The Lombard effect in male ultrasonic frogs: Regulating antiphonal signal frequency and amplitude in noise

Acoustic communication in noisy environments presents a significant challenge for vocal animals because noise can interfere with animal acoustic signals by decreasing signal-to-noise ratios and masking signals. Birds and mammals increase call intensity or frequency as noise levels increase, but it is unclear to what extend this behavior is shared by frogs. Concave-eared torrent frogs (Odorrana tormota) have evolved the capacity to produce various calls containing ultrasonic harmonics and to communicate beside noisy streams. However, it is largely unclear how frogs regulate vocalization in response to increasing noise levels. We exposed male frogs to various levels of noise with playback of conspecific female courtship calls and recorded antiphonal signals and spontaneous short calls. Males were capable of rapidly adjusting fundamental frequency and amplitude of antiphonal signals as noise levels increased. The increment in fundamental frequency and amplitude was approximately 0.5 kHz and 3 dB with every 10 dB increase in noise level, indicating the presence of noise-dependent signal characteristics. Males showed the noise-tolerant adaption in response to female calls in noise level from 40 to 90 dB SPL. The results suggest that the noise-dependent signal characteristics in O. tormota have evolved as a strategy to cope with varying torrent noise.

The Chinese concave-eared torrent frogs (Odorrana tormota, Ranidae) 1,2 inhabit vegetation along mountain streams at elevations of approximately 150-700 m, where the ambient noise generated by running water is broadband (about 0.1-48 kHz) with a peak near 1.7 kHz. During breeding season at male's calling sites, the background noise intensity is usually between 80 and 95 dB SPL (sound pressure level, 0 dB re. 20 μPa RMS). Males 3-5 and females 6,7 produce high-pitched calls containing spectral energy in the ultrasonic range (above 20 kHz), and their hearing extends to approximately 35 kHz for males 5,8 and 16 kHz for females 7 . This exceeds the upper limits of most anurans' frequency sensitivity, e.g., 8.2 kHz, as determined by the auditory sensitivity of nerve fibres innervating the basilar papilla 9 , or 5 kHz, as supported by current psychoacoustical data 10 . However, it is largely unclear whether or not concave-eared frogs exhibit noise-dependent vocal characteristics as a short-term response. An animal increases its vocal amplitude and changes other acoustic parameters in response to an increase in background noise, called the Lombard effect [11][12][13] and found in mammals [14][15][16][17][18][19] and some birds [20][21][22][23][24][25] . It is clear that some species exhibit the Lombard effect 26 , some don't 27 , and that it is unclear how widespread the phenomenon is in frogs.
However, some frogs living in habitat of high levels of stream noise produce vocalizations containing remarkable frequency modulations and/or high frequencies beyond the noise spectral range to overcome noise interference 3,[28][29][30][31] . Males of the Puerto Rican treefrog Eleutherodactylus coqui cease responding to synthesized calls if they are accompanied by intense broadband noise 32 . Males of the frog Eupsophus calcaratus increased their call rate and duration in response to exposures of band-pass noise between 700 and 2,700 Hz of various intensities (66-78 dB SPL RMS) 33 . Green frogs increase the peak frequency of their calls when suddenly exposed to loud anthropogenic noise 34 . Playback treatments caused spring peepers Pseudacris crucifer to produce shorter calls, the high-frequency noise treatment caused them to lower the frequency of their calls immediately after the noise ceased, and call rate did not change in response to playback 35 . Males of Cope's grey treefrog Hyla chrysoscelis did not regulate call amplitude, although increasing noise levels ranging between 40 dB and 70 dB induced predictable changes in call duration and rate 27 . The results from the grey treefrog do not support the hypothesis that voice amplitude regulation is a generic vertebrate mechanism for coping with noise 27 . A recent research demonstrates that male túngara frogs (Physalaemus pustulosus) exhibit the Lombard effect, increasing call amplitude linearly with noise level in response to low-frequency noise (overlapping the species' call range) but not to high-frequency (nonoverlapping) noise 26 .
