Report

Bats perceptually weight prey cues across sensory systems when hunting in noise

See allHide authors and affiliations

Science  16 Sep 2016:
Vol. 353, Issue 6305, pp. 1277-1280
DOI: 10.1126/science.aaf7934

Abstract

Anthropogenic noise can interfere with environmental information processing and thereby reduce survival and reproduction. Receivers of signals and cues in particular depend on perceptual strategies to adjust to noisy conditions. We found that predators that hunt using prey sounds can reduce the negative impact of noise by making use of prey cues conveyed through additional sensory systems. In the presence of masking noise, but not in its absence, frog-eating bats preferred and were faster in attacking a robotic frog emitting multiple sensory cues. The behavioral changes induced by masking noise were accompanied by an increase in active localization through echolocation. Our findings help to reveal how animals can adapt to anthropogenic noise and have implications for the role of sensory ecology in driving species interactions.

Anthropogenic noise is a globally rising environmental pollutant that has been linked to lower survival and reduced reproductive success of many animal taxa (13). Noise can mask environmental cues, making it difficult to hear moving prey or approaching predators, and can interfere with the perception of acoustic communication signals (36). Signal producers may be able to reduce the masking impact—for example, by calling louder (79)—but such signaling strategy is unavailable to receivers. Some receivers can depend on perceptual strategies to maintain cue detection and thereby adapt to noisy environments (10, 11).

Predators such as bats and owls are highly specialized to hunt prey by ear (12); thus, noise that masks prey sounds severely hampers their foraging success (4, 5). However, predators may be able to adapt to masking levels of anthropogenic noise by actively shifting their attention or emphasis placed on processing cues from different sensory modalities from the same prey (1316). We refer to this as cross-modal perceptual weighting (17).

We studied the effect of masking noise on the attack behavior of the fringe-lipped bat (Trachops cirrhosus), a neotropical species that is specialized to find frogs by eavesdropping on their mating displays (18, 19). Bats can passively locate their prey using only prey-generated sounds, but their performance is severely hampered when exposed to noise (20). A male frog, however, provides additional, multimodal cues to hunting bats, as it inflates and deflates a vocal sac while calling (14, 16). Bats can detect the frog’s vocal sac with their echolocation (i.e., processing the ultrasonic echoes that return from their prey), but only when the sac is dynamically inflated; bats do not detect a static vocal sac (21). Echolocation can provide highly accurate spatial information about target stimuli, and we therefore expected that these bats are capable of adapting to masking interference through a change in their perceptual weighting of sonic cues versus ultrasonic cues (22).

We used robotic frogs that emitted either multimodal cues (sound plus moving vocal sac) or control cues (sound plus static vocal sac) (Fig. 1A). Individual bats were given a choice to attack one of two models under three different noise treatments: (i) a masking noise overlapping the main frequency range of the frog call, (ii) a nonmasking noise, and (iii) an ambient control noise condition (Fig. 1B and supplementary materials). Although the signal-to-noise ratio of the frog call was strongly reduced during masking noise, the signal remained audible to the perched bat (Fig. 1B). We predicted that bats rely more on echolocation when presented with masking noise and would consequently make more attacks on the multimodal frog model and alter their echolocation behavior [see scenarios in Fig. 1, C and D, or (23)]. Objects surrounding a target also return echoes, and this so-called background clutter is known to interfere with detection and processing of echolocation target cues (4, 24). We thus tested bats additionally on their attack behavior when dried leaves were added around the frog models (clutter treatment; fig. S1).

Fig. 1 Perceptual strategies to deal with prey signal masking.

(A) Graphic representation of our experimental design. Bats can passively listen to frog sounds (channel 1) broadcast from speakers underneath the robofrog models; they can also actively use their echolocation (channel 2) to detect the dynamic vocal sac. (B) Bats were tested under nonmasking noise, masking noise, and control conditions (no noise) broadcast from a speaker placed above the frog models. Shown are two spectrograms of a frog call with frequency regions of noise treatments superimposed on it (4.0 to 8.0 kHz, 0.1 to 4.0 kHz). (C) Bats can rely on passive listening to their prey’s mating sound (channel 1) as well as on active listening by processing multiple echoes returning from the frog’s moving vocal sac (channel 2). Shown are the typical frog call amplitude profile (channel 1), the inflation and deflation of the frog’s vocal sac (channel 2, gray symbol), and bat calls (red symbols) and echoes (blue symbols) overlapping as well as nonoverlapping in time with the vocal sac cue. (D) When noise masks the prey call, bats may increase their echolocation effort (scenario 1) or alter call design (scenario 2) to maintain target localization.

