Field study

During the summers of 2017, 2018, and 2019, we tagged and recaptured 11 post-lactating, female greater mouse-eared bats (Myotis myotis) with on-board tags [18] (inset in Fig. 2a) to sample their acoustic and foraging behavior. The bats were caught in the early mornings with a harp trap upon their return to the roost at Orlova Chuka cave, close to Ruse in northeast Bulgaria. The tags were wrapped in balloons for protection and glued to the fur on the back between the shoulders with skin-bond latex glue (Montreal Ostomy Osto-bond, Vaudreuil-Dorion, QC, Canada). The bats spent between one and 14 days equipped with the tags until we recaptured them at the cave or until the tags were detached from the bats and fell to the ground below the colony.

The tags included a Knowles ultrasonic microphone (Knowles, FG-23329, 2.6 mm diameter, Itasca, IL, USA) as well as triaxial accelerometers (inset in Fig. 2a). The accelerometers sampled at 1000 Hz (16-bit resolution) while the audio data were recorded at a sample rate of 187.5 kHz (16-bit resolution) and then filtered with a 10-kHz 1-pole high-pass filter and an anti-alias filter of 80 kHz.

All wild tag recordings were manually analyzed by displaying the acoustic and the acceleration data in 7-s segments with an additional option of playing back the audio [3]. The visualization included three separate panels with synchronized data: (i) an envelope of audio data filtered with a 20-kHz 4-pole high-pass filter to detect the echolocation calls; (ii) a spectrogram of audio data filtered with a 1-kHz 1-pole high-pass filter to visualize the full-bandwidth acoustic scene showing echolocation calls, conspecific calls, chewing sounds, and wind noise; and (iii) a plot of triaxial acceleration. Times of rest were separated from flight periods through the identification of wingbeat epochs as cyclic oscillations in the z-axis dimension of the accelerometer data [14]. By first applying a band-pass, delay-free, and symmetric finite impulse response filter (4–25 Hz, filter length: 1024 samples, sample rate: 100 Hz) on the z-axis dimension, we identified wingbeat epochs as time intervals where a running mean of 50 s was above 20 m/s2. All calls during flight periods were automatically detected by a call detector and visually inspected to ensure correct detection. Buzzes were defined as call sequences where call intervals (CIs) were below 14 ms [19] and were used to identify aerial prey attacks while excluding landing buzzes [3]. The onset of buzz I and II was defined as two consecutive calls with CIs below 14 and 7 ms respectively [19]. In addition, each prey attack was marked as “successful” or “unsuccessful” based on the presence or absence of chewing sounds [14].

Laboratory experimentLaboratory animals

During the summer of 2021, we captured three male and 15 post-lactating, female, greater mouse-eared bats at Orlova Chuka cave by using a harp trap when the bats exited the cave to forage in the evening. The bats were kept in temporary captivity for 2 months at the Siemers Bat Research Station in Tabachka. They had continuous access to water and were kept on a diet of live mealworms (larvae of Tenebrio molitor), supplemented with live moths captured with light traps in the vicinity of the station. At the end of the experimental period, the bats were released back into the wild at the site of capture.

Experimental setup

We trained the bats to fly in an anechoic room (5.3 × 2.8 × 2 m) and intercept a mealworm (hereafter termed “target”) hanging from a small prolate spheroid (semimajor axis 1.5 cm, target strength = − 22 dB re. 10 cm) that was attached with a thin fishing line (0.14 mm diameter) at 0.7 m from the ceiling (inset in Fig. 2b). The room was lined with 0.1 m deep acoustic foam (− 30 dB reflectivity re. hard wall at frequencies > 10 kHz) and the spheroid was covered with black electrical tape to exclude the use of visual cues by the bats. During the experiments, the room was kept dark, except for dim, red light from the headlamps of the two present researchers and the laptop computer screen at the recording station, which was oriented away from the target at a distance of 1.5 m. At 0.25 m from the ceiling, the fishing line with the spheroid was connected via a different fishing line to a target moving unit (drylin ZLW-1080, Igus, Cologne, Germany, step motor: NEMA23XL, step motor drive: ES-D808, Leadshine Technology Co., Shenzhen, China) outside the anechoic room. This step motor was controlled via a digital switch in the anechoic room and could move the target moving unit and hence the target two different distances (0.08 m and 0.24 m) and at four different speeds (0.5, 1, 1.3, and 1.5 m/s, Additional file 2: Movie S1) based on values from moths during evasive maneuvers [20]. The different treatments were divided in three groups: stationary (control), easy (0.08 m/0.5 m/s, 0.08/1, 0.08/1.3, 0.24/0.5), and difficult (0.24 m/1 m/s, 0.24/1.5). To assess the effect of the noise generated from the moving unit, we conducted mock trials where the unit was activated without any subsequent target movement (see section “Control experiment” below).