Our study subject, Odorrana tormota, is a nocturnal ranid species living beside noisy streams in southern China 1 . Males of this species are unique and their eardrums are deeply sunken at the far end of "ear canals" with unobstructed openings at the body surface 3 . Males produce ultrasonic calls, preventing frequency overlap with stream noise, as long-term acoustic adaptations [5][6][7] . Recently, an efficient form of courtship in the frog was reported, in which males answer the female courtship calls (FC) with short-latency antiphonal calls with different fundamental frequencies (F 0 ) in the field and indoors, that are positively correlated with the level of ambient noise, and receptive males exhibited hyperacute phonotaxis toward FC playbacks 6 . Antiphonal vocalization has been regarded to have an important function in nocturnal or subterranean animals living in an environment where visual cues are unreliable 36 . In this study, we recorded and compared antiphonal signals (ASs, elicited with FC playbacks) and spontaneous short calls (SSCs, in the absence of FC playback, obtained in respective noise) in response to increasing noise levels to investigate whether or not and how males of O. tormota cope with various levels of noise exposures.

Results
A total of 114 receptive males of O. tormota were tested for each condition in which broadcasting randomly with combined FC noise at different SNR of 32, 22, 12, and 2 dB, respectively, and a total of 12 FC playbacks for each trial (see METHODS). The antiphonal signals (ASs, elicited with FC playbacks) and spontaneous short calls (SSCs, in the absence of FC playback, obtained in noise) were recorded. The AS number per frog was from 0 up to 8 (median = 2.36) dependent on SNR, but only one AS per frog was collected for analysis at each SNR condition. The total number of ASs was 269, i.e., 36, 100, 93, and 40 ASs at the SNR of 32, 22, 12 and 2 dB respectively. Natural torrent noise. The natural torrent noise was recorded at the upper reaches of the Taohua creek (elevations 650-700 m). Figure 1A shows the spectral characteristics and amplitude envelop of the noise at 83 dB SPL (0 dB = 20 μPa RMS). The noise contains prominent acoustic energy in low-frequency range with a peak near 1.7 kHz, decaying by 5 dB at 100 Hz, by 20 dB at 8.0 kHz, by 37 dB at 20 kHz, and by 57 dB at 48 kHz. The spectrum does not change with the average noise level. Figure 1B illustrates a clip of the combined FC noise with the signal-to-noise ratio (SNR) of + 2 dB, depicting 5 harmonics of the FC (duration 140 ms) with a fundamental frequency (F 0 ) at about 9.0 kHz. Apparently, the natural noise overlaps partly with FC's harmonics only in a low-frequency range (0.1-20.0 kHz).
Two basic types of antiphonal signals. We recorded antiphonal signals (ASs) (elicited with FC playbacks) and spontaneous short calls (SSCs) (in the absence of FC playback, obtained in noise) from individual receptive males by using various combined FC noise exposures with 4 different SNRs (32,22,12, and 2 dB, respectively; respective noise level: 53, 63, 73, and 83 dB SPL; FC playback at 85 dB SPL). The ASs were single-note short calls (duration <100 ms; delay times or latency <400 ms, measured from the onset of FC playback) of two types, a constant-frequency tone (CF-type) (63.2%, 72 of 114 males) with an unmodulated-frequency F 0 ( Fig. 2A; F 0 6.9 kHz) and a frequency-modulated sweep (FM-type) (36.8%, 42 males). Most of the FM-type ASs were a Noise-dependent fundamental frequency regulation of antiphonal signals. The effects of combined FC noise with different SNRs on antiphonal signals were evident. For the same male, when the SNR of