Ultrasonic Vocalizations of Male Mice Differ among Species and Females Show Assortative Preferences for Male Calls

Male house mice (Mus musculus) emit ultrasonic vocalizations (USVs) during courtship, which attract females, and we aimed to test whether females use these vocalizations for species or subspecies recognition of potential mates. We recorded courtship USVs of males from different Mus species, Mus musculus subspecies, and populations (F1 offspring of wild-caught Mus musculus musculus, Mus musculus domesticus (and F1 hybrid crosses), and Mus spicilegus), and we conducted playback experiments to measure female preferences for male USVs. Male vocalizations contained at least seven distinct syllable types, whose frequency of occurrence varied among species, subspecies, and populations. Detailed analyses of multiple common syllable types indicated that Mus musculus and Mus spicilegus could be discriminated based on spectral and temporal characteristics of their vocalizations, and populations of Mus musculus were also distinctive regardless of the classification model used. Females were able to discriminate USVs from different species, and showed assortative preferences for conspecific males. We found no evidence that females discriminate USVs of males from a different subspecies or separate populations of the same species, even though our spectral analyses identified acoustic features that differ between species, subspecies, and populations of the same species. Our results provide the first comparison of USVs between Mus species or between Mus musculus subspecies, and the first evidence that male USVs potentially facilitate species recognition.


Introduction
Ultrasonic vocalizations (USVs) in rodents have been recognized for more than 100 years (in theory: [1], first recording [2]), and recent analyses of the spectrographic features of the USVs of male laboratory house mice (Mus musculus) have revealed surprising complexity [3]. Male house mice emit USVs during courtship, and their vocalizations may facilitate mating [4].
To test whether USVs in house mice mediate species, subspecies, and population recognition, we recorded the courtship USVs of males from eight different wild-derived house mouse populations, all members of the M. musculus species group (F1 offspring from wild-caught M. m. musculus from 4 locations, M. m. domesticus, as well as two colony-reared hybrid lines of M. m. musculus x M. m. domesticus crosses, and M. spicilegus). We analyzed spectral and temporal parameters of male USVs to determine whether Mus species, M. musculus subspecies and M. m. musculus populations can be discriminated using classification models. In addition, we conducted playback experiments to assess whether females are able to discriminate between courtship calls produced by males from different species, subspecies or other populations from the same species. If male USVs facilitate species recognition and females show assortative preferences for male USVs, then these vocalizations potentially facilitate hybridization avoidance and speciation [43].

Subjects and housing
Animals in our study were colony-bred offspring of wild-caught house mice and mound-building mice. House mice were originally trapped at four sites surrounding Vienna, Austria, including Safaripark (48°18 ' House mice were reared in mixed-sex family groups until weaning at 21 days of age. After weaning, males were individually housed to prevent fighting, and females were kept as sister pairs in type II cages (size: 26.5 x 20.5 x 18 cm, plus high stainless steel covers, mesh width 1 cm) with bedding and nesting material (ABEDD, Vienna, Austria: aspen wood chips and shavings) and shelter (paper rolls). Mound-building mice were kept in same-sex as well as in mixed-sex litter groups in interconnected type II cages (size: 26.5 x 20.5 x 18 cm, plus high stainless steel covers, mesh width 1 cm) with bedding and nesting material (ABEDD, Vienna, Austria). Home cages were kept at standard conditions (mean temperature 20 ± 1°C and 12:12 h light: dark cycle; lights on at 07:00 a.m.). Food (Altromin, Lage, Germany) and water were provided ad libitum. All experimental mice were sexually mature (> 8 weeks). Animals were not acoustically isolated prior to the experiment (except for M.s., which were bred in a different colony room) and were kept within the same colony room. males of M.s. continued to live in mixed sex family groups through adulthood, no further social experience treatment was performed.

Urinary stimulus collection
To trigger male USV emission for recording, fresh female urine was used as a stimulus [46]. Female urine was collected from donor females placed on a surface covered with clean aluminum foil. Handling was usually sufficient to induce urination, but urination could also be triggered by ventral stroking in an anterior-to-posterior direction [47]. Urine was collected immediately and fresh urine was subsequently pipetted onto a clean cotton swab (storage 5 min, volume 60 μl). In addition, soiled bedding (12 g) of the same female was collected from her home cage and presented simultaneously by placing it in front of the test animal. Urine and bedding were derived from different unfamiliar, randomly selected adult females from our captive mouse colony regardless of their reproductive stage, as there is mixed evidence that female estrus affects male USV emission [47,48].