We trained 12 wild-caught bats to attack our robotic models in an outdoor flightcage (21). Bats always started their attack flight toward one of the frog models from their perch and only in response to stimulus playback. Attack latency was strongly influenced by our noise treatment (generalized linear mixed models, Nbats = 12, Ntrials = 432, χ2 = 22.07, P < 0.001; Fig. 2A and table S1). Post hoc independent contrasts revealed that bats were slower in making their attacks under masking noise relative to nonmasking noise (z value = 5.57, P < 0.001) and ambient conditions (z value = 7.56, P < 0.001). We did not find a significant effect of the clutter treatment on attack latencies (χ2 = 0.09, P = 0.76). On the other hand, the time between leaving the perch and attack on the model (hereafter, flight duration) increased during the clutter treatment (χ2 = 17.04, P < 0.001; Fig. 2B) but was not affected by the noise treatment (χ2 = 0.24, P = 0.89). The number of attacks on the multimodal frog model relative to the control frog model was significantly affected by the noise treatment (χ2 = 7.63, P = 0.022; Fig. 2C). Post hoc binomial tests revealed that bats had a clear preference for the frog model displaying multimodal cues under masking noise (z value = 2.34, P = 0.019), but not under nonmasking noise (z value = –0.94, P = 0.35) or ambient conditions (z value = 1.17, P = 0.24). Clutter treatment had no significant effect on the probability of bat attack on either model (χ2 = 0.60, P = 0.43; Fig. 2C), nor did we find any significant interaction effects between noise and clutter treatment (all response variables P > 0.5).

Fig. 2 Masking noise alters bat attack and echolocation behavior.

(A) Attack latency (natural log-transformed scale on the y axes) was significantly affected by noise treatment. Bats took longer to start their attack from the perch when the frog call was masked by noise. (B) Flight duration significantly increased during clutter treatment. Noise treatment had no effect on flight duration. (C) Bats made proportionally more attacks on the multimodal robofrog model relative to the control model, but only during masking noise treatment. The y axis depicts proportion of attacks made on the multimodal model; the horizontal dashed line indicates chance level (probability of 0.5); ns, not significant. (D) Bats increased call rates during masking noise. Box plots depict distribution of latency and duration across all trials [(A) and (B)], the proportion of attacks averaged over noise and clutter treatment groups (C), and the number of calls emitted in a 1-s time frame based on model estimates (D). Red and blue box plots indicate clutter treatment and no-clutter treatment trials, respectively.

We obtained ultrasonic recordings for a subset of six individuals and analyzed all calls made on the perch between stimulus onset and start of the attack flight. Bats made on average 7.69 ± 2.78 calls on their perch, the majority (77%) during the last second. We selected three calls from a 1-s portion of the recording (shortly before bats had taken flight) to test for an effect of experimental treatment on echolocation behavior. We found the number of calls produced at the perch, as well as the call rate, to be significantly affected by noise treatment (number of perch calls, Nbats = 6, Ntrials = 146, χ2 = 6.44, P = 0.039; call rate during last second, χ2 = 7.78, P = 0.022; Fig. 2D). Bats increased their use of echolocation during masking noise relative to nonmasking noise (5 of 6 bats on average increased call rate; z value = 2.34, P = 0.032) and relative to ambient noise conditions (6 of 6 bats increased call rate; z value = 3.46, P < 0.001). We did not find any differences in call peak frequency (χ2 = 0.27, P = 0.87) or call duration (χ2 = 2.85, P = 0.24) between noise treatment groups. Clutter treatment had no significant effect on the number of calls emitted from the perch (χ2 = 1.53, P = 0.22) or on call rate (χ2 = 1.26, P = 0.26), call peak frequency (χ2 = 1.53, P = 0.22), or call duration (χ2 = 0.84, P = 0.36).

