the total number of choices possible in that context. For instance, if a given simulated male was chosen in three of the four possible eight-choice trials, his relative fitness score for that particular choice environment would be 3/8 or 0.375. Across the two-, four- and eight-stimuli tests, therefore, relative fitness always varied between a minimum of 0.0 (indicating a simulated male that was never chosen) and a maximum of 0.5 (indicating a simulated male that was chosen in every choice test in which he appeared).

We randomly assigned simulated males to speakers between stimulus sets such that there was no systematic relationship between the relative spatial locations of stimuli and their signal properties across the experiment. The minimum angular separation between any two adjacent speakers in all trials was kept constant at 45° (figure 1D,H,L) to control for the possibility that subjects might have more difficulty discriminating between sounds that were closely spaced than between sounds separated in space [32]. Note that grey treefrogs regularly distinguish sounds at angular separations of less than 45° [50–52].

(d). Statistical analysis

Statistical analyses were carried out in R v. 4.3.3 ‘Angel Food Cake’ [53]. Our approach used generalized linear mixed models (GLMM) to estimate linear, quadratic and correlational selection gradients [54] on the calling rate (cpm) and calling regularity (CV_w) of simulated males. Prior to analysis, we converted the cpm and CV_w values of each simulated male phenotype to z-scores so that we could directly compare the magnitude of the selection gradients on calling rate and regularity in like terms. The linear regression coefficients from GLMM were interpreted as linear selection gradients (β) for the main effects of calling rate and regularity and allowed us to make inferences about directional selection. We estimated nonlinear selection gradients ($\gamma$) using the quadratic and correlational terms in the model. Quadratic selection gradients (calling rate² and regularity², or $\gamma$_CR and $\gamma$_CVw) measure stabilizing and disruptive selection on individual traits, while correlational selection gradients (here, calling rate × regularity, or $\gamma$_{CR × CVw}) measure selection on combinations of more than one trait. We fit three GLMM using the nlme package in R [53], one model for each of the two-, four- and eight-stimulus environments. We included female identity nested within stimulus set as random effects in each model. This was done to account for (i) possible among-individual differences in female behaviour and (ii) the possibility that a signalling phenotype’s relative fitness was influenced by the random assemblage of the other stimuli in its set. Raw output from the three GLMM models is available in the electronic supplementary material.

3. Results

In two-choice tests, subjects exerted strong directional selection for faster calling rates (β_CR = 0.071, d.f. = 870, t = 9.421, p < 0.0001) and greater call regularity (β_CVw = −0.054, d.f. = 870, t = −7.305, p < 0.0001; table 1; figures 3A and 4). Note that because smaller values of CV_w indicate greater regularity, negative estimates of β indicate directional selection for greater regularity. Significant stabilizing selection was also acting on regularity (γ_CVw = −0.016, d.f. = 870, t = −2.485, p = 0.013; figure 3B) but not on calling rate (γ_CR = −0.010, d.f, = 870, t = −2.485, p = 0.078; figure 3A). We did not detect significant nonlinear correlational selection on the interaction between calling rate and regularity (γ_{CR × CVw} = −0.013, d.f. = 870, t = −1.607, p = 0.108; figure 4). Together, these results suggest that, in two-choice tests, selection for higher calling rates is purely linear, while selection on regularity is curvilinear or threshold-shaped, with diminishing returns of increasing regularity as CV_w approaches zero.

Table 1.

Estimated selection gradients from the three generalized linear mixed models estimating the effect on relative fitness of calling rate, calling regularity (coefficient of variation within males, CV_w) and their quadratic (represented by superscript 2) and correlational effects (represented by ×). Values in bold are statistically significant at the α = 0.05 level.

		choice set size
selection gradient type	effect	two	four	eight
linear	rate	0.071	0.025	0.018
	regularity	−0.054	−0.015	−0.013
quadratic	rate2	0.01	0.001	0.001
	regularity2	−0.016	0.000	−0.001
correlational	rate × regularity	−0.013	−0.015	−0.003

Figure 3.

Effects of (A) calling rate (calls per minute) and (B) regularity (coefficient of variation within males, CV_w) on relative fitness. Lines show second-order polynomial fits of the marginal effects for each number of stimuli (blue, two; orange, four; magenta, eight) and the shading shows standard error.

Figure 4.

