How do tiger moths jam bat sonar?

Bats and moths are intertwined in a predator–prey relationship that has spanned at least 50 million years (Miller and Surlykke, 2001). In response to the heavy predation pressure of echolocating bats, many moths have evolved simple ears that alert them to the danger of attack and initiate unpredictable evasive maneuvers (Roeder, 1967). Tiger moths (superfamily Noctuoidea, subfamily Arctiinae) also produce ultrasonic clicks whose defensive functions vary depending on the acoustic and chemical properties of the moth species. Most species produce short bursts of clicks that maximally occupy 1–12% of time with sound (Corcoran et al., 2010). These low-duty-cycle clicks are well suited for advertising the toxic chemistry many tiger moths acquire as caterpillars (Hristov and Conner, 2005a; Hristov and Conner, 2005b; Nishida, 2002; Ratcliffe and Fullard, 2005). After bats learn this aposematic association, they can be misled by palatable tiger moths that mimic their chemically defended relatives (Barber and Conner, 2007; Barber et al., 2009). Moth clicks can also startle bats unaccustomed to sonic prey; however, bats typically habituate to moth clicks after only a few exposures (Bates and Fenton, 1990; Miller, 1991). A minority of tiger moth species produce copious bursts of clicks capable of filling 25–52% of time with broadband noise (Corcoran et al., 2010). We recently demonstrated that one such species – Bertholdia trigona – defends itself by jamming the sonar of bats (Corcoran et al., 2009). This is the only known example of such a defense in nature. It was demonstrated using a behavioral learning paradigm (Barber and Conner, 2007; Hristov and Conner, 2005a), whereby naïve bats (i.e. never having experienced moth clicks) were pitted against clicking moths over several consecutive nights. Against aposematic moths, bats first catch the distasteful moths before learning to avoid them (Barber and Conner, 2007; Hristov and Conner, 2005a). The reverse pattern can be observed for a startle defense – the bats are startled at first, but habituate after only a few exposures to clicking moths (Bates and Fenton, 1990). Only for jamming should the defense be effective throughout the experiment, and this was found to be the case for bats attacking the abundantly clicking and palatable B. trigona (Corcoran et al., 2009).

Three mechanisms have been proposed for how moths jam bat sonar. First, the phantom echo hypothesis states that moth clicks that are sufficiently similar to bat calls could be misperceived by bats as echoes from objects that do not exist (Fullard et al., 1979; Fullard et al., 1994). If perceived as clutter, bats should veer away from phantom objects that occur on the flight path leading to the moth; or, if perceived as another prey item, bats should be observed trying to capture ‘phantom targets’. Second, the ranging interference hypothesis holds that clicks that overlap or closely precede echoes may diminish a bat's precision in determining target distance and, therefore, prevent the bats from properly coordinating their final capture maneuvers (Miller, 1991). Ranging interference has been demonstrated for bats conducting standardized tasks in psychophysical experiments, but only when clicks occur in a 2 ms window preceding returning echoes (Miller, 1991; Masters and Raver, 1996). Finally, the masking hypothesis suggests that if clicks are sufficiently numerous and intense, they may prevent the bat from detecting the moth's echo entirely (Møhl and Surlykke, 1989; Troest and Møhl, 1986). This would make the moth temporarily invisible to the bat. Although other mechanisms of sonar jamming could be proposed, we aimed to test these three hypotheses, which have been the focus of the literature on the subject to date.

The three jamming hypotheses can be differentiated by what the bat perceives: multiple objects surrounding the moth for the phantom echo hypothesis, a blurred target for the ranging interference hypothesis, and no target for the masking hypothesis. Therefore, the bats' flight and echolocation behavior should vary according to each hypothesis. Here we expand upon our previous work demonstrating the existence of a sonar-jamming defense (Corcoran et al., 2009) by analyzing details of bat three-dimensional (3-D) flight and echolocation behavior to address the question of how moths jam bat sonar.