combined FC noise decreased from 32 dB to 12 dB (i.e., respective noise level increased from 53 to 73 dB SPL), the average F 0 of an approximate CF-type AS rose from 8.0 kHz (Fig. 3A) to 8.7 kHz (Fig. 3B), the duration shortened from 106 to 66 ms and the sound level increased from 86 to 89 dB SPL. As for all CF-type ASs recorded from 72 males, the population mean of F 0 of CF-type ASs exhibited statistically significant differences and positively correlated with increasing noise levels as determined by one-way ANOVA ( Fig. 3C; F 3, 176 = 29.093, P < 0.0001, n = 27, 66, 60, and 27 males, respectively, for noise level group of 53, 63, 73, and 83 dB SPL) with a dynamic range of F 0 from 5.14 to 6.72 kHz when noise level changed in a range from 53 to 83 dB SPL. A Tukey HSD post-hoc test revealed that there were statistically significant differences in the f 0 of CF-type AS among the 53 dB SPL group (5136.1 ± 130.2 Hz), the 63 dB SPL group (5830.0 ± 89.8 Hz) (P < 0.0001), the 73 dB SPL group (6305.0 ± 83.2 Hz) (P < 0.0001), and the 83 dB SPL group (6716 ± 133.5 Hz) (P < 0.0001) (Fig. 3C). There were no statistically significant differences in the f 0 of CF-type AS between the 73 dB SPL group and the 83 dB SPL group (P = 0.052). Under the same noise level, the mean F 0 of ASs was significantly greater than that of SSCs (P < 0.0001), while the F 0 of SSCs did not differ among noise treatments (F 3, 170 = 0.618, P = 0.604).
The F 0 increments of both ASs versus SSCs rose significantly from 0.96 to 2.49 kHz with the noise level increasing, an average of 0.5 kHz with every 10 dB increase in noise level ( Fig. 3D; F 3, 176 = 30.354, P < 0.0001; P = 0.034 between the 73 dB SPL group and the 83 dB SPL group). The data demonstrate that the F 0 variation of CF-type ASs was a function of the ambient noise level and males could regulate fundamental frequency of antiphonal signals in direct proportion to the level of noise exposure.
The FM-type ASs elicited from males by combined FC noise exposures with various SNRs also exhibited noise-dependent signal characteristics in the average and maximum of F 0 . As an example shown in Fig. 4A-C, when noise level increased from 63 dB SPL to 73, and 83 dB SPL, the corresponding maximum of F 0 of ASs increased from 7.5 kHz (Fig. 4A), to 8.0 kHz (Fig. 4B) and 10.3 kHz (Fig. 4C), respectively. Similarly, the average F 0 increased from 6.0 kHz to 6.3, and 7.4 kHz, respectively, and the bandwidth of FM sweeps became broader, increasing from 2.1 kHz to 2.7, and 5.1 kHz, respectively, whereas the minimum remained nearly unchanged between 5.2 and 5.4 kHz. Statistical analyses of all FM-type ASs recorded from 42 males show that both the population average and maximum F 0 exhibited significant differences and positively correlated from 6.06 to 7.13 kHz for the average and from 7.70 to 9.15 kHz for the maximum, respectively, as noise levels increased (Fig. 4D) (F 3, 84 = 5.418, P = 0.002 for average; F 3, 84 = 3.587, P = 0.017 for maximum; n = 8, 34, 33, and 13 males, respectively, for noise level group of 53, 63, 73, and 83 dB SPL). A Tukey HSD post-hoc test revealed that there were statistically significant differences in average f 0 of FM-type AS among the 53 dB SPL group (5988.6 ± 120.0 Hz), the 73 dB SPL group (7078.7 ± 198.4 Hz) (P = 0.016), and the 83 dB SPL group (7130 ± 223.6 Hz) (P = 0.032), and between the 63 dB SPL group (6433.7 ± 131.5 Hz) and the 73 dB SPL group (7078.7 ± 198.4 Hz) (P = 0.024) (Fig. 4D). There were no statistically significant differences between the 53 dB SPL group and the 63 dB SPL group (P = 0.533), and among the 63 dB SPL group, the 73 dB SPL group and the 83 dB SPL group (P = 0.157).