Fitness surfaces showing how the relative fitness of simulated male phenotypes as a function of their calling rate (calls per minute) and regularity (CV_w) changes across two-, four- and eight-choice environments. Sexual selection was measured separately in two-choice tests (A, top surface; B), four-choice tests (A, middle surface; C) and eight-choice tests (A, bottom surface; D). Data in this figure have been loess-smoothed and plotted using MATLAB v. 2023b.

In four-choice tests, calling rate and regularity became less predictive of the relative fitness of signalling phenotypes. Subjects continued to exert directional selection for faster calling rates (β_CR = 0.025, d.f. = 870, t = 4.646, p < 0.0001) and greater regularity (β_CVw = −0.015, d.f. = 870, t = −2.843, p = 0.005; table 1; figure 1), but this selection was markedly weaker than in two-choice tests. Directional selection on calling rate was only about $\frac{1}{3}$ the strength of that in two-choice tests (table 2), while directional selection on calling regularity was only about ¼ the strength. Neither calling rate (γ_CR = 0.001, d.f. = 870, t = 0.161, p = 0.872) nor regularity (γ_CVw = −0.000, d.f. = 870, t = −0.091, p = 0.928) was under significant quadratic selection in the four-choice environment. A significant correlational effect of the two traits (γ_{CR × CVw} = −0.015, d.f. = 870, t = −2.683, p = 0.007) indicates that the positive effect of calling rate on relative fitness decreased as regularity decreased (i.e. fitness selection for faster calling rates diminished as calling became more irregular). This effect appeared only in the four-choice context.

Table 2.

Magnitude of the linear selection on calling rate and calling regularity (measured as coefficient of variation within males, CV_w) expressed as proportions of the magnitude of selection on the same call property in two-choice tests.

	choice set size
property	two	four	eight
rate	1	0.352	0.254
regularity	1	0.278	0.241

In eight-choice tests, subjects continued to exert directional selection on calling rate (β = 0.018, d.f. = 870, t = 6.551, p < 0.0001; table 1; figures 1A and 3) and call regularity (β = −0.013, d.f. = 870, t = −4.849, p < 0.0001; figures 1B and 3). However, this selection was only about ¼ the strength of the linear selection on both traits in the two-choice context (table 2). No detectable nonlinear selection was acting in the eight-choice environment (calling rate quadratic: β = 0.001, d.f. = 870, t = 0.316, p = 0.752; calling regularity quadratic: β = −0.001, d.f. = 870, t = −0.464, p = 0.643; calling rate × calling regularity interaction β = −0.003, d.f. = 870, t = −1.054, p = 0.292).

We visualized the three-dimensional fitness surfaces in each of the three choice contexts (figure 4). The resulting fitness surfaces had different overall heights, reflecting a decrease in receiver performance as choice set size increased and suggesting that average male fitness is lower in larger assemblages. Male relative fitness was maximized by high calling rates with high regularity, especially when only two simulated males were present.

4. Discussion

Our data unequivocally support the predictions of the choice overload hypothesis. Grey treefrog females exerted multivariate phenotypic selection simultaneously on two temporally emergent properties of male signals, calling rate and calling regularity. However, the strength and shape of the selection they imposed were context-dependent. Male relative fitness was maximized when signals were produced at faster rates, with greater regularity across choice environments; that is, the same male phenotypes were attractive across environments [55]. However, male relative fitness was maximized when there were only two signallers present. Selection weakened and became more nonlinear when there were more options available, and some quadratic and correlational effects were detectable only in the larger choice sets of four and eight, which represented more complex choice environments. Overall, having more available options decreased receiver performance and changed the resulting sexual selection landscape. The broad implication of these results is that that choice overload in complex choice environments—such as leks—can reshape the selective landscape in ways that allow relatively unattractive males to mate. These findings are an important counterexample to the idea that direct comparisons among potential mates in leks may increase the strength of sexual selection [37].