Bertholdia trigona (Grote 1879) were collected using BioQuip (Rancho Dominguez, CA, USA) black lights set over riparian corridors at the Southwestern Research Station near Portal, AZ, in July 2008. Moths were shipped overnight to our lab at Wake Forest University where they were held at 12.5°C until experimentation. Noctuid control moths were collected at lights on the campus of Wake Forest University. Three naïve pre-volant juvenile big brown bats [Eptesicus fuscus (Palisot de Beauvois 1796); J1, J2 and J3] and one adult big brown bat (A1) were taken from a roost in Forsyth County, NC, and held in captivity for the duration of the experiment. Three of the four bats were found to readily eat B. trigona, whereas one did not (Corcoran et al., 2009). To exclude the possibility that bats were responding to clicks as aposematic signals, only bats that ate B. trigona were used in our analysis. Animal care has been described elsewhere (Corcoran et al., 2009) and was conducted in accordance with Wake Forest University's Animal Care and Use Committee (ACUC #A04-188). After each bat could reliably capture silent control moths [female Galleria mellonella (Linnaeus 1758)] off a monofilament line in our anechoic indoor flight room (5.8×4.0×3.0 m), they were put through a seven-night experiment. Each night, 16 moths – four B. trigona, four similarly sized noctuid novelty controls, and eight G. mellonella – were tethered to a monofilament line, one at a time, and presented in random order to an individual bat in the flight room. The line was 60 cm long and attached to the ceiling at a point 1.2 m from each of two walls. Moths flew freely on the tether, eliminating the possibility that bats could memorize the moth's exact location from previous trials. Bats were allowed 1 min or five attack attempts on each moth; however, here we only analyze the first attack attempt on each moth (N=47). Attacks on moths that did not click (N=19) and attacks that were not fully captured on video (N=16) were excluded from our analysis.

All interactions were recorded with two digital high-speed video cameras (Photron FastCam PCI 500, Tokyo, Japan) sampling 250 frames s^–1 directly to the hard drive of a desktop computer. Nine infrared LED arrays (Wildlife Engineering, Tucson, AZ, USA) illuminated the interaction space, as well as a low intensity deep red light for behavioral observation by the experimenter. An ultrasonic microphone (Pettersson D940, Pettersson Elektronik, Uppsala, Sweden; ±8 dB frequency response from 20 to 80 kHz) was placed on the ceiling, facing downwards directly above the tethered moth to record audio of all interactions. The microphone was connected to a laptop computer via a National Instruments (Austin, TX, USA) 6062 PCMCIA A/D card. Audio was sampled at 250 kHz using BatSound Pro v.3.3 (Pettersson Elektronik) and externally triggered to record synchronously with the high-speed video.

Four calibration frames constituting 192 points and a volume of 1.4×1.5×2.1 m were placed in view of both cameras to calibrate the interaction space. All points were digitized from frames taken from each camera using a custom MATLAB program (Natick, MA, USA) (Hedrick, 2008), which fit a set of direct linear transform (DLT) coordinates to the data. These DLT coordinates were then used in a second MATLAB program (Hedrick, 2008) to calculate the 3-D coordinates of objects occurring in view of both cameras. For each interaction, the ‘center-of-object’ of bat and moth were determined for each frame while both animals were visible in each camera.

A third custom MATLAB program (BATracker.m; coded by B. Chadwell) fit a quintic smoothing spline (MATLAB SPAPS routine) to the 3-D coordinates. The bat and moth spline functions were then used to determine flight vectors, distances between bat and moth, and vectors from bat to moth for each point of each interaction. For all time parameters, time zero was taken as the time of bat–moth contact or when the distance between bat and moth was minimized. φ is defined as the angular deviation between the bat's flight vector and the vector from bat to moth (Fig. 1A). To quantify the trajectories of bats completing successful attacks, φ-values were measured from 20 successful bat attacks on noctuid control moths. For each frame of the attack (occurring at 4 ms intervals) from t=–600 ms to t=0 ms, the mean, upper and lower 95% confidence intervals of φ were calculated. For each interaction between a bat and a clicking moth, the time when φ first exceeded the upper 95% confidence limit for control attacks was taken as an indication of when an attack was aborted [time of flight abort (T_Fabort); Fig. 1].