To enhance female responsiveness to male USVs during acoustic playback experiments, we collected urine from several males, which we pooled and presented to females, as an additional stimulus. Equal amounts of fresh urine (5 μl per male) were mixed in a 0.2 ml PCR tube and stored at -20°C. Here, due to the number of males (10), it was not possible to simultaneously collect fresh aliquots of all males for each pool. A urinary pool consisted of aliquots of 10 unfamiliar male mice of each population or species (50 μl each). For tests including M.s., the urine pool for the comparing stimuli (M.s. and M.m.m.1) was reduced to 12.5 μl (1.25 μl per male), as the amount of urine obtained from M.s. was generally low. Experimenters wore clean gloves at all times and metal foil and gloves were exchanged immediately after each urine collection.

Apparatus, calibration and recording procedure
Recordings were made in an acoustically isolated recording room with no other animals present. A sound-insulated chamber (65 cm x 65 cm, inside lining: acoustic pyramid foam (5 cm)) was used for recording USVs from individual males within their home cages. The insulated lid provided an opening for the condenser microphone (UltraSoundGate CM16/CMPA, 15-180 kHz, flat frequency response (± 6 dB) between 25 and 140 kHz; Avisoft Bioacoustics, Berlin, Germany) which was fixed 20 cm above the cage in the middle of the box. For monitoring USVs, we used an UltraSoundGate 116 (Avisoft Bioacoustics) and an external soundcard (Edirol UA-101, 24-Bit/192 kHz 10-in/10-out Hi-SPEED USB (USB 2.0) audio interface for multitrack computer recording). Settings included sampling rate at 250 kHz and a format of 16 bit. Recordings were transferred to a sound analysis system (Avisoft-SASLab Pro, Version 4.40, Avisoft Bioacoustics) for processing. Spectrograms were generated with a fast Fourier transformation (FFT)-length of 512 points and a time window overlap of 50% (100% Frame, FlatTop window). A noise reduction by 90 dB below-52 dB was used to reduce background noise whilst not compromising automated call detection. We calibrated recording equipment by recording a pure tone of 440 Hz (commercially available tuning fork) and comparing the frequency in the spectrogram with the actual frequency recorded.
Before each recording session, a clean cage lid was put on each male's home cage (type II; without food pellets and water bottle to reduce sound interference). The cage with the mouse was placed in the center of the recording chamber. Following a 5 min habituation period, test trials began after the introduction of female urine and soiled-bedding stimuli. Recording trials lasted 30 min each after stimuli introduction and after each trial, the recording chamber was cleaned using a handheld vacuum cleaner. The temperature of the testing room was held constant at 20.4°± 0.9°C. During recording, no one was present in the recording room.

Experiment 1-USV variability among populations and species of male mice
We compared USV characteristics of males from different mouse populations, subspecies and species. In addition to the presence or absence of vocalizers within populations, distinct USV parameters and syllables (discrete sound units separated by silence from each other [3,25]; Fig  1), spectrographic features, and vocal repertoire size were compared. Altogether 127 socially experienced, mature males (255.4 ± 97.3 SD d of age) were recorded until a sample size of 10 males per population was reached. Due to high proportions of non-vocalizing males and limited number of experimental animals, samples sizes were lower in the hybrid lines (Hybrid 1: N = 5; Hybrid 2: N = 7) and in the M.m.d. population (N = 4). Only vocalizers were used for further spectrographic analyses (N = 66; 269.3 ± 103.8 SD d of age).