Our study extends previous work on mate choice in two important respects. First, our results address an important and persistent limitation in experimental studies of mate choice. As Dougherty [25] noted in a recent review on the design of mate choice studies, no-choice and two-choice tests are commonly used experimental designs that poorly reflect the ecological realism of complex choice scenarios like those found in leks. Such designs can lead to incorrect estimates of selection acting on signals in natural populations [25]. Second, instead of focusing on a single feature of a single signal, we examined multivariate selection on two temporally emergent properties of signalling behaviour, signalling rate and regularity. One of these properties—signalling rate—often has profound influences on mating decisions, at least in simple choice contexts, and is under directional selection by female choice across numerous and diverse animal taxa (reviewed in [11]). The second property—signalling regularity—can impact a female’s ability to accurately assess the first [14] and may itself be under directional selection [15]. The multivariate approach used in the present study reveals how both the strength and the shape of sexual selection on these two temporally emergent properties can vary as choice complexity increases.

Animals (including humans) may frequently reach their cognitive processing limits as the complexity of choice environments increases [27,28,30]. Choice overload occurs when animals are confronted with an overwhelming number of options simultaneously [29,31]. Choice overload is frequently quantified by the changes in decision making it induces, one of which is a change in option selection [27] or the extent to which individuals choose options they would not choose under simpler circumstances. Sharma & Isvaran [34] recently showed how choice overload influences the decisions female mosquitoes make in selecting oviposition sites. Female preferences for optimal oviposition sites were overall weaker when choice set size increased. In Pacific field crickets, females become less likely to express their strong preference for songs with more ‘long chirp’ elements (seven pulses versus two) as the number options in the choice set increases from two to eight [32]. That study was the first, to our knowledge, to explicitly address choice overload in the context of acoustically mediated mate choice. We extend that work in several key ways, by measuring selection on (i) a large sample of male phenotypes (512 versus 2 in Tanner & Simmons [32]) that (ii) varied continuously in two dimensions (versus discretely in one dimension) and (iii) included temporally emergent traits that required receivers to assess a stream of repeated signals to estimate a trait value.

While they have not explicitly addressed the concept of choice overload, a handful of previous studies of mate choice in frogs have also shown that the selectivity of females for a preferred acoustic feature of individual calls, such as lower frequencies [56–58], longer durations [59] and greater amplitudes [56,60], generally weaken as the number of options increases beyond two, although preferences can be stronger in two-choice tests compared with no-choice tests [20]. Receivers in the present study showed changes in option selection as the number of stimuli increased from two to four and eight, consistent with the interpretation that receivers in four- and eight-choice tests experienced choice overload compared with the two-choice tests. If frogs routinely experience choice overload under natural conditions, these changes in option selection at the level of individual decisions should scale up to re-shape sexual selection at the population level [31].

As the number of signallers in a choice set increased from two to four and eight, the strength of selection declined dramatically. Consistent with previous studies, females exerted multivariate selection favouring faster calling rates and greater calling regularity [14,15,20]. However, the strength of linear selection on both traits measured in the larger choice sets was only a fraction of what it was in two-choice tests. In contrast, a previous study of grey treefrogs showed that two-choice tests revealed stronger selection on some traits than no-choice tests of the same subjects, using the same stimuli [20]. Thus, in line with the more is better hypothesis, some choice may indeed be better than no choice at all, and adding an option (i.e. changing from one to two stimuli) can increase estimates of the strength of selection. Interestingly, Stratman et al. [61] showed in the sister taxon Hyla versicolor that females avoid lone calling males, instead preferentially approaching groups of three calling males and then assessing their call properties. Those results also are consistent with the more is better hypothesis, at least for small numbers of signals (choice set sizes of one and three). In one study of bumblebees [36], increasing the choice set size from two to four improved forager selectivity for a preferred flower type. Thus, in at least some contexts and species, increasing the number of options in the choice set (or males in the lek) may increase receiver selectivity [36] or attract more receivers [61,62]. In contrast, by increasing to four and eight options, the present study showed a decrease in selection in line with the choice overload hypothesis. Taken together, the two studies of treefrogs (this article and [20]) suggest that there may be an inverted U-shaped relationship between the number of calling males in an assemblage and the strength of sexual selection imposed on signal traits by receivers. Some models of evolution under mate choice suggest that sexual selection should be stronger when more mates are sampled (e.g. [63]). If receivers are better able to find mates with preferred combinations of traits when they sample more potential mates, that could generate selection for females that choose larger groups of displaying males over smaller groups. Our results suggest that, at least in some species, receivers may be less likely to gain the benefits of mate choice when they must choose in larger assemblages.