Masking noise increased attack latencies during our experiment, but bats could reduce this effect when using multimodal cues. We reanalyzed the attack latency data and added robofrog choice (control or multimodal) to our statistical model as an additional factor. We found a significant interaction effect between noise treatment and robofrog choice (Nbats = 12, Ntrials = 432, df = 2, χ2 = 11.82, P = 0.003; Fig. 3 and table S1). Bats were faster in attacking the multimodal model relative to the control model, but only under masking noise levels (z value = –3.78, P < 0.001). Robofrog choice had no effect on attack latencies under nonmasking noise (z value = 0.86, P = 0.78) and ambient noise conditions (z value = –0.26, P = 0.99).

Fig. 3 Multimodal cues reduce attack latency under masking noise levels.

Attack latency showed a significant interaction between noise treatment and multimodal cue use. Bats were faster in attacking the multimodal model relative to the control model only during masking noise. Box plots depict model estimates per noise treatment group and attack choice (on either the control or the multimodal model). Gray lines depict fitted slopes per individual per noise treatment. ***P < 0.001.

Anthropogenic noise can affect predator-prey dynamics through masking of acoustic cues or by distracting or disturbing individuals (2, 25, 26). Our results confirm a masking impact of noise on bat attack latencies, thereby giving frogs more time to escape predation. More important, the results show that bats can actively compensate the masking impact by making more use of cues available to them in an additional, less noisy sensory channel. We also found a factor of 2 increase in attack preferences on the multimodal versus the unimodal model, which suggests that noise can drive selection pressures acting on sexual signals.

A previous study on a bat species that echolocates silent prey showed that the negative impact of noise on hunting success is enhanced when individuals are tested in a highly cluttered environment (4). Clutter treatment in our experiment affected flight duration but surprisingly had no effect on attack choice or latency. Fringe-lipped bats can detect and localize the frog’s vocal sac at distances up to 6 m from their perch (21, 22), and it is likely that its movement allows bats to discriminate target echoes from the stationary background (27), such as the dried leaves we placed around our models during clutter treatment.

In conclusion, we showed that bats preferred multimodal displays to unimodal displays, but only under masking noise conditions. Such cross-modal perceptual weighting reduces the masking impact of noise and could be a general receiver strategy (11, 28). A shift in the use of signals and cues across sensory systems will also alter selection pressures acting on sexual displays (29, 30). Thus, in noisy human-impacted areas such as in cities or along highways, we would expect to find a change in the multimodal content of communication signals (11). We may also expect a shift in species composition in noisier areas based on perceptual as well as communicative traits. Species that can rapidly alter their perceptual mechanisms will likely do better in noise-impacted areas, and this in turn has consequences for their predator and prey species that emit different signals and cues. Human-induced changes to the sensory ecology of particular habitats can thus be an important factor in driving species interactions and ultimately determining community assemblages.

SUPPLEMENTARY MATERIALS

www.sciencemag.org/content/353/6305/1277/suppl/DC1

Materials and Methods

Fig. S1

Table S1

References (3133)

REFERENCES AND NOTES

Acknowledgments: For support in the field, we thank the Gamboa Bat Lab, specifically L. F. Gómez-Feuillet. We thank B. Klein, P. Clements, and Moey Inc. for fabricating the pneumatic robotic frog system, and J. Ellers and H. Goerlitz for comments that substantially improved the manuscript. Supported by a Smithsonian fellowship (W.H.), NSF grant IOS 1120031 (R.C.T., M.J.R., and R.A.P.), and the Smithsonian Tropical Research Institute (R.A.P.). All research reported here complied with STRI IACUC protocols (2015-0209-2018; 2014-0101-2017). We obtained all required permits from the Government of Panama (SE/A-86-14). The authors report no conflict of interest. Raw data are available at the Dryad Data Repository (dx.doi:10.5061/dryad.5gk8j).
View Abstract

Navigate This Article