An Acoustic Analysis of the Genus Microhyla (Anura: Microhylidae) of Sri Lanka

Vocalizing behavior of frogs and toads, once quantified, is useful for systematics, rapid species identification, behavioral experimentation and conservation monitoring. But yet, for many lineages vocalizations remain unknown or poorly quantified, especially in diversity rich tropical regions. Here we provide a quantitative acoustical analysis for all four Sri Lankan congeners of the genus Microhyla. Three of these species are endemic to the island, but Microhyla ornata is regionally widespread. Two of these endemics, M. karunaratnei (Critically Endangered) and M. zeylanica (Endangered), are highly threatened montane isolates; the other, M. mihintalei, is relatively common across the dry lowlands. We recorded and analyzed 100 advertisement calls from five calling males for each species, except for M. zeylanica, which only had 53 calls from three males suitable for analyses. All four species call in choruses and their vocal repertoires are simple compared to most frogs. Their calls contain multiple pulses and no frequency modulation. We quantified eight call characters. Call duration and number of pulses were higher for the two montane isolates (inhabiting cooler habitats at higher altitudes) compared to their lowland congeners. Microhyla zeylanica has the longest call duration (of 1.8 ± 0.12 s) and the highest number of pulses (of 61–92 pulses). The smallest of the species, Microhyla karunaratnei (16.2–18.3 mm), has the highest mean dominant frequency (3.3 ± 0.14 kHz) and pulse rate (77 ± 5.8 pulses per second). The calls separate well in the Principal Component space: PC1 axis is mostly explained by the number of pulses per call and call duration; PC2 is mostly explained by the pulse rate. A canonical means plot of a Discriminant Function analysis shows non-overlapping 95% confidence ellipses. This suggests that some call parameters can be used to distinguish these species effectively. We provide detailed descriptions for eight call properties and compare these with congeners for which data is available. This work provides a foundation for comparative bioacoustic analyses and species monitoring while facilitating the systematics of Microhyla across its range.

92 pulses). The smallest of the species, Microhyla karunaratnei (16.2-18.3 mm), has the highest mean dominant frequency (3.3 ± 0.14 kHz) and pulse rate (77 ± 5.8 pulses per second). The calls separate well in the Principal Component space: PC1 axis is mostly explained by the number of pulses per call and call duration; PC2 is mostly explained by the pulse rate. A canonical means plot of a Discriminant Function analysis shows non-overlapping 95% confidence ellipses. This suggests that some call parameters can be used to distinguish these species effectively. We provide detailed descriptions for eight call properties and compare these with congeners for which data is available. This work provides a foundation for comparative bioacoustic analyses and species monitoring while facilitating the systematics of Microhyla across its range.

Introduction
Among amphibians, anurans (frogs and toads) are conspicuous for their calling (vocalizing) behavior [1][2][3][4]. Acoustically well-studied vocal repertoires of anurans provide vital information on many fronts: ascertaining species identities, especially when cryptic species are involved [5][6][7][8][9][10][11]; acoustic experimenting and understanding specific acoustic signaling in different behavioral contexts [12][13][14][15]; bioacoustic monitoring, surveys, remote sensing and population studies [16][17][18]. However, for this, the vocal repertoires of frogs need to be understood statistically. Such bioacoustic analyses for tropical regions of the world, where most of the anuran diversity exists, are almost non-existent [19,20]. This is also true for the genus Microhyla, widespread throughout the tropics across India, Sri Lanka, Indonesia, archipelagos of Ryukyu of Japan and Sulu of Philippines [21]. So far, 39 species of the genus have been identified [10,21], of which four (ca. 10%) are found in Sri Lanka. The island is now a well-recognized hotspot of amphibian diversity and endemicity [9,22,23], containing a unique assemblage of fauna and flora, distinct from Indian mainland, with only a few shared species between the two regions. The genus Microhyla also shows the same pattern of distribution, with three species being endemic to the island, and one shared species [10].
Due to their widespread nature, only 4 of the 39 species are listed as threatened by the IUCN, 2015 [24]; however, of these, two species, M. karunaratnei (Critically Endangered) and M. zeylanica (Endangered), are confined to two mountain peaks of the Central Hills and Rakwana Hills of Sri Lanka. Furthermore, M. karunaratnei is considered an EDGE (Evolutionary Distinct and Globally Endangered) species [25], the only species amongst Microhyla and the only Sri Lankan amphibian to be included in this category. Though these species are prioritized for conservation, their ecology, behavior and natural history is virtually unknown. A deeper understanding of these threatened species will enable their effective conservation.
Here, we provide a quantitative description of the vocalizations for the three Sri Lankan endemic species, M. karunaratnei, M. zeylanica, M. mihintalei, and a Sri Lankan population of M. ornata. We provide descriptions of eight call properties measured for 353 calls from 18 individuals, which also help us evaluate the intraspecific variation within these species. We discuss and compare the vocalizations of these species and point out the knowledge gaps, thus providing the basis for species monitoring and systematics of Microhyla across its range.