We used a custom version of Sonobat v.2.9 (DnDesign, Arcata, CA, USA) to semi-automatically detect and measure parameters from echolocation calls from bat attacks. The automated processing routines of Sonobat were overseen by an observer to manually confirm the accuracy of all measurements. The time of each echolocation call and the time interval between successive echolocation calls, or pulse interval, were graphed for each attack. Two response variables were measured from each graph – the minimum pulse interval (PI_min) and the time when attack echolocation was aborted (T_Eabort; Fig. 2). T_Eabort was taken as the time of the last echolocation pulse interval of less than 35 ms. If no pulse interval was less than 35 ms, the time of minimum pulse interval was taken instead. Two moth click parameters were measured to test whether they predict bat response: the number of click modulation cycles (Fig. 3) and the time of first clicks (T_click).

To determine whether and how bats modify their echolocation in response to jamming, parameters were measured from four echolocation calls from each interaction: one call preceding the first moth click and three calls following the first moth click. Two parameters were measured from each echolocation call: duration and peak frequency. To calculate these measurements, Sonobat generates a spectrogram of each recording using a 196-point window with a Hanning function. It then creates a time-frequency trace of the first harmonic (or fundamental) of the bat call. The start and end points are defined as the time points when the call intensity decreases below –20 dB relative to the maximum intensity measured in the bat call. All intensity values are measured from points on the spectrogram. Call duration equals the time between the start and end points. Peak frequency was measured from the point of maximum intensity in the spectrogram. This method was used in favor of measuring the peak of a power spectrum generated from the entire call in order to isolate the peak frequency of the fundamental. In previous tests, the power-spectrum method sometimes led to measuring a peak associated with the frequency where the upper portion of the fundamental and the lower portion of the second harmonic cumulatively produced a greater spectral level than that of the lower portion of the fundamental (Fig. 3A). This added unnecessary variation to the analysis. To account for the directionality of bat emissions, the vertical and horizontal angular deviations between bat and moth and bat and microphone were calculated for all attacks at the time when bat calls were measured. The horizontal deviation in angle was 7.6±6.7 deg and the vertical deviation was 23±6.7 deg. These values are within the –6 dB bandwidth of E. fuscus (Ghose and Moss, 2003; Hartley and Suthers, 1989).

To quantify what factors affected bat echolocation and flight responses to moth clicks, we conducted a series of multivariate linear regressions on each of four response variables: minimum bat-moth distance, T_Fabort, PI_min and T_Eabort. Five predictor variables were used for each regression: bat experience (nights), moth click modulation cycle number, T_click, J1 dummy and J2 dummy. Each dummy variable codes the attacks of J1 and J2 separately, allowing us to test and account for individual differences. Predictor variables had low correlations (R²<0.12). For each response variable, we created a single regression model that added all five predictor variables simultaneously. We also analyzed correlation values between all predictor and response variables and compared these results with our multiple regressions to look for interaction effects among predictor variables. This approach avoids the numerous problems associated with stepwise multiple linear regression (Tabachnick and Fidell, 2007).

To determine how bats responded acoustically to moth clicks, we conducted a series of repeated-measures ANOVAs on echolocation call parameters measured from one echolocation call preceding moth clicks and three calls following clicks. In cases where sphericity could not be assumed, the Greenhouse–Geisser correction was applied. Post hoc pairwise comparisons were made between all groups with Bonferroni correction applied to adjust P-values for multiple comparisons. Because of heterogeneity of variance, we used a non-parametric Friedman's ANOVA to test for differences in pulse interval values before and after moth clicks. Non-parametric post hoc tests with adjusted P-values for multiple comparisons were used to test for differences between mean values.