Despite prior noise reduction (by 90 dB below-52 dB), recordings still contained a considerable amount of 'non-USV' sound, and this background noise was manually removed from spectrograms. Parameters were determined automatically (Avisoft-SASLab Pro; version 4.40), including minimum and maximum frequency, peak frequency, peak amplitude, entropy and bandwidth, which were derived from start, center, and end spectrum of the entire syllable. Peak frequency is defined as the frequency at the location of maximum amplitude. Peak amplitude is defined as the point with the highest intensity within the spectrum. Entropy is a measure of the bandwidth and uniformity of the spectrum of a sound; tonal (whistle-like sounds) usually have low entropy (< 0.3), while broad-band (noisy) sounds have higher entropies (> 0.4). Bandwidth is calculated as the frequency difference between maximum and minimum frequency. Mean call duration was measured as a temporal parameter of syllables.
To examine between-species and between-population differences in extracted spectral and temporal characteristics of USVs, we calculated individual mean values for all USV types of a given species or population. To reduce the dimensionality of the 25-variable data set, we extracted 5 principal component axes (2 frequency, 1 amplitude, 1 bandwidth, and 1 entropy) from spectral and temporal characteristics. Prior to the PCA analysis we identified multivariate outliers in the dataset, using squared standardized distance. This measure allowed us to focus on potentially influential points to consider for removal. We removed the 12 most extreme outliers. To standardize across different measurement scales, a correlation matrix was used in the PCA.
To assess differences in USV characteristics among groups, we made the following comparisons. Lastly, we compared all populations, at all taxonomic levels, with each other, for a total of 8 groups. We performed all comparisons considering all USV types combined and considering each call type independently.
Data analyses and statistical tests. We confirmed the assumptions of models and tests before applying, and when data were non-normal and could not be transformed, nonparametric tests were applied. To test whether populations differed with respect to their ratio of vocalizing/non-vocalizing males, χ 2 tests with false discovery rate (FDR) controls were conducted. A binominal test was applied, setting the expected proportion to 0.5 (50%) to analyze the frequency of vocalizers/non-vocalizers within populations. Latency until first ultrasonic call was measured as a temporal feature of vocalizations. A nonparametric Kruskal-Wallis-H test was applied followed by Mann-Whitney U tests to compare latencies of the different populations. The total number of uttered syllables per individual per 30 minutes recording period was counted by visual inspection of spectrograms. An effect of age on latency and / or USV emission rate was tested via Spearman rank correlation for nonparametric data. Repertoires were compared using nonparametric Kruskal-Wallis-H, followed by Mann-Whitney U tests. In order to test the classification of populations based on USV features, we compared two classification methods: support vector machine (SVM) and discriminant function analysis (DFA). We used the 5 principal component scores and the mean duration variable as a reduced dataset in separate analyses. For the species comparison, we also used the raw, 25 variable data set. For the SVM, each population dataset was randomly divided into training (80%) and testing (20%) datasets. SVM models were separately tuned with the respective training dataset to obtain the best gamma and cost parameters for each model using 10-fold cross validation. Data from all USV types were included in the analysis. DFA and multivariate analysis of variance (MAN-OVA) was used to test for differences in the USV characteristics among populations with populations used as a classification factor. For the DFA, Wilk's lambda was used to estimate discrimination among individuals and an F-test was used to determine its significance [52]. DFA cross-validation classification rates are presented for comparison with SVM classification rate. For all groups where SVM and DFA classification occurred, we also compared the variables using either Kruskal-Wallis-H or applied Mann-Whitney U tests. All statistical analyses were performed using SPSS (version 15.01 for Windows) and R (3.0.2) [53]. Results are reported as mean ± SE, unless otherwise specified. In all cases p < 0.05 was considered to be statistically significant, and tests were two-tailed. FDR was used to control for Type I errors due to multiple testing [54].