The averages of F 0 were significantly different from those of SSCs at respective noise levels (P < 0.0001). However, the minimum F 0 did not change statistically from 4.66 to 5.5 kHz with noise treatments (F 3, 84 = 2.638, P = 0.055). Furthermore, the increment of average F 0 of FM-type ASs relative to SSCs increased significantly from approximately 1.39 to 2.67 kHz in the noise level range from 53 to 83 dB SPL ( Fig. 4E; F 3, 84 = 5.518, P = 0.002), having an average increment of 0.44 kHz per 10 dB of noise level increase. Figure 4F illustrates the relationship between the bandwidth of FM-type ASs and noise levels, indicating the bandwidth presented a marked increase from 2.83 to 4.33 kHz with increasing noise levels (F 3, 84 = 5.262, P = 0.002). It is evident that the both the F 0 average and maximum of FM-type ASs were also regulated depended on the level of noise exposure. Noise-dependent amplitude regulation of antiphonal signals. To clarify whether or not the Lombard effect occurs in frogs, we analyzed the amplitudes of males' ASs elicited by combined FC noise exposures with various SNRs. We found that males increased amplitude of ASs in response to increasing noise level significantly and positively correlated with noise level, on average, with an amplitude increment of 3 dB with every 10 dB increase in noise level as determined by one-way ANOVA (F 3, 146 = 11.13, P < 0.0001, n = 22, 66, 51, and 11 males, respectively, for noise level group of 53, 63, 73, and 83 dB SPL for CF-type ASs; F 3, 65 = 6.405, P < 0.001, n = 4, 33, 24, and 8 males, respectively, for noise level group of 53, 63, 73, and 83 dB SPL for FM-type ASs) (Fig. 5). A Tukey HSD post-hoc test revealed that there were statistically significant differences in amplitude of CF-type AS among the 53 dB SPL group (62.7 ± 1.3 dB), the 63 dB SPL group (68.0 ± 0.9 dB) (P = 0.005), the 73 dB SPL group (71.3 ± 0.7 dB) (P = 0.0001), and the 83 dB SPL group (72.6 ± 1.6 dB) (P = 0.0001), and between the 63 dB SPL group and the 73 dB SPL group (P = 0.029) (Fig. 5). There were statistically significant differences in amplitude of FM-type AS among the 53 dB SPL group (61.4 ± 4.5 dB), the 63 dB SPL group (69.5 ± 1.1 dB) (P = 0.032), the 73 dB SPL group (73.1 ± 0.7 dB) (P = 0.001), and the 83 dB SPL group (72.8 ± 0.8 dB) (P = 0.006) (not shown in Fig. 5). In fact, increases in noise level resulted in significant rises in the amplitude of SSCs (in the absence of FC playbacks) as well (F 3, 115 = 28.35, P < 0.0001, n = 41, 41, 32, and 5 males, respectively, for noise level group of 53, 63, 73, and 83 dB SPL for SSCs). Under the same level of noise, there were no significant differences in amplitude between ASs and SSCs (P > 0.11), except the comparison between FM-type ASs and SSCs at 83 dB SPL, in which mean amplitude of FM-type ASs was lower than that of SSCs (P = 0.02) (Fig. 5). The minimum signal-to-noise ratios of the recorded AS and SSC were − 8.3 dB and 1.5 dB, respectively. The findings indicate that males of O. tormota emit signals of higher intensity as noise levels increase, revealing the presence of the Lombard effect in the frog. Furthermore, the duration of all CF-type and FM-type ASs at each SNR condition was analyzed. For SNR of 32 dB, the average duration was 56.6 ± 15.5 ms (mean ± SD; n = 36); 59.2 ± 20.7 ms (n = 100) for SNR of 22 dB,  59.7 ± 21.2 ms (n = 93) for SNR of 12 dB, and 60.2 ± 19.5 ms (n = 40) for SNR of 2 dB, respectively, showing a little longer about 3 ms with increasing noise levels with the exception of data for a single CF-type AS (shown in Fig. 3). Statistically, the population mean of AS duration was 59.2 ± 20.1 ms and there was no significant difference in the AS duration between each SNR condition (ANOVA, F 3, 265 = 0.24, P = 0.868).
