Sound signatures and production mechanisms of three species of pipefishes (Family: Syngnathidae)

Experimental setup for sound recording

Six adult Syngnathoides biaculeatus, four adult Doryichthys martensii and three adult Doryichthys deokhatoides with mean heights (±standard deviation) of 19.4 ± 1.0 cm, 12.5 ± 2.0 cm and 10.9 ± 8.5 cm, respectively, were acquired from a local fish hobbyist shop and kept in separate aquariums (by species) for four weeks prior to the experiment. Experiments with Syngnathoides biaculeatus were conducted in an acoustic dampened tank (160.0 cm × 100.0 cm × 45.0 cm) filled with seawater, while experiments with either Doryichthys martensii or Doryichthys deokhathoides were conducted in smaller acoustic dampened tanks (60.0 cm × 45.0 cm × 40.0 cm) filled with freshwater. Both marine and freshwater tanks were lined on the inside with 1-inch polystyrene foam and air-filled packing wraps, with the tank bottom filled with sand to reduce resonance and reflection (Wysocki & Ladich, 2002). Each experimental tank was placed on a 2-inch thick foam block to further reduce resonance from background noise.

Sound recordings of individual pipefishes were conducted one at a time, over a period of two weeks. The alligator pipefish was first confined in a plastic mesh cage (30.0 cm × 20.0 cm × 45.0 cm; 0.3 cm mesh opening) placed inside the seawater acoustic tank and allowed to acclimatize for 48 h before sound recordings were made. To induce feeding clicks, the pipefish inside the mesh cage was fed with live poecilid fish larvae. Both freshwater pipefishes were not confined in any mesh cage inside the tank, and were fed with live brine shrimp nauplii. All mechanical filters and heaters were shut down two hours prior to sound recordings.

Audio signals emitted during feeding were recorded using a hydrophone (Model C55-F2-LAB: Cetacean Research Technology, Seattle, WA, USA) with a frequency range of 0.006–203 kHz. The hydrophone was omnidirectional with a sensitivity of −165 dBre 1 V/µPa; preamplifier gain: 20 dB connected to a compact flash recorder (Fostex FR-2 24 bit/192 kHz). The hydrophone was placed at mid-water level at the centre of the recording tank. The calculated minimum attenuation distance was 30.1 cm for the saltwater tank and 21.9 cm for the freshwater tank with a minimum resonant frequency of 2353.2 Hz for saltwater and 2802.8 Hz for freshwater. Both distances were below calculated levels based on the equations of Akamatsu et al. (2002). Background noises were pre-recorded as control before the experimental recordings were performed.

All experimental protocols involving the live seahorses were approved by the Institutional Animal Care and Use Committee, University of Malaya (UM IACUC), with Ethics References No. ISB/14/08/2012/ALCO(R). No fish deaths resulted from the experiments, and fish were returned following the experiment to their aquaria housed in the University of Malaya’s Marine Culture Unit.

Microtomography

Two specimens each of Doryichthys martensii and D. deokhathoides were studied by microtomography. The pipefish’s head was scanned and analysed using the microtomography equipment (SkyScan 1172, high-resolution micro-CT) and services provided by the Nuclear Agency of Malaysia. The head was mounted onto a polystyrene holder and placed into the scanner to fit into the field of view. Scan layers were set to 10 µm on a 180° rotation to acquire frames. The image frames were collated and reconstructed using a 3D imaging software (CTvox) to determine the cross section and morphological structure of the pipefish’s cranium particularly around the supraoccipital (SOC) region. Syngnathoides biacuelatus was not studied due to its large head which could not be mounted onto the scanning equipment.

Clearing and Staining

The cranial anatomy of alcohol-preserved pipefishes was further examined by using a clearing and double-staining method for bone and cartilage (Dingerkus & Uhler, 1977). For each species, three pipefish heads were cleared in 35.0 ml saturated sodium borate, 65.0 ml distilled water and trypsin powder for 24 h; and then double stained in 100.0 ml 1.0% KOH solution with 1.0 mg Alizarin Red stain for 48 h, and 30.0 mg Alcian Blue, 60.0 ml ethanol and 40.0 acetic acid for 36 h, respectively. Since the whole head could not be viewed in the same field under the microscope, parts of the same head were photographed under a stereo microscope (Leica M125) attached to a digital camera system (Leica MC170 HD; Leica, Wetzlar, Germany). All photos were then stitched together to reproduce the whole pipefish head using the Leica Application Suite (LAS) (version 4.3). Interpretations and descriptions of histological and microtomographic images are primarily based on the terminologies used by Leysen et al. (2011).