An infrared illuminator (96 LED, C&M Vision Technologies, Houston, TX, USA) was placed at the back end of the anechoic room to illuminate the flight path of the bats before target interception, which we recorded with an infrared camera (Intel RealSense Depth Camera D435i, Santa Clara, CA, USA) on the side wall of the room. The video data were sampled at a rate of 90 frames per second with an A/D converter channel in a multi-purpose USB device (USB-6251, National Instruments, Austin, TX, USA) and were used to estimate the time zero of movement and assess if the bats chased the target. Target movement and recording of video data were controlled via a custom-written LabVIEW (National Instruments) program (K. Beedholm).

Training procedure

To train the bats for intercepting the target below the spheroid, we first trained them to attack live, tethered moths placed in multiple sites within the anechoic room (including the position of the target) at 0.7 m from the ceiling. If a bat could not complete this task after three consecutive training days, it was excluded from the training procedure. For successful bats, the moths were replaced with mealworms until the bats were presented with only one tethered mealworm at the position of the target. Upon successful completion of this step, a staircase method [21] was used to introduce the final setup in two successive steps: (i) fishing line with a spheroid above the target and (ii) target movement (all distances and speeds). After 10 days of training, we selected four bats [three females (id #1–3) and one male (id #4) with mean mass 24.5 g over the captivity period] that could successfully complete the final task. While training with target movement, we used two criteria to maintain animal motivation: (i) a training session always started and ended with a stationary trial (i.e., no target movement) and (ii) a stationary trial was interspersed with three movement trials.

Tagging

During data collection, we attached a tag [18] (inset in Fig. 2a) to each bat to sample its acoustic behavior. The tags had the same specifications with the ones used for the field study and were glued to the fur on the back between the shoulders with a small piece of Velcro (1 × 2 cm) and skin bond latex glue (Montreal Ostomy Osto-bond). The tags weighed 2.8 g and therefore represented ca. 11% of the body mass of the bats. The bats were tagged during four (bat id #1–3) or six daily sessions (bat id #4) with each session lasting up to 45 min. After each session, the tags were removed via the Velcro connection and the general health status and skin condition of the bats were monitored. To test the effect of the tags, we randomly selected ten trials with and without tags deployed on each bat and compared their flight behavior from the video data. The absence of behavioral difference and the short duration of each tagging session suggest that the welfare of the bats was not compromised during the experiments [3, 22].

Experimental protocol

After the training period, we collected data on four and six consecutive days for bats id #1–3 and #4 respectively and all six movement treatments (i.e., 0.08 m/0.5 m/s, 0.08/1, 0.08/1.3, 0.24/0.5, 0.24/1, 0.24/1.5) were part of each session. We carried out one session per day and bat and conducted in total 631 trials. Before each session, the trial combinations were mixed in the same way for all bats via a pseudo-random Gellermann schedule [23]. The same criteria for animal motivation were used during data collection as during training, until four trials per movement treatment were completed. Two researchers were present in the room during data collection without handling the bats. One researcher started each trial by initiating the recording while the other researcher guarded the target to prevent the bat from intercepting it. Four seconds after the start of each trial, a 1-s long, trial-specific synchronization sound (frequency range: 3–6 kHz, hereafter termed “sync sound”) was emitted from a custom-build speaker (Aarhus University, Department of Biology Electronics Workshop) on the side wall of the anechoic room. Two infrared light-emitting diodes in the speaker were excited at the same time to allow for synchronization of the tag recordings with the infrared camera. After the sync sound, the researcher withdrew from guarding the target to allow for the bat to intercept it. During non-control trials, the other researcher initiated the movement of the target with the digital switch when the bat crossed a line on the floor 0.6 m in front of the target (Additional file 2: Movie S1). At this distance, the bat was already approaching the target and had initiated or was about to initiate the buzz phase of echolocation (Additional file 2: Movie S1). The first interception from the front of the target (Additional file 3: Movie S2) signified the end of the trial and the recording of video data. Upon successful mealworm interception, a new mealworm was placed below the spheroid, which was also returned to starting position after non-control trials. Three trials during which the first interception after the sync sound was instead from the back of the target were invalid and were discarded from further analysis.