The delay times of ASs measured from the FC onset at each SNR condition was analyzed as follows: 283.0 ± 40.8 ms for SNR of 32 dB (n = 36), 273.8 ± 44.0 ms for SNR of 22 dB (n = 100), 272.0 ± 38.1 ms for SNR of 12 dB (n = 93), and 278.2 ± 43.7 ms for SNR of 2 dB (n = 40). When SNR was reduced from 32 dB to 22 or 12 dB, the AS delay times became shorter of 10 ms, whereas there was a delay of about 5 ms for SNR of 2 dB. Statistically, there was no significant difference in AS delay times between each SNR condition (ANOVA, F 3, 265 = 0.700, P = 0.553).

Discussion
Antiphony enables to acknowledge that one's signal was received with certainty 36 . Antiphonal signals (ASs) evoked from males due to FC stimulation play a vital role in courtship for mate attraction, particularly for nocturnal frogs 6,7 . Thus, it is of significance to know the influence of noise levels on ASs and how males adjust ASs to ambient noise. The present data first demonstrate that males can simultaneously regulate noise-dependent frequency and amplitude of ASs. First, the increases in noise level result in a significant increase in the fundamental frequencies (F 0 ) of ASs, about 0.5 kHz with every 10 dB increase in noise level (Figs 3D and 4E). This frequency shift may provide a masking release of 1-3 dB, a case seldom found in other anuran amphibians 13,27,37 . Second, the amplitudes of both ASs and SSCs were positively correlated with noise level (see Fig. 5), about 3 dB with every 10 dB increase in noise level. These findings reveal the presence of the Lombard effect in the concave-eared frogs. This dynamic response of males may reduce noise masking of their response to females. Third, average population ASs duration of around 60 ms was little affected by noise in O. tormota, with the exception of data for a single CF-type AS (shown in Fig. 3). By contrast, increases in noise level resulted in significant increases in call duration in Cope's grey treefrog 27 and in leptodactylid frogs of the temperate austral forest 33,38 ; in both cases, call amplitude did not change, though a little support for noise-dependent voice amplitude regulation was found in leptodactylid frogs. However, male túngara frogs exhibited a Lombard effect in the range of 1-3 dB with every 10 dB increase in noise, showing call amplitude linearly with low-frequency noise level, but no frequency shift in the frog 26 .
Next, we usually recorded ASs from receptive males when the noise levels were between quiet (approximately 40 dB SPL) indoors and noisy conditions (up to 90 dB SPL) in field 6 . It suggests that the noise-tolerant adaptations in response to female calls have developed in the auditory and vocal pathways of O. tormota. We found that various levels of natural noise exposures (ranging from 35 to 85 dB SPL) have little effect on hearing in males 39 , implying the males are noise-tolerant in the noise level range. In addition, AS was seldom elicited from males when the signal to noise ratio (SNR) was set to less than − 5 dB.