For all tests, a separate statistical analysis was conducted for each category of bat attack (for categories, see Results) and on 20 randomly selected attacks on silent control moths. We selected four calls (one before clicks and three after clicks) from each control attack so that they matched the timing of calls selected from attacks on clicking moths. Measured T_click values were selected at random from attacks on clicking moths and applied to control attacks. For example, if a T_click value of –450 ms was selected, the control attack was processed as though clicks occurred 450 ms before the point of contact. Statistical analysis was conducted in SPSS 16.0 (SPSS, Inc., Chicago, IL, USA); α was set at 0.05 and data are reported as means ± s.d.

To test the plausibility of the three jamming hypotheses (phantom echo, ranging interference and masking), we simulated attacks according to assumptions that follow from each hypothesis. According to the phantom echo hypothesis, bats perceive click bursts not as extrinsic sounds, but as echoes from objects. We assume that the bats would perceive these sounds in the same directional vector as that of the sound source – the moth. Because the interval (325 μs) between moth clicks (Corcoran et al., 2009) is less than the bat's 400 μs temporal resolution (Simmons et al., 1989), clicks within a burst would blur into one continuous sound, and are therefore modeled as one object. Due to the mechanism of sound production, moth click bursts come in pairs. The first click burst (the active half-modulation cycle) decreases in frequency over time, as does the sweep of an echolocation call (Fig. 3). The second burst (the passive half-modulation cycle) changes frequency in the opposite direction. We therefore modeled only the active half-modulation cycles as phantom echoes. To calculate the distance from the bat to each phantom object, the time interval from the beginning of the echolocation call preceding a click burst to the beginning of the click burst was divided by the speed of sound (∼343 m s^–1) and again divided by two for the two-way travel distance. Measurements were made on click bursts from each of five randomly selected attacks from each of the three bats used in the experiment.

Ranging interference occurs only when clicks are present within a 2 ms interval preceding the return of echoes (Masters and Raver, 1996; Miller, 1991). For each attack, we determined how often clicks occurred within this crucial time window. To do this, we first used the 3-D reconstruction of the attack to determine the distance between bat and moth at the time each call was produced. We then calculated the two-way travel time of sound covering this distance and marked whether moth clicks occurred within the 2 ms time interval surrounding the estimated time of the returning echo. Masking requires sound to overlap the short time window of returning echoes (Møhl and Surlykke, 1989). Although the masking window is slightly offset from the ranging interference window, they are of similar durations, and therefore our measurements on ranging interference should roughly apply to the hypothesis of masking.

Bat flight response to clicking moths took three forms: (1) in direct attacks, bats completed flight paths typical of successful attacks on control moths (N=11; Fig. 4A,D) and either captured the moths or made contact after making at least a partial capture attempt; (2) in close-range attacks, bats continued their attacks after clicks and narrowly missed the moth without making a capture attempt (N=24; Fig. 4B,E); and (3) in avoidances, bats aborted attacks soon after hearing clicks and did not make capture attempts (N=14; Fig. 4C,F). The three attack types were distinguished by their φ versus time plots (Fig. 4D–F). In direct attacks, φ-values (the angle between the bat's flight vector and the bat–moth vector) did not exceed the upper 95% confidence limit of control attacks on non-clicking moths; close-range plots showed decreasing or constant φ-values (indicating continued moth pursuit) after the moth clicked, followed by a breach of the upper 95% confidence level of control attacks; avoidances showed immediate increases in φ-values towards the upper 95% confidence level of control attacks after the moths clicked.