Our experiment revealed that selection also changes shape across choice contexts, with nonlinear selection appearing only in some contexts. Only in two-choice tests, we detected some negative quadratic selection on calling regularity. Combined with the linear selection for more regular calls, that result suggests a threshold-shaped preference for more regular calls, or diminishing returns for increasingly regular calls as the CV_w approaches zero. Additionally, only in four-choice tests, there was significant correlational selection on the product of rate and regularity. That effect suggests better receiver performance at estimating calling rate when signals are regular, in line with previous studies [14]. For males, this receiver performance likely translates to a stronger relationship between fast calling rates and relative fitness when calling is also more regular. Why this correlational selection was detected only in four-choice tests remains uncertain. As a consequence of statistical sampling, attractive correlational combinations of rate and regularity are more likely to occur as the number of calling males increases. We speculate that correlational selection was stronger in four-choice test compared with two-choice tests for this reason, but that choice overload in the eight-choice tests compromised females’ ability to discriminate based on combinations of calling rate and regularity. That the strength of correlational selection might vary in a nonlinear way as a function of choice set size argues for measuring sexual selection in complex environments more often.

At present, the neural underpinnings of choice overload remain poorly understood, even in humans (e.g. [64]). Thus, we can only speculate about possible underlying perceptual or cognitive mechanisms that might explain the weakening of female preferences we observed as choice set increased. Using playbacks through a multi-speaker array deployed in a natural breeding pond, Schwartz et al. [46] estimated that female grey treefrogs likely assess the calls of potential mates for about 2 min before making their mating decision. In order to choose a male calling with a fast and regular rate from among multiple potential mates, females would have to recognize competing and potentially overlapping vocalizations as conspecific advertisement calls, assign each call to its correct source male, track the calls of different males long enough to assess differences in their rate and regularity of calling and then localize at least one of the males. Hence, memory, attention and perceptual grouping (e.g. auditory streaming) are likely key elements of a female’s ‘psychological landscape’ [65,66] involved in comparing differences in calling rate and regularity among potential mates in a chorus. Past studies of frogs indicate that memory [67,68], attention [69,70] and perceptual grouping [71,72] play important roles in vocal communication, but these processes have not been investigated in the context of choice overload. It is certainly plausible that, as the mate choice set size increases, these cognitive and perceptual processes might become limiting and lead to choices of less preferred signallers.

In summary, we conclude that mate choice studies based on using small choice sets—including the no-choice and two-choice tests that make up the bulk of mate choice studies [25]—are likely to overestimate the strength of selection and may also overgeneralize or misestimate the shape of the fitness landscape in natural choice contexts. Multi-choice experiments like the present study may allow more realistic inferences about how sexual selection acts in nature. However, we note that controlled laboratory experiments like this one necessarily reduce the complexity of natural choice environments, in which receivers must discriminate among many more signallers that vary within and between individuals in multiple signalling traits against the noisy backdrop of a multispecies chorus. In highlighting the importance of temporally emergent features like rate and regularity, our results also serve to emphasize that future studies of selection on animal signals should refrain from divorcing signals from their associated signalling behaviours: sometimes the male’s signal is less important than how he uses it. Future research should also take into account that natural choice environments vary in their complexity across both space and time [23]. As our data show, more complex communication environments can reshape sexual selection, allowing different individuals to mate. Receiver perception and cognition are potentially limiting factors, but future studies of memory, attention and perceptual grouping are needed to fully understand how receivers navigate realistically complex social environments [66,73].

Ethics

This work was done under scientific collection permits granted by the Minnesota Department of Natural Resources. Our research protocol was approved by the University of Minnesotas Institutional Animal Care and Use Committee under protocol 2301-40692A. We followed all of the Animal Behaviour Society Guidelines for the Ethical Treatment of Non-Human Animals in Behavioural Research and Teaching [74].

Data accessibility

Our data are available via the Dryad Digital Repository [75].

Declaration of AI use

We have not used AI-assisted technologies in creating this article.

Authors’ contributions

J.C.T.: conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, project administration, software, supervision, visualization, writing—original draft, writing—review and editing; M.S.-I.: conceptualization, funding acquisition, resources, writing—review and editing; K.L.H.: conceptualization, funding acquisition, project administration, writing—review and editing; M.A.B.: conceptualization, funding acquisition, project administration, resources, supervision, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

This work was supported by the National Science Foundation under Grant No. 215403 and the United States-Israel Binational Science Foundation (BSF) Grant No. 2021743.