To summarize, we have shown that male concave-eared frogs have evolved the capacity to regulate both frequency and amplitude of ASs to answer female courtship calls under different noise levels as a strategy to deal with varying torrent noise. Nevertheless, these findings are dissimilar to those in echolocation bats 19 and in urban birds 40 , in which call amplitude and frequency shifts occurred independent of one another. The presence of the Lombard effect in male concave-eared frogs (they change both frequency and amplitude of ASs in different levels of noise exposure; For SSC, only amplitude is modified, not frequency) and túngara frogs (they change call amplitude, not frequency, in response to noise) 26 shows that there are at least two different mechanisms to deal with noise, as suggested by Slabbekoorn et al. 40 . It could support the speculation that the Lombard effect may be 'a synapomorphy of all amniotes, ' not only 'the outcome of a convergent evolution in birds and mammals' 13 . Noise acts as a strong selective force in sculpting acoustic communication systems, though the essential neuronal mechanisms mediating noise-dependent signal characteristics are less known. was recorded using a high-resolution digital recorder (Type 722, Sound Devices, USA; frequency range 10 Hz-96 kHz), a 1/4-inch pre-polarized condenser microphone (Type 40 BE, frequency response 4 Hz-100 kHz, + /− 3 dB, G.R.A.S., Denmark) and a preamplifier (Type 26 CB, G.R.A.S., Denmark) positioned at 45° from the water surface, at a distance of 3 m from the noise source and a height of 1.5 m. This is at a typical location of a calling male. A 1,000 Hz tone at 94 dB SPL produced by a sound level calibrator (Type 4231, Brüel & Kjaer, Denmark) was used as a reference signal. Compared with the reference signal, the sound pressure levels of various natural noise exposures were adjusted to 4 different levels ranging from 53 to 83 dB SPL in steps of 10 dB (time window 140 ms, the same duration as the signal FC length) by the Praat software package for the analysis of phonetics (from University of Amsterdam, The Netherlands, and maintained by Paul Boersma and David Weenink) 40 , and the spectrum of the noise exposure does not change with the average noise level. Then, acoustic stimuli consisting of the female call playbacks (FC, duration 140 ms at the level of 85 dB SPL) 6 combined with 4 different levels of natural noise (noise duration of 12 s; FC playback started 9 s after noise onset), here after referred to as "FC noise" with different SNR of 32, 22, 12, and 2 dB, respectively. The noise did not have a short ramp-on and off.

Methods
Acoustic playback experiments. Experiments were carried out from 1930 to 2200 h in a quiet and darkened indoor room with background noise at 40 dB SPL, ~1 km from the frog's natural habitat. Combined FC noise with 4 different SNRs was broadcast through a power amplifier and a full range speaker unit (FE87E, reproduction frequency response: 140 Hz to 30 kHz, 3 inch, 8 Ω, Fostex, Japan) with 3-s intervals of silence at the rate of 1 FC per 15 s over a 3-min period, a total of 12 FC playbacks for each trial, and 3-min silent periods between trials. A total of two 3-min trials per individual male were conducted for each condition. During recording sessions, the speaker unit was positioned 1 m away from a captive male frog, which remained motionless on the side of plastic basket covered with gauze. The same microphone and preamplifier were mounted on a tripod and positioned about 45° from the ground surface and 10 cm from the frog to monitor its vocalizations. In the experiments, 4 different FC noises were randomly applied, and antiphonal signals (ASs, elicited from males with FC playbacks) and spontaneous short calls (SSCs, without FC playbacks) were recorded using the same recorder (a sampling rate of 96 kHz). The main parameters for ASs and SSCs, e.g., basic types of calls, sound spectrograms, peak-to-peak amplitude (mean, maximum, minimum in dB SPL), duration and pitch (F 0 , maximum, minimum, average and sweep bandwidth), were measured with the Praat software package 41 and displayed using SELENA, a custom-designed program (S. Andrzheevski, St. Petersburg) 4,42 . The sound pressure levels of respective ASs and SSCs buried in various levels of noise were determined based on the decibel calculation of combining sound sources 43,44 . Frogs did not call during the 3 sec silent intervals.

Statistics.
To analyze the effect of the exposure to natural noise, we measure and compared the main parameters of ASs elicited with FC playbacks presented at the same 85 dB SPL and SSC from the males of O. tormota at 4 different SNR conditions. The AS number per trial per male was dependent on a given SNR in the range from 0 up to 8 (median = 2.36), but only one of the first or second AS in two trials per individual was collected for analysis at each SNR condition. There are the six assumptions for ANOVA: (1) Dependent variable should be continuous; (2) Independent variable consists of two or more categorical, independent groups; (3) Independence of observations, using independent individuals as subjects; (4) There should be no significant outliers; (5) The dependent variable is normally distributed in each group; and (6) Homogeneity of variances. We can check assumptions #4, #5 and #6 using SPSS Statistics. Before doing this, we made sure that our data meets assumptions #1, #2 and #3. Then, we performed a one-way ANOVA test and a Tukey's honestly significant difference (HSD) test, as a standard post-hoc test, for multiple comparisons using the SPSS Statistics 17.0 software package 45 .