Wavelet analysis

Several studies have used the wavelet transform analysis as a signal processing recognition tool (Selin, Terunen & Tanttu, 2007; Ng, Muniandy & Dayou, 2011; Martinez et al., 2013). The continuous wavelet transform of a signal X(s) is defined as (Daubechies, 1992) ${\displaystyle W\left(a,t\right)=\frac{1}{\sqrt{a}}\int \psi \left(\frac{t-s}{a}\right)X\left(s\right)ds,}$ where the localized function ψ(⋅) is called the mother wavelet, a is the scale parameter and t is the time shift parameter. Scale parameter a functions to dilate (a > 1) and compress (a < 1) the mother wavelet ψ, hence creating a time-varying multiscale bases. The scale parameter is inversely proportional to frequency. When the analysed signal is differentiable, a mother wavelet of high regularity will be selected and vice versa. The corresponding ‘energy distribution’ to spectrogram is called wavelet scalogram, and is defined as S(a, t) = |W(a, t)|² (Rioul & Flandrin, 1992). This time-scale representation has an equivalent time-frequency expression that can be obtained by scale-frequency conversion $a=\frac{f_0}{f}$, where f₀ is the central frequency of the mother wavelet ψ. Scalogram has a varying window size at different scales in contrast to fixed-window size of the spectrogram thus preserving the time and frequency resolution.

Prior to analysing sound data, the background signal peak at 30 Hz was removed using wavelet detrending. The raw signal was decomposed into twelve dyadic scale levels using multiresolution decomposition with Daubechies N =5 mother wavelet. Then, the 30 Hz signal was reconstructed based on scale level 9 decomposition, and this so-called band-passed interference signal was removed from the raw signal to give the detrended signal.

The wavelet multiresolution was carried out using the Matlab Time-Frequency toolbox developed by Auger et al. (1996). Morlet wavelet of 50 half-length was used to analyse the wavelet scalogram. The scalogram was calculated over 128 analyzed voices and was bounded between 0.001–0.04 Hz normalized frequencies. Squared magnitude of continuous wavelet transform with threshold value over 0.5 was depicted in the scalogram. The energy spectral density that gives the frequency marginal of the scalogram was plotted side by side with the scalogram.

The parameters of the measured sound wave studied included the dominant frequency, which is the highest amplitude value of the measured signal, and the duration of the signal, which is the temporal length of the sound.

Statistical analysis

The mean of the measured sound parameters was calculated for each species. Sound parameters did not conform to parametric requirements of normality using Shapiro–Wilk’s test and homoscedascity using Levene’s test, even after logarithmic transformation. Hence, differences in the measured sound parameters between three species were tested using Kruskal–Wallis non-parametric test. The Mann–Whitney U-test was used to test the differences of the sound parameters of the low frequency component between the congeneric species D. martensii and D. deokhathoides. All statistical tests were carried out using Statistica 10.0 software (StatSoft Inc., Tulsa, Oklahoma, USA).

Sound characteristics

All three species of pipefishes (Doryichthys martensii, Doryichthys deokhathoides and Syngnathoides biaculeastus) produced high frequency, short broadband clicks during feeding strikes (Table 1) (Figs. 1–3). One click per head movement was consistently observed in all three species.

The feeding click waveform of D. martensii consists of a single pulsed burst at 0.54 ± 0.03 kHz which decays rapidly (10.37 ± 1.10 ms) (Figs. 1A and 1B). However, its scalogram also revealed a lower energy burst at higher dominant frequency of about 1.07 ± 0.04 kHz but with a shorter duration of 6.56 ± 2.47 ms (Fig. 1B) (Table 1). The power spectrum revealed an additional low frequency component of 0.17 ± 0.01 kHz, with duration of 26.05 ± 7.02 ms (Figs. 1C and 1D). This component was however observed in 77.8% of the total recorded signals (Table 1). The dominant frequency (Kruskal–Wallis test statistic, $H_{df\left(3\right),N\left(18\right)}^{\prime }=4.65$) and duration ($H_{3,18}^{\prime }=2.76$) of all recorded clicks were not significantly different (p > 0.05) among four individuals within species. The low frequency component also did not record any significant differences (p > 0.05) in its sound characteristics in terms of frequency ($H_{2,19}^{\prime }=1.30$) and duration ($H_{2,19}^{\prime }=4.42$) within species.

The waveform of D. deokhathoides revealed symmetrical low-amplitude wave before and after a high-amplitude spike (Fig. 2C). The scalogram revealed a broad band, high-frequency modulation energy burst of between 1.5 and 4.5 kHz which was of short duration (4.01 ± 1.18 ms) (Fig. 2B) (Table 1). This broad band signal appeared to be the merging of two components, 2.04 ± 0.3 and 3.84 ± 0.16 kHz with dominant or modal frequencies of 2.55 ± 0.15 kHz (Fig. 2C). A low frequency component was recorded at 0.20 + 0.03 kHz with an average duration of 10.35 ± 6.40 ms (Figs. 2A and 2B). However, this low frequency component was observed in only 29.2% of the total recorded clicks for the species (Table 1). Both dominant frequency ($H_{2,14}^{\prime }=5.97$) and duration ($H_{2,14}^{\prime }=3.94$) of recorded feeding clicks of D. deokhathoides were not significantly different (p > 0.05) among the three individuals. Similarly, the low frequency component displayed no significant difference (p > 0.05) for frequency ($H_{2,13}^{\prime }=0.70$) and duration ($H_{2,13}^{\prime }=0.36$) within species.

