Ten P. kuhlii bats were captured under permit from the Israeli National Park Authority (permit no. 2016/41421). They were housed at Tel Aviv University’s Zoological Gardens under a reversed light-dark cycle and a temperature of 23–26°C. Experimental protocols and procedures were approved and performed according to the Institutional Animal Care and Use Committee of the Israel Ministry of Health (Ethics permit: 04-18-026).
Experimental setup and training
The flight room where the bats were flown comprised a 5.5 × 4.5 × 2.5 m3 room with acoustic foam on the walls and ceiling (Additional file 1: Fig. S1). Audio recordings were made using 46 ultrasonic wide-band microphones (CM16) connected to four Hm1216 AD converters (Avisoft Bioacoustics), synchronized by injecting an SMPTE code (Horita) into the least significant bit of their first channel. The microphones were evenly spread in four rows around the perimeter of the room at 100 cm intervals and at heights of 0 cm, 60 cm, 120 cm, and 180 cm. The bats were trained to search for and land on a platform where mealworms were offered, except during the mouth-gape assessment experiments, in which there were no mealworms on the plate in order to prevent the bats from chewing and interfering with data collection. The landing platform was a 15 cm diameter styrofoam sphere mounted on a ca. 100 cm high pole.
A total of ten bats took part in the experiments, five in the gape tracking experiments and five in beam reconstruction, in order to establish the relationship between the direction of the head and the beam (see below). The data obtained from the latter are presented in Additional files 3 and 4: Figs. S2-S3 reveal a very strong correlation between the direction of the “beam” and that of the head (R = 0.91, Pearson correlation). Panel A in Fig. 1 presents the data from one of these bats.
Tracking was performed using a Motion - Analysis Corp system. Sixteen cameras (12 Raptor 1280 × 1024-pixel cameras and 4 Raptor-12 4096 × 3072-pixel cameras) tracked the bats at a frame rate of 300 fps to a spatial accuracy of less than 1 mm. We confirmed experimentally that the system was able to track a moving reflector with an accuracy of ~1 mm and to detect movements as small as 2 mm between two markers (see paragraph below and ). Spherical reflectors were glued to the target (6 mm marker) and to the bats using skin bond latex cement (OTSO-BOND Montreal Ostomy Corp.). Two reflectors (1.6 mm diameter, 3X3 Designs Corp.) were glued on either side of the bat’s mouth, and two reflectors (2.4mm diameter, 3X3 Designs Corp.) were mounted on the center of the bat’s head in a cross-shape to enable tracking the head azimuth. Head azimuth relative to a target was defined as the angle between the horizontal direction vector of the head and the vector from the head to the target. The distance between the two mouth markers was adjusted by subtracting a bias of ~5mm—which was the additional distance we added due to our positioning of the markers slightly above and below the edge of the lip (we measured the bias for each bat when the mouth is closed and subtracted the exact individual bias from the measurement results). Even though the echolocation signal can be as short as ~1 ms, tracking the mouth at 300 fps is sufficient to monitor the changes in mouth gape because the lip movement is at least an order of magnitude slower (i.e., the period of a mouth opening cycle is in the order of 40 ms, as we validated from 800 fps videoing, Additional file 7: Movie S1).
Synchronization of movement and audio
The tracking system included a synchronized audio channel. The microphone connected to this channel was mounted on the platform next to one of the CM16 microphones (Avisoft Bioacoustics). Cross-correlating the two channels allowed us to synchronize the two systems to an accuracy of < 1 ms .
The echolocation calls were detected and marked using a custom-made MATLAB-based software “batalef” [32, 33]; they were then manually scrutinized.
Selection of mouth gape related calls
For the acoustic analysis of the gape condition we chose, for each beam, the microphone with the loudest calls; hence, the highest SNR and an approximation of an on-axis call. We then extracted for each call its ICI (inter-call-interval—defined as the time between the start of one call to the start of the following call in a sequence), duration (defined by the call segment in which the envelope drops by 12 dB relative to the peak), and the peak frequency (the frequency with most energy in the spectrum).
The gape was measured during call emission for each individual bat. Flight sequences in which the gape was measured were divided into approach and search phases according to an ICI of either more or less than 40 ms, respectively. In most analyses, we used trials that included both search and approach phases (totaling 45 trials for all 5 bats). For some analyses (i.e., head azimuth (Fig. 1D), mouth opening and speed (Fig. 1F and Additional file 5: Fig. S4) and call peak frequency (Fig. 1H)), we added trials incorporating only search behavior (a total of 23 additional trials for all bats).
Controlling for the tracking system accuracy
The accuracy of the tracking can be finer than the size of the marker (e.g., a < 1-mm tracking accuracy can be achieved with a 1.6-mm diameter marker), due to the system tracking the center of the spherical marker (as confirmed experimentally below). Therefore, as long as the two markers do not come close enough to be confused by the system, the size of the marker does not limit the tracking accuracy. This allowed us to track two markers as close as 6–7 mm apart (one on either side of the lips) using 1.6 mm markers. The system has already been used in several published studies for tracking movement on similar scales (e.g., [32, 34]). However, to validate its accuracy, we ran a series of control experiments to ensure that our system was able to measure movements in the order of a few millimeters (see Supplementary Figure 1 in Eitan et al, 2019 ). In our control experiments (previously published), we measured the position of a stationary marker in different locations in the flight room, with the position tracked over time to estimate the jitter (the standard deviation of the position). In addition, to assess possible error in estimating the distance between two stationary markers, we placed a pair of markers ~5 mm apart at 10 different positions in the room and tracked them over time. Next, to test the robustness of tracking the distance between two markers on a moving object, the same two markers were placed on the pendulum of a mechanical metronome (Wittner) which was moved around the room, while the two markers remained stationary relative to one another (~5 mm apart). This was repeated in nine positions in the room. Because, in this study, we were tracking two markers moving relative to each other (the lips) on a flying bat, we ran an additional control experiment (Additional file 8: Fig. S6). We placed one marker (1.6 mm diameter) on the bottom of the pendulum and another on the fixed base of the metronome (7 mm apart). We first estimated the distance between these two markers while the metronome was stationary and then while it was moved through the room. We repeated this for 10 positions in the room and the estimated error was in the order of 0.06 mm (Additional file 8: Fig. S6). All of the above controls were performed with the same camera setup and frame-rate as in the experiments. Finally, to demonstrate the performance of the tracking system, we provide a film showing the tracking on a metronome pendulum (Additional file 9: Movie S2).