Experiment 2 -Female discrimination of male USV playbacks
We tested females for their ability to discriminate USVs of different mouse populations, subspecies and species using USV playbacks. The females in our choice experiments originated from a single M. musculus population (M.m.m.1), and mice from this population were used in previous USVs experiments [5]. Females were tested in or near estrus, which was determined by vaginal smears [55,56] collected 5-7 h before the trial to prevent handling stress which might affect the experiment. To synchronize and increase the number of estrous females, 12 g of male soiled bedding (unrelated to the females and the subsequent testing stimuli) was introduced as a priming stimulus to female home cages 3 days prior to their use in the experiment [57]. The choice apparatus was constructed from a type III cage (42.5 x 27 x 20 cm), specially modified for assessing females' attraction to playbacks through one of two speakers at one end ('Speaker zone') (see S1 Fig). Individual females were placed at the opposite end ('Neutral zone') from which they could move into one of two equal-sized compartments, which both contained a fenced speaker for playing ultrasound recordings (Ultrasonic Dynamic Speaker Magnat, dominant frequency range 1-55 kHz; impedance 4 ohm, Avisoft Bioacoustics, Berlin, Germany; using an external soundcard (Edirol UA-101) covering ultrasonic frequency ranges) and were separated by acoustic foam (Pur Skin, 10 mm, SONATECH). Acoustic insulation (pyramid foam, 5 cm) ensured that playbacks played on one side were only heard in the corresponding arm and 'Neutral zone', allowing females to hear playbacks from both compartments at the onset of the experiment. Signal quality was verified by spectral comparison of recordings from original and played-back USVs. For analyzing females' relative attraction, compartments were further divided into different proximity zones: 'Fence' (mice in contact with the fence in front of the speakers); 'Speaker zone' (area 0-9 cm from the speaker/fence); 'Middle zone' (area 9-18 cm apart); and 'Neutral zone'. As previously mentioned, females show limited interest when playbacks were the only stimulus and therefore, male urine was also provided as an 'enhancing stimulus' [58]. This stimulus consisted of two different pools of urine of 10 males (unfamiliar and unrelated to females), which were pipetted on filter paper (Whatman 3MM Chr, 0.34 mm thick, 4 x 4 cm) in front of the speakers. The urine stimuli in each compartment were identical in volume and composition, i.e., urine from males belonging to the species (or population) of both test populations was placed on both sides to prevent biasing the results. After females were placed in the 'Neutral zone', habituation lasted as long as the animal explored both compartments and returned to the 'Neutral zone' (all females did) after which playback was initiated.
Playbacks consisted of 310 sec of previously recorded uttered syllable bouts, and were standardized by the number of syllables (399 ± 3 SD, Avisoft SASLab Pro; version 4.40) to avoid preferences on the basis of performance-related traits [59]. Artificial inter-syllable durations were the same as those found in mice (< 1 sec, range 0.03-0.94), and duration between each individual phrase (sequence of syllables uttered in close succession) was 1 sec. Playback amplitude was standardized for both sides.
Experiments were digitally videotaped (Sony Handycam DCR-SR30) and analyzed blindly regarding female identity and trial number (using 'The Observer 7.0', Noldus). Retention times in the designated areas of the apparatus (Neutral zone, Middle zone, Speaker zone and Fence) were measured and compared. Self-grooming and sniffing at the enhancing urinary stimulus and their relative proportions were measured as behavioral parameters, providing further indicators for mating preferences [60].
As a positive control, we confirmed that females are generally attracted to male USVs by testing a USV playback versus silence (no-USV playback Population playbacks (pooled USVs) were composed of a collection of alternating calls of different males to reflect the complete population repertoire. Pooled playbacks consisted of 10 different 30 sec segments of syllable bouts (Avisoft SASLab Pro; version 4.40). When sample size < 10 males, segment duration/individual was increased to reach 310 sec of total playback length (e.g., in M.m.d.: four 76.5 seconds segments of uttered syllable bouts due to a low number of vocalizers (N = 4); Hybrid 1 & 2 were merged as test population to reach N = 10). Pooled USV playbacks from 4-10 males were used as a composite to control for individual variation in male USVs [8]. We conducted an experiment to investigate female responses to the pooled versus non-pooled playbacks to determine whether there is anything unusual about how females react to pooling individuals males.  (Table 1). Here, we define a trial as a single test of a female's response to two different populations and/or species. Females were tested once per trial in 5 different trials. Each trial took between 2-6 days to complete depending upon females' estrus status. Time between trials varied between 2-3 weeks. The order of trials was the same for all females (M.m.d., M.m.m.2, M.m.m.3, M.s., Hybrids). For each trial a new pooled USV playback of their own population was composed. The side of playback presentation was balanced within subjects and reversed in subsequent trials. Data analyses and statistical tests. All behavioral trials were checked for a general side bias using nonparametric Wilcoxon signed-rank tests. The initial preference (side of first entry), latency of first entry, and the number of visits in each arm were also recorded. Initial preferences were tested for significance using a binomial test with an expected proportion of 0.5 (50%). For comparisons of latency of first entry the time for the not chosen arm was set to 310 s (total time of a trial). When the means of retention times in proximity zones were nonnormally distributed, Wilcoxon signed-rank tests were applied. The time females spent sniffing at urine stimuli was determined with repeated measures ANOVAs with 'male USV' as the between subject factor and female as covariate. Sniffing behavior at the enhancing odor stimuli did not differ between cues (population-pools) and sides of the experimental apparatus indicating that these olfactory stimuli in general do not affect timed USV preferences (data not shown).