Fieldwork and acoustic recordings
Calling males of Microhyla karunaratnei were recorded between 20:00-23:00 hours on 9 th December 2014 from a population in Morningside forest reserve, Suriyakanda, Rathnapura district (6.4075°N, 80.6094°E, 1050 m.a.s.l). They were calling around a breeding pool, an abandoned gem-pit in a regenerating forest patch with tall grasses and shrubs. Fairly common, calls of M. mihintalei and M. ornata were recorded from temporary breeding pools in Mihintale, Anuradhapura district (8.3548°N, 80.5054°E, 120 m.a.s.l) on 27 th September 2014 and Maakandura, Kurunegala district (7.3245°N, 79.9887°E, 30 m.a.s.l) on 25 th August 2014 respectively. Calls of M. zeylanica were recorded from a shallow ephemeral pool in a grassland habitat at Horton Plains National Park (6.7963°N, 80.8179°E, 2135 m.a.s.l) on 22 nd and 23 rd August, 2014 between 21:00-02:00 hours (Fig 1). Samples of call recordings used in the analyses with collection numbers of each species are provided as supporting information (S1-S4 Audio), which will help direct and wider comparison with other taxa [34].
Calls of five males from each species (except M. zeylanica, N = 3) were recorded using a digital recorder, Marantz PMD 620 MKII (sampling rate 44.1 kHz, 16-bit resolution) and a directional Sennheiser-ME66 microphone equipped with a foam windscreen. Microphone was handheld to maintain a distance of 0.3 to 0.5 m between calling male's snout and the microphone tip. Gain setting of the recorder was adjusted prior to each recording and maintained until the end of the given recording. Ambient temperature of the calling site was taken immediately after each recording using a handheld Kintrex IRT0421 non-contact infrared thermometer to the nearest 0.1°C. Snout vent length (SVL) and body weight (BW) of all recorded males were measured in situ using a precision digital caliper and a portable digital balance to the nearest 0.01 mm and 0.01 g respectively; one specimen from each species were taken as a reference sample and all other animals were released back to their original habitat upon taking measurements. Collected male was placed in a moist plastic container (200 ml) with some grass blades or leaf litter and transported to the field station in less than 3 hours. It was then ornata DZ1427). The species identification was done using morphology [10,[35][36][37][38]. Tissue samples (thigh muscles or liver) were taken immediately after euthanization and stored in absolute ethanol at -20°C for further analyses at the Department of Molecular Biology and Biotechnology, University of Peradeniya (reference numbers: DZ1530, DZ1420, DZ1445, DZ1427). Video recordings of vocalizing males were collected for all four species using a Canon EOS 60D digital SLR Camera and a Sony DCR-SR45 camcorder using the night vision mode.