Estimating time zero of movement, detecting calls, and scoring chase rate

To estimate the time zero of movement (t0) in each non-control trial, we visually inspected frame by frame the video recordings, which were imported into MATLAB (The Mathworks Inc., Natick, MA, USA) using the librealsense library (https://github.com/IntelRealSense/librealsense/tree/master/wrappers/matlab) and custom-written scripts. For each trial, we also inspected the video recordings with Intel RealSense Viewer v2.44.0 (Santa Clara, CA, USA) to score if the bats chased the target during the course of its movement. The bootstrap mean chase rate and 95% confidence intervals were generated after bootstrapping 1000 times for each treatment. Three trials during which it was not clear if the bats chased the target were excluded from further analysis.

For non-control trials, all calls belonging in a time window 2 s before and 1 s after t0 were automatically extracted and visually inspected for correct detections via a custom-written call detector. The same CI criteria used in the field recordings were applied to find the onset of buzz I and buzz II [19]. For control trials, we inspected the video recordings with the software Intel RealSense Viewer to estimate the time of the first interception from the front of the target and used this as a cue to select a time window 2 s before and 1 s after in the tag recordings. To generate a mock t0 for control trials, we made a distribution of the last CI just before t0 for all non-control trials. A mock t0 was estimated for each control trial when the bat reached a randomly selected CI value from this distribution.

To describe the change in normalized CIs after t0, we fitted an exponential decay model (\(normalized\;CI=0.6\cdot e^\frac+0.3\), t: time in ms since target movement) to the combined raw data from all treatments. All signal processing analyses were performed with custom-written scripts in MATLAB 2023a.

Control experiment

The noise generated from the moving unit could have an effect on the vocal behavior of the four bats. To control for this confounding factor, we carried out mock trials during which the moving mechanism of the unit was activated but no target movement took place. Data were collected on a single day for each bat at the end of the data collection period and all seven treatments were part of each session. With the same pseudo-random schedule, four trials per treatment and per bat were completed. During 108 mock trials and irrespective of treatment, all four bats chased the target while producing a buzz just before target interception.

Statistical analysis

All statistical analyses were implemented in RStudio (RStudio Team, 2023, version 2023.09.1 + 494, Boston, MA, USA). We used the lme4 package [24] to fit four generalized linear (mixed) models (GLMMs): one model was fitted to the data from the field to test if hunting success differed with buzz II duration (model 1); three models were fitted to the data from the laboratory experiment to test if the probability of chasing the target (model 2) and buzz duration (I and II separately, models 3 and 4) differed between the different treatment groups (i.e., stationary, easy, difficult). Models 1 and 2 had a logit link function, the response variable of hunting success or chase (either 0 or 1) and buzz II duration or treatment type as the independent variable. Models 3 and 4 had the response variable of either buzz I or II duration, the independent variables for the categorical predictors of treatment type and bat id and the identity link function. We used the 0.05 criterion for statistical significance and treated within-bat correlation in models 1 and 2 with a random intercept. Within-session correlation in the response variable of models 2–4 was also treated with a random intercept.

Image credits

Irilena Linardaki, Astrid Særmark Uebel, and Stefan Nahnsen for insets in Fig. 2a and b.