Female responses were further analyzed with respect to playback call composition. Although playbacks were standardized by the total number of syllables, 'Frequency-Step' syllables appeared irregularly in male vocalizations. The effect of individual variation in the numbers of 'Frequency-Step' syllables within playbacks on female behavior was tested using linear mixed models for repeated measures with the individual number of 'Frequency-Step' syllables as a fixed effect and trial number as a random effect.

Ethics
The study was approved by the institutional ethics committee of the University of Veterinary Medicine, Vienna, in accordance with Good Scientific Practice guidelines and national legislation. Mice were trapped on private land with the permission of the owners. Trapping of the founder individuals was conducted overnight with Sherman live traps. Each trap was set with food and nesting materials. Traps were checked twice during the night for occupancy and trapped individuals were immediately removed and placed individually into standard mouse cages (type II). Trapping was performed in accordance with national legislation and approved by the MA 22 (Municipality for Environment and Conservation of Vienna).  Populations differed in the proportion of different syllable types produced (Fig 3).  Table). Syllable type '1-Frequency-Step' did not differ significantly among populations (χ² = 10.971, p = 0.140).

Experiment 1-USV variability among populations and species of male mice
Spectral and temporal characteristics of USVs. Spectral and temporal characteristic data were reduced to 6 variables (5 PC components and duration). Two extracted frequency PC components explained 89% of the variation in the 12-frequency variables in the data set ( Table 2). The amplitude PC component explained 83% of the variation in the 4-amplitude variables in the data set ( Table 2). The bandwidth PC component explained 42% of variation in the 4-bandwidth variables, while the entropy PC component explained 66% of variation in the 4-entropy variables in the data set (Table 2). Female Preferences for Species-Specific Male USVs SVM exhibited higher discriminating efficiency at the interspecific (Table 3) and intersubspecific level (Table 4), but DFA classified vocalizations to intraspecific populations more consistently than SVM ( Table 5). As expected, both SVMs and DFAs performed better when classifying higher order taxonomic groups (interspecific and intersubspecific levels; Tables 3 and 4) than when classifying populations within a subspecies (Table 5). Across all 8 populations considered, both SVM and DFA had classification rates lower than 46% (43.12% and 45.8% respectively) because of considerable overlap among all M.m. populations and overlap between M.m. and M.s. populations (Fig 4).
We found striking variation in the USVs of different males (Fig 4), and although there was much overlap between different subspecies and populations, M.m. USVs differed from those of M.s. when considering all USVs syllable types, indicating species differences in spectral and temporal characteristics USVs (Table 3). Specifically, the parameters that differed most significantly between Mus species were duration and the PC axes describing frequency, amplitude, and entropy (Table 3) Table 3, see Fig 4). In addition, M.s. USVs were shorter in duration than M.m. USVs Female Preferences for Species-Specific Male USVs (0.030 ± 0.001 vs. 0.0325± 0.001 sec). These species differences, especially regarding frequency and amplitude were generally consistent across USV syllable types (Table 3). Both SVM and DFA classification approaches were effective at classifying over 85% of USVs to the species level (Table 3). Intraspecific comparisons between subspecies (M.m.m. and M.m.d.; Table 4) and within M. m.m. populations (Table 5) show differences in USVs based on spectral characters. However, SVM and DFA classification were much more effective at discriminating M.m. subspecies compared to populations within M.m.m. While both SVM and DFA classification were equally effective at discriminating two different subspecies (over 79%; Table 4) as for classifying two species (over 86%; Table 4), discriminating mice from different populations within M.m.m. was less than 42% for SVM and less than 53% for DFA (Table 5). Sample sizes were generally too small to examine intrasubspecific variation within particular USV syllable types (Table 4) (Table 5).   Table 6). Similarly, females did not discriminate between USV playbacks of males from different M.m.m. populations (Table 6), nor did they show differences in grooming behavior (data not shown). Also, we found no other differences in female responses to USVs from different populations in initial preference, latency to enter one stimulus compartment, or number of visits (data not shown).