Acoustical analyses
Only calls having a high signal to noise ratios that were free from overlapping calls of nearby males were used for the analysis. A total of 100 calls were measured from each species (20 calls per individual) except for one species, M. zeylanica, for which only 53 calls from three individuals were used due to the paucity of non-overlapping calls. Calls emitted by all species contained multiple pulses. Two species, M. karunaratnei and M. zeylanica, vary their call by dropping pulses (discontinuing the pulse train while maintaining longer duration between pulses several times within a call); however, this was observed only in five and four calls of the two species respectively. We excluded such calls from the analysis.
We measured eight call properties for this study (Table 1). These included temporal properties (call duration, call rise time, call fall time, 50% call rise time, 50% call fall time, number of pulses per call and pulse rate) and a spectral property (call dominant frequency by averaging spectrum over an entire call). Call characters were measured using methods and terminology from previous studies [1,19,20] as illustrated in S1 Fig. Raven Pro 1.4. was used to measure the call characters; temporal call characters were measured using Raven's waveform display and spectral properties were measured by averaging the spectrum over the entire duration of a call (256 pt. fast fourier transform, Hanning window). Descriptive statistics of the call characters; mean ( X), standard deviation (SD), range and percent coefficient of variation (CV ¼ SD= X Â 100; calculated as within species CV, where SD is divided by the mean value of a species) were computed using Microsoft Excel 2010. Median and interquartile range were calculated for indivisible characters (number of pulses per call). Systat version 11 (SYSTAT 11) was used to conduct a principal component analysis (PCA) on the correlations matrix, using four call property variables. Call variables with high CV values (i.e. having CVs above 35%); Call rise time, Call fall time, 50% call rise time, 50% call fall time were excluded from the PCA analysis. We also did a discriminant Table 1. Descriptions of acoustic properties measured. function analysis (DFA) using six call variables (two highly correlated call variables to call duration, call rise time and 50% call rise time, were removed from the DFA analysis) and produced a canonical means plot using the first two canonical variables to verify the results of the PCA. For all four species, we assessed the relationship of the call variables with temperature, SVL and body weight using Pearson product moment correlation analyses (correlation coefficient = ρ).

Results
Among Hierarchical organizations of calls (i.e. call groups and call bouts) were not considered as they vary widely across the recordings. For these species, calls were emitted as duets or in groups. A single male always initiated calling following periodic pauses, which induced other males to call; this was observed for all four species. Waveform, spectrogram and power spectrum of the most common call type (hereafter referred to as advertisement call) of the four Microhyla species are illustrated in Fig 2. Pulsatile structure of the call, emitted as a short series and the presence of a single prominent frequency band is common to all species (Fig 2). Frequency modulations were not observed in any species.
The advertisement call of M. karunaratnei contains between 50 to 95 pulses having an average pulse rate of 77 pulses per second. The call duration ranged between 700 and 1172 ms. ( X ¼ 866 AE 101 ms). Call dominant frequency ranged between 3.1 and 3.4 kHz ( X ¼ 3:3 AE 0:1 kHz). It takes an average of 661 ms to reach its maximum amplitude and decreases rapidly within the next 205 ms. Call duration of M. zeylanica ranged between 1500 and 2000 ms ( X ¼ 1852 AE 123 ms). It contains 61 to 92 pulses having an averaged pulse rate of 44 pulses per second. The call rises gradually within 1273 ± 243 ms and decreases slowly over the last 575 ± 178 ms. The advertisement call of M. mihintalei and M. ornata were short, less than half a second. Advertisement calls of M. mihintalei consist of 9 to 15 pulses, which had a rate of 58 pulses per second. Call duration ranged between 141 and 245 ms ( X ¼ 187 AE 24 ms). Call typically reached its full amplitude under 144 ms ( X ¼ 80 AE 30 ms) and decreased in amplitude over the last 106 ± 44 ms. Dominant frequency ranged between 1.3 and 2.6 kHz. Call duration of M. ornata was 300 ± 44 ms (210-795 ms). It composed 9 to 14 pulses, having a pulse rate of 41 pulses per second. Dominant frequency ranged between 2.2 and 3.4 kHz. Average call rise time and fall time were 154 ± 43 ms and 145 ± 61 ms respectively (Table 2).
Four temporal characters, call rise time, call fall time, 50% call rise time and 50% call fall time of the four species show the highest variation within species (CV range between 19% and 86.4%), whereas other four characters, call duration, dominant frequency, pulse rate and pulses per call were less variable, CV below 19% ( Table 1).
The Principal Component analysis (Fig 3) on the correlations matrix, using four call property variables (Call duration, Dominant frequency, Number of pulses per call and Pulse rate) from the four species shows that none of the species overlap in PC space. PC 1 axis explains 53% of the total variation, mostly by the number of pulses per call (factor score = 0.985) and call duration (factor score = 0.874); PC 2 axis is explained by the pulse rate (factor score = -0.805), which explains 25% of the variance. Microhyla zeylanica overlaps with M. karunaratnei in PC1 axis, but does not overlap with any other form in PC2 axis. Microhyla karunaratnei slightly overlaps with both M. ornata and M. mihintalei in PC2 axis. Though M. ornata and M. mihintalei are separate on the PC space, they overlap in both axes. Further the stepwise backward DFA (Fig 3) identified each species more than 99% as correct (100% of M. karunaratnei and M. zeylanica and 99% of M. ornata and M. mihintalei, Wilks' lambda 0.0002); similar results were obtained for the jackknifed dataset. Three discriminant functions were generated, and eigenvalues of the first two variables were 59.09 and 15.76 respectively. The centroids are