The feeding click of S. biaculeatus displays a multimodal sinusoidal waveform of higher amplitude (Fig. 3A) in contrast to the feeding clicks of both Doryichthys species. The waveform displays initial small precursor signals, followed by a gradual build-up of large wave signals in two or more overlapping pulses before they followed an extended decay. The double-pulse comprised an initial higher-amplitude (0.2 Pa) pulse of 4 ms, followed very quickly by a second lower-amplitude (0.1 Pa) pulse of 5 ms. The scalogram of S.biaculeatus displayed a localised energy burst at a dominant frequency of 2.09 ± 0.06 kHz with an average duration of 3.40 ± 0.75 ms (Fig. 3B), followed by repeated broadband energy releases in the frequency range of 1.8 kHz–6.6 kHz and over a duration of 8.64 ± 0.61 ms (Fig. 3C) (Table 1). Only one recorded click shows an additional low-frequency component which was detected with a dominant frequency of 0.82 kHz and duration of 11.79 ms (not shown in figure). The dominant frequency ($H_{5,43}^{\prime }=2.08$) and duration ($H_{5,43}^{\prime }=2.52$) of all recorded clicks of S. biaculeatus was not significantly different (p > 0.05) among the six individuals of the same species.

Statistical analysis revealed interspecies differences for both click frequency and duration (Table 1). Syngnathoides biaculeatus was significantly different ($H_{2,94}^{\prime }=54.35$) (p < 0.05) from D. martensii and D. deokhathoides in dominant frequency but only D. martensii for duration. The dominant click frequency and duration of D. martensii and D. deokhathoides was significantly different (p < 0.05) (Table 1). The recorded frequency of the low-frequency component did not display any significant difference (Mann–Whitney U-test: Z′ = 1.94; p > 0.05) between D. deokhathoides and D. martensii, except for the duration (Z′ = − 2.93; p < 0.05) (Table 1).

Cranial bone morphology and kinesis

All clicks (n = 94) were associated with head movement during feeding strikes. Microtomograph images of the head of D. martensii features a prominent 1st postcranial plate (POC1) positioned between the supraoccipital bone (SOC) and 2nd postcranial plate (POC2) (Figs. 4A and 4B). The POC1 (equivalent to the coronet in the seahorse) vaguely resembles a blunt arrow head with an anterior end that is slightly grooved and a caudal end that is bifid. The plate rests on respectively the caudal and rostral extensions of the SOC and POC2. Both extensions of the SOC and POC2 are slightly depressed and V-shaped or taper to the tip. All three bones have a dorsal ridge or carina. Only the dorsal carina of the POC1 runs from anterior to posterior end, while the dorsal carina of both SOC and POC2 runs short of reaching their posterior or anterior edge, each ending in a raised, short or blunt spine (Fig. 4A). The lateral section of D. martenssi further revealed that POC1 is not fused to, or articulated with the adjacent bones (Fig. 4B). The POC1 shows an arched structure resembling a fused double-chevron in transverse section (Fig. 4C). The long twin sesamoid bones in epaxial tendons (SEM) which run anteriorly from the neck to ligamentously join the SOC are clearly visible beneath the POC1. The arched lateral arms of the POC1 are close to both the post-temporal bone (POSTT) of the cranium and the cleithrum (CL) of the pectoral girdle. The linear system of guided interlocking cranial bone plates which are neither fused nor articulated facilitates bone movements. As the head of the pipefish flexes backwards when the SEM pulls on the SOC, the latter slides beneath the POC1 pushing it towards POC2. The SOC’s carina and raised posterior edge provides the “push” against the POC1 while POC2’s carina provides the “brake.”

Similar to the cranial structure of D. martensii, the POC1 of D. deokhathoides is visibly present between the SOC and POC2 (Fig. 5A). However, the bone morphology of all three bones differs from that of D. martensii. In D. deokhathoides, the SOC’s dorsal carina ends caudally as a raised blunt spine, while its inferior caudal extension is expanded like a fan with a long medial spine. Both POC1 and POC2 are narrow bone plates located between the twin SEM which run laterally on both sides. The POC1 has more pronounced caudal bifids than in D. martensii (Fig. 5A). The fan-like caudal extension of the SOC of D. deokhathoides is tucked underneath the POC1 (Fig. 5B). Similar to D. martensii, backward movements of the head would result in the SOC sliding under the POC1, and the latter (bifids) sliding over the POC2.

Clearing and staining of the head of S. biaculeatus revealed the presence of a V-shaped plate at the posterior end of the SOC and on the anterior end of the POC2, just like in D. martensii. However, the POC1 plate is distinctly absent leaving a space of about 2 mm between the SOC and POC2 in a fish of 193 cm height (Fig. 6A). Also clearly different from the two Doryichthys species is the absence of a dorsal carina and terminal spine on both the SOC and POC2 of S. biaculeatus. The twin SEMs are largely exposed between the SOC and POC2. During head flexion, the SOC is pulled backward by the SEM towards POC2. The large, caudal extension (beak-like) of the SOC thus slides over the rostrum of the POC2 which provides the rough stridulating surface (Fig. 6B).