Discussion
Our analysis revealed significant differences in several features of male USVs between M. m. musculus and M. spicilegus, including latency to vocalize, syllable repertoire, spectral and temporal characteristics. We also found that female M. m. musculus were able to distinguish between the USVs of conspecific and heterospecific males, and spent significantly more time near calls from conspecific males. These results provide the first evidence that the USVs of males differ among Mus species and that females can discriminate and prefer the USVs of conspecific over heterospecific males. As playbacks were composed of recordings of a pool of individual males, rather than a single individual male on each side, these results cannot be explained by a simple preference for more dissimilar or less familiar calls due to females' imprinting on paternal USVs [21]. Future studies are needed to test whether female preferences for the USVs of conspecific males are enhanced when combined with other stimuli from other sensory modalities (multimodal integration) [7,61] and whether USVs are subject to sexual selection mediated by female choice. If male USVs mediate such inter-sexual interactions, then USVs would help to avoid genetically incompatible matings and provide a pre-zygotic reproductive isolation mechanism. Females' preferences for the USVs of conspecific males may be controlled by innate recognition or females may positively imprint on species-specific features of vocalizations of males in their family-as found in birds [62]. The innate preference of female laboratory mice for the USVs of males from a different strain has been shown to disappear when they are reared without fathers, emphasizing the necessity of close exposure to USV [21] and disassortative preferences can be reversed by cross-fostering [21], suggesting that female preferences for male USVs are controlled by classical (paternal) imprinting, and that females negatively imprint on the USVs of males in their own family. If females positively imprint on species-specific features of USVs from males in their family, then such dual imprinting may allow females to avoid the extremes of inter-specific hybridization and close inbreeding (optimal outbreeding) [63]. We compared USVs of male M. musculus on an intraspecific level, and we found significant differences between two subspecies (musculus and domesticus) and also between hybrids and their parental subspecies. The classification of these groups of mice in our models (both SVM and DFA) was relatively high and comparable to classification levels at the species level. Yet, despite these differences in male USVs, we found no evidence that females discriminated between the playbacks of male USVs of M. musculus subspecies (M.m.m. versus M.m.d.) or between conspecific males versus hybrids of these subspecies. Thus, the lack of female M. m. musculus discrimination does not appear to be due to a lack of differences in male USVs between subspecies. Alternatively, pooled USV playbacks might hamper female discrimination, as they are more variable than natural calls and thereby might diffuse the subspecies signal. To rule out such potential experimental artifacts, female preferences should be tested with individual playbacks. Previous odor preference tests found that M. m. musculus, but not M. m. Female Preferences for Species-Specific Male USVs domesticus, mice from parapatric and sympatric populations showed significant assortative odor and partner preferences for their own subspecies [34,40] but no assortative preference was found in allopatric populations [41]. House mice from the subspecies hybrid zone produce distinct urinary signals that are more pronounced compared to other areas (i. e. character displacement) [42]. The mice used in our study were collected far from the hybrid zone, and therefore, studies are needed to test whether females from sympatric populations show assortative preferences for subspecies male USVs.