Discussion
All four species studied here have pulsatile calls, which also seem to be a characteristic feature of the genus [26][27][28][29][30][31][32][33]. Hence the vocal repertoire within the genus is less complex when compared to other taxa studied from the region, such as most rhacophorids, ranids and bufonids [9,14,20,39,40]. Most of the call characters can be explained in relation to phylogenetic relatedness and by habitat occupation. For microhylids, the phylogenetic relatedness seems to have played a greater role than the habitat influence in the evolution of call characters [41]. Almost all members of the genus are pool breeders where the physical resources needed for females are concentrated. The males advertising in choruses from such optimal habitats are well known for Microhyla [26,29,31]. Choruses are also useful in several aspects such as increasing attraction to females and reducing predation risk by finding refuge in numbers [1]. But the cost of calling is higher in choruses due to individual reproductive success; hence the males make periodic pauses, which are essential to recover and conserve their energy.
A well-known strategy for adapting to increasing ambient temperature is shortening the call duration and reducing the number of pulses [42,43]. This is evident for the cool-adapted M. karunaratnei and M. zeylanica in having comparatively longer calls with higher number of pulses when compared to the members inhabiting the warmer drier lowlands.
Among the four Sri Lankan species, M. mihintalei shows the highest within species variation in all call characters except for call duration and call fall time. However wider comparisons cannot be done because most other studies have considered only a single individual.
Vocalization of M. karunaratnei and M. zeylanica were distinct from all others especially having longer call durations where the call duration of all other members were below 0.85 s (S1 Table). In fact M. zeylanica has the longest call duration (1.5 to 2 s) among all microhylids. Pulse rate of M. rubra from India has been reported as having the highest (108 pulses/s) among the clade (S1 Table). This may have occurred as a result of following a different terminology when characterizing the pulses. Microhyla petrigena is distinct within the group in having several unique characters such as the smallest in body size (SVL = 14 mm), highest dominant frequency (5100 Hz) and highest pulse rate (89 pulses/s). Call characters of M. ornata from India are similar to that of the Sri Lankan population. Vocalization (call characters) of M. laterite and M. sholigari are closer to the two Sri Lankan species M. karunaratnei and M. zeylanica than to other members of Microhyla. Microhyla laterite, so far, has the highest number of pulses per call among all congeners (S1 Table).
The only known locations of M. karunaratnei had been subjected to small-scale gem mining. These frogs however now are utilizing those abandoned gem pits, now overgrown with vegetation, for breeding. It shows that they are somewhat adaptable to changing conditions. Acoustic exploration of potential habitats can be used for rapid and accurate identification of new populations of this cryptic frog.
Acoustic analysis made here facilitates the comparative understanding of vocalization for the genus. Furthermore, this work will help non-destructive anuran surveys, identification of new populations, population monitoring and behavioral experimentation.