Bat echolocation responses (Fig. 4G–I) mirrored the differences in flight responses (Fig. 4D–F). Bats in direct attacks typically completed echolocation sequences with a buzz, or trill of calls with low (∼6 ms) pulse intervals. Buzzes were always present in attacks on control moths (Fig. 5A). Close-range attacks were dominated by short pulse intervals (10–40 ms) that are typically used for bats investigating nearby objects. Buzzes in these attacks were often highly abbreviated (Fig. 4H, Fig. 5B). In avoidances, bats began elongating pulse intervals early in the attack (∼250 ms before minimum bat–moth distance; Fig. 4I, Fig. 5C) and did not buzz.

Bat experience was a consistent predictor of bat flight and echolocation response to moth clicks (Table 1); bats continued attack flight and attack echolocation longer, and flew closer to their prey, as they gained experience. This trend appears to be driven by the higher proportion of avoidances and lower proportion of direct and close-range attacks on the first two nights of the experiment (Fisher's exact test, χ²=6.66, P=0.04; Fig. 6). This suggests that bats were initially startled by the clicks and habituated after approximately two nights. After the habituation period, close-range attacks were the most frequent result (Fig. 6).

The number of click modulation cycles was also important in predicting bat response (Table 1); bats flew closer, used attack echolocation later into the attack and achieved shorter minimum pulse intervals against moths that clicked less. This result appears to be driven by the difference in number of modulation cycles between direct attacks and other attack categories (18.7±16.7 modulation cycles in direct attacks versus 38.0±23.7 modulation cycles for all other attacks; Student's t-test, t=–2.52, d.f.=47, P=0.02). Moths that clicked less were less successful at deterring their predators. The timing of clicks was also found to be important in when bats aborted attack echolocation, but not any of the other three response variables.

Finally, J1 dummy was significant in the T_Fabort model; that is, one juvenile bat aborted flight earlier than the adult bat. J2 dummy was not significant in any model. The relatively small impact of individual dummy variables in our models indicates that the three bats tended to respond similarly to jamming. Together, these models account for approximately one-third of the variation in bat responses to clicking moths. In summary, bat flight and echolocation responses to moth clicks demonstrated a high level of agreement on three results: (1) bats aborted attacks more frequently early in the experiment; (2) bats completed successful attacks more often against moths that clicked less; and (3) after the initial habituation period, the most frequent result was for bats to narrowly miss their prey.

We looked for differences in parameters from echolocation calls made prior to and just after moth clicks to determine the bats' acoustic reaction to jamming (Table 2). As expected, in control attacks, the bats decreased their call durations and, although not statistically significant, a trend towards decreasing pulse intervals and decreasing peak frequencies was observed. In contrast, in close-range attacks and avoidances, bats increased call durations and pulse intervals in response to moth clicks (Table 2; Fig. 5). These results show a reversal of the changes in echolocation emissions typical of attack sequences. No changes in peak frequency were found. Interestingly, the changes of echolocation observed in close-range attacks were short-lived – the bats frequently advanced echolocation towards a buzz a second time as they continued pursuing their prey (Fig. 5B).

To determine the plausibility of each sonar-jamming mechanism, the frequency of clicks occurring at critical time periods was calculated according to assumptions made for each hypothesis. Moths began clicking 379±190 ms before minimum bat–moth distance and produced 33.7±23.6 modulation cycles per attack. Moth clicks occurred in 13.5±5.8 of 16.1±6.8 pulse intervals (83.4%) that occurred after moths began clicking. The locations of phantom echoes relative to actual targets were simulated for 15 randomly selected attacks that occurred after the habituation period. Of the 385 simulated phantom echoes, 307 (79.9%) occurred 25–2624 cm beyond the target; 39 (1.0%), or 2.6 phantom echoes per attack, occurred within 25 cm of the target; and 38 (1.0%), or 2.5 phantom echoes per attack, occurred 25–134 cm in front of the target. These figures are roughly what would be expected by chance, as the time intervals that would place phantom objects between bat and moth or within 25 cm of the moth are much shorter than the time interval that would place phantom objects distant to the moth. Fig. 7A,C and supplementary material Movie 1 show the modeled locations of phantom objects, as well as the measured locations of the bat and moth of an example attack. The locations of phantom objects are randomly distributed spatially and lack spatial continuity over time. If perceived in such a way, phantom objects would flash briefly in disparate locations. Although a small number of simulated phantom objects do occur near the bat's flight path, the bat's flight trajectory does not appear to be influenced by them.

The frequency of clicks occurring in the critical ranging interference window was measured for all attacks. Moth clicks occurred in 11.6±5.5 critical call windows per attack (85.3% of calls with clicks in preceding call intervals). Fig. 7B,C and supplementary material Movie 2 show an individual attack with overlaid ranging errors based on previous interference estimates (Miller, 1991). In this incidence, and in close-range attacks overall, the bat missed its prey by a distance similar to the ranging error predicted by the ranging interference hypothesis (∼15–20 cm) and no capture attempt was made.

Substantial variation was observed in the flight and echolocation responses of bats to moth clicks (Fig. 4). A primary predictor of this variation was bat experience (Table 1), with three of four response variables demonstrating that bats came closer to completing attacks as they gained experience. This pattern suggests that bats were startled by the moth clicks early in the experiments. The bats appeared to habituate to the clicks after approximately two nights (Fig. 6). Bats typically habituate to moth clicks in one to three trials (Bates and Fenton, 1990; Miller, 1991). However, naïve bats attacking tethered, low-duty-cycle moths under nearly identical conditions to that of this study showed little to no startle response (Barber and Conner, 2007; Barber et al., 2009; Hristov and Conner, 2005a).

In the first two nights of our experiments, each bat had up to 40 exposures (four moths per night and up to five attack attempts per trial) to B. trigona moth clicks prior to habituation. Therefore, bats appeared to take much longer to habituate to the high-duty-cycle clicks of B. trigona than the low-duty-cycle clicks of other tiger moths. Bertholdia trigona and other low-duty-cycle moths tested in our experiments have similar peak equivalent sound pressure level values (∼80–85 dB at 5 cm) (Corcoran et al., 2010). The densely produced B. trigona clicks would have a 6 dB increase of power as perceived by the bat because of the number of clicks present within the bat's integration window of 2.4 ms (click measurements based on mean values) (Corcoran et al., 2010; Surlykke and Bojeson, 1996). However, this does not appear to sufficiently explain the observed differences in habituation rates. We suggest that the bats require longer to habituate to B. trigona clicks because of their jamming effect. This is akin to the situation for Catocala moths flashing their colorful hind wings to blue jays; predators take longer to habituate to more conspicuous startle signals (Ingalls, 1993).

After bats habituated to the clicks, they were still largely unsuccessful in capturing moths (30% attack success on nights three to seven; Fig. 6). The few successful attacks were typically on moths that produced few clicks. This supports the notion that only clicks produced in sufficient number are effective at jamming (Corcoran et al., 2009; Corcoran et al., 2010; Hristov and Conner, 2005a; Miller, 1991; Tougaard et al., 1998). The most frequent bat response to jamming after the initial habituation period was a close-range attack (Fig. 6). In these attacks, bats missed their prey by 15.7±7.9 cm and did not make capture attempts. These results are consistent with predictions of the ranging interference hypothesis. The absence of capture behavior suggests that bats were not trying to capture phantom objects, whereas the low percentage of avoidances on nights after the habituation period suggests that bats were not trying to avoid flying into phantom objects. Finally, the proximity of the bats to their prey and the continued use of attack echolocation at relevant times demonstrate that the moth's presence was not completely masked by the moth clicks. We take these results as strong evidence in support of the ranging interference hypothesis.

The 3-D simulations demonstrate how ranging interference may affect an attacking bat (Fig. 7; supplementary material Movie 2). At a distance, the bat is able to gather enough positional information on its prey to direct its flight trajectory appropriately. However, the bat is insufficiently aware of the prey's precise location to coordinate its final capture maneuver. Bats attacking tethered prey exhibit a stereotyped wing beat pattern for the last 400 ms of the attack, demonstrating that bats begin coordinating their capture attempts well in advance of contact (Wilson and Moss, 2004). This time period closely matches the time when moths jam bats. The coordination required to successfully capture prey would likely be obstructed by an imprecise knowledge of the prey's location.

Bertholdia trigona clicks frequently occurred in the time window required for ranging interference to occur. In a two-alternative forced choice experiment involving bats performing tasks from a platform, Miller demonstrated that clicks occurring in as few as 10% of critical time windows were sufficient to degrade the ranging ability of bats (Miller, 1991). We found that during clicking bouts, clicks occurred in 85% of critical windows. When the moths click at full capacity there is rarely a millisecond absent of sound (Fig. 3). This capacity appears well adapted for ensuring that interfering sound occurs in the narrow time intervals when echoes are returning to the bat. Based on spectrogram correlation analysis, B. trigona clicks have been found to be highly disruptive of the acoustic properties of echoes (Corcoran et al., 2010). Also, clicks occurring in such critical windows have been demonstrated to cause latency ambiguity and suppression of neural responses to auditory stimuli in the nuclei of the lateral lemniscus of E. fuscus (Tougaard et al., 1998; Tougaard et al., 2004). These cells are specialized for coding the precise timing of echoes and, therefore, target distance (Covey and Casseday, 1991). In summary, we now have behavioral evidence from bats performing standardized laboratory tasks and bats attacking their natural sonar-jamming prey, acoustic signal processing evidence, and a neurophysiological mechanism, all supporting the ranging interference hypothesis.

Previous research has provided little evidence in support of the phantom echo and masking hypotheses. Behavioral and neurophysiological experiments have generally rejected the phantom echo hypothesis (Miller, 1991; Møhl and Surlykke, 1989; Surlykke and Miller, 1985) and, from an acoustic signal processing perspective, moth clicks are far from similar enough to bat calls to be mistaken by bats as echoes (Corcoran et al., 2010; Surlykke and Miller, 1985). Recent work has demonstrated that bats can distinguish even the subtle differences of calls made by individual bats (Yovel et al., 2009). Therefore, it is difficult to imagine bats mistaking moth clicks, which only have superficial similarities to bat calls, for echoes (Fig. 3).

Masking has previously been rejected as a function for low-duty-cycle moth clicks based on the low probability of a sufficient number of clicks arriving simultaneously with returning echoes (Møhl and Surlykke, 1989). Although B. trigona clicks met this requirement (overlapping 12 echoes per attack on average), the behavioral response of bats in our study demonstrates that complete masking was not achieved. However, we cannot eliminate the possibility that partial masking of echoes has an effect on bat perception. For example, the broadband moth clicks may disrupt spectral notches in echoes that may be indicative of multiple-wave-front echoes, which provide information about the shape of targets (Simmons et al., 1989). These effects would be in addition to the ranging interference that appears to occur.

Bats attacking prey use a stereotyped pattern of changes in echolocation whereby pulse intervals, call durations, call intensities and call frequencies decrease as bats near their prey (Griffin et al., 1960; Kick and Simmons, 1984; Wilson and Moss, 2004). These acoustic alterations are highly adaptive for bats pursuing prey. The rapid emission of pulses late in the attack allows the bats to quickly update their ‘sonar screens’ as to the position of their prey. The decreasing call durations prevent overlap between calls and echoes, which occur closer together as bats near their prey (Wilson and Moss, 2004). The decreased intensity of call emissions keeps returning echoes at a relatively stable level, which may help bats avoid ranging errors caused by time-intensity trading neurons, and allow bats to isolate amplitude changes caused by prey movements (Hartley, 1992; Hiryu et al., 2