We also found several features of male USVs that differ among M. m. musculus populations, but found no evidence that M. m. musculus females discriminate between USV playbacks of males at the population level, as predicted by the paternal imprinting hypothesis. Females are capable of distinguishing the USVs of siblings versus unrelated males (from within the same population), indicating that females can recognize subtle differences in male USVs [5]. As the males in our experiment were unrelated to the choosing females, and USV playback consisted of pools of several males, each USV population pool might have been perceived as equally  Female Preferences for Species-Specific Male USVs distant at the intraspecific level. Nonetheless, our findings indicate that there are intraspecific differences in components of male USVs, especially within 'Constant modulated' and 'Frequency upsweep' syllable types, which likely exceed those found among laboratory mouse strains [26]. M. musculus mice in our study were reared in the same room, but mice from different populations were kept on separate racks. The hearing threshold of mice [64], and the high attenuation of ultrasound [65] makes it highly unlikely that mice were able to perceive USV from distances > 3m,. Nevertheless, it is possible that the lack of female discrimination for USVs of males from different M. m. musculus populations may be due to young females imprinting on the USVs of males from different cages and different racks. Female discrimination of individual males from differing populations on a small geographic scale is required to test the imprinting hypothesis. Also, further studies are needed to examine the acoustic preferences of females from different subspecies and populations. Female M. m. domesticus differ from M. m. musculus by being indiscriminant towards the urinary scent of males from their subspecies [34,40,66]. Therefore, females from other M. m. musculus populations, subspecies and species need to be tested with respect to their USV preferences.
Spectral features that differentiated the two species in our study were amplitude and frequency, whereas subspecies and populations mainly varied in duration and frequency, the latter two being consistent with differences among inbred mouse strains [49] and between wild and laboratory Peromyscus mice [25]. The calls of mound-building mice were quieter and of higher frequency than those of house mice. In addition, we also found evidence that variation in frequency, entropy, and to a lesser extent duration, discriminated species. Within M. musculus subspecies, frequency, duration, and entropy were important discriminating features. At population levels, we identified amplitude, frequency, entropy and duration as being important discriminatory parameters.
In addition to female preferences for conspecific male USVs (species recognition) and species differences in male USV characteristics, we found several results that should be considered in future studies on mouse USVs. First, our spectrographic analysis of USVs had a robust classification rate between species and subspecies and supported taxonomic differentiation between M. m. musculus and M. m. domesticus, but further research is needed to determine how well variation in USVs and which features coincide with phylogeny [67]. We used discriminant function analysis (DFA) and support vector machines (SVM) to discriminate species, subspecies, and populations based on the spectrographic features. In previous studies on bats, SVMs provided better classifications and accuracies than DFAs [68,69]; however, in our study, both methods provided similar results. SVMs may not have outperformed DFAs because our data set, that needed to be divided into testing and training sets for the SVM, was small relative to previous studies. Regardless, both SVM and DFA approaches performed well when classifying Mus USVs at the species and subspecies level. Second, we found significant variation in USVs among geographic populations of M. m. musculus, consistent with other evidence that male USVs are innate [20,70] and influenced by genetic differences [8,19,49,71,72]. Syllable repertoire varied among M. m. musculus populations, as well as between Mus species. Geographic variation in song repertoire has been found in songbirds [73], bats [74] and wild singing mice (Scotinomys: [11]), which may also be due to genetic differences, as well as learning ('dialects') [75][76][77]. In our study, emission rate of 6 out of the 7 syllable types differed between populations with 'Frequency upsweep' being the most common syllable type in all populations (Fig 3). We found that latency to vocalize varied significantly between the Mus species and among M. m. musculus populations. Variation in latency and vocalization rate are known to differ among laboratory mouse strains [19,49,[78][79][80], with dominant alleles coding for phenotypes with high vocalization rates and low latency [71]. Thus variation in USV production in our study may be influenced by genetic differences. Finally, our results have implications for understanding whether male USV phenotypes provide reliable indicators of their quality or condition to rivals or potential mates. We found no evidence that female house mice prefer more complex 'Frequency-Step' syllables even though 'Frequency-Step' syllables are emitted by males at higher rates under conditions of sexual motivation from female scent [5]. We also found no evidence that male age (range 65-543 d) had any effect on latency to vocalize or vocalization rate. In rodents, serum testosterone concentration diminishes with age [81,82], suggesting that androgen-mediated USVs should decrease with age [83], as does the expression of condition-dependent male secondary sexual traits in other species [84]. Future studies are clearly needed to examine which features of male USVs are attractive to females and whether they provide indicators of male quality or condition [7].
In summary, our study is the first to demonstrate significant variability in USVs among wild Mus species, subspecies, and M. m. musculus populations, and provides evidence that females can potentially use these signals to identify conspecifics. Thus, in addition to providing potential signatures of individual and kin recognition [5,8], house mouse females may use male USVs for interspecific discrimination.