Estimating beam direction
We reconstructed 2D emission beams to assess the bat’s acoustic gaze. An 8-ms window was automatically defined around each peak and its spectrogram was estimated (using an FFT window of 1,024 samples with a flat-top window of 512 samples and an overlap of 480 samples). Any spectral content outside the main frequency range of the bat (35 kHz to 90 kHz) was also nullified. By applying MATLAB’s medfreq() function to the spectrogram (after thresholding), we obtained our primary assessment of the signal’s ridge—a frequency over time vector representing the center of the signal in the spectrogram. This function estimates the median frequency of the spectrogram for each time sample. Based on our preliminary knowledge of the downward chirp-like shape of the bat’s calls, we searched for the start of a monotonic decrease in the above function’s output that marked the beginning of a call. The call was terminated either when reaching the end of the ridge (i.e., a 6 dB decrease from its peak) or when its frequency started climbing, suggesting that it had reached the start of an echo. This procedure was iterated several times.
Following estimation of the signal’s ridge, we were able to estimate signal intensity for each microphone and each frequency. Finally, based on the location of the bat when the signal was emitted, we compensated for both the spread loss and air attenuation and calculated the position of each of the microphones relative to the bat during the emission, which provided the samples used later in the beam reconstruction process. Due to the limited 3D spread of the microphones, we only reconstructed the 2D cross-section of the beam. Next, the direction of each beam was obtained by fitting the samples (recorded in all channels) to a Gaussian beam model and taking its maximum as the direction of the emission. Any beam that did not present a good fit to this model (i.e., the R-square in the least mean square analysis was lower than 0.7) was excluded. This procedure was aimed at removing low-quality beams and resulted in the exclusion of ~ 15% of the data.
Boundary element model simulation
Employing a previously used method , we CT scanned (VECTor4 CT, MILabs) five deceased P. kuhlii individuals (the best preserved, with one having died less than 12 h earlier). The bat’s mouth was open during the scan and, based on the data obtained, the two bats with the best scans were selected. The two selected scans were converted into a triangular stl mesh using Amira 6.2.0 (Thermo Fisher Scientific) and down-sampled using 3dsMax (Autodesk) and Meshmixer (Autodesk). The bat’s internal mouth and lips had the highest mesh resolution in the model (0.1 mm; 30-fold shorter than the shortest wavelengths used), and only this part was shown to contribute significantly to the beam. Mesh edge length was increased progressively to 1.2 mm on the exterior features, which had their normals directed away from the measurement area.
The triangular mesh was verified to be closed in on itself, that is, to have no holes, no non-manifold vertices, and with all faces being coherent. We used BEMFA (boundary element modeling) to calculate the beam emitted from the bat’s mouth at a distance of 0.5 m using 8281 measurement points spaced 2° apart from − 90 to 90 °, vertically and horizontally. We verified that pressure generated by the 1 Pa source inside the rear of the bat’s mouth did not leak into the head itself (which could happen if the mesh was faulty). Simulations of the sound field emanating from the mouth were performed at frequencies from 35 to 110 kHz with steps of 5 kHz. We highlighted the bats’ peak signal frequencies during search and approach, respectively (40 and 50 kHz). In each simulation, the Helmholtz equation simulating the complex sound pressure inside the scene, with the entire head as boundary, was solved using the CHIEF method . For a complete description and benchmark testing of BEMFA software, see Boonman et al. . Using 3dsMax, we modified the bat’s gape by deforming the mesh by means of slight rotation and translation to within a range of natural postures with 10 different gape widths (the degree of opening was based on the observations carried out on the real bats), from each of which we calculated the sound-beam. In essence, parts of the inner mouth, such as tongue, palate, and inner cheeks, became acoustic radiators depending on the resonance at the given frequency, while the gape width represented variable apertures in elevation, from which these radiators can emit into the far field. We compared our results with a simple piston model with circular aperture whose radius or frequency can be varied (see supplementary materials for Additional file 10: equation and Additional file 11: Fig. S7).
For both the acoustic simulations and the measurements taken from live bats, gape width refers to the minimum distance from upper lip to lower lip during call emissions.
For the GLM analysis of the ICI, duration, mouth gape, and frequency (Fig. 1B, C), we used time bins of 0.05 s. We then examined whether there was an effect of the time (aligned to “0”) on any of the four parameters. We used a mixed effect generalized linear model (GLM) with ICI/duration/frequency/mouth gape as the explanatory parameter, time as the fixed factor, and the individual bats as a random effect. Random effects were set as intercepts (See also Additional file 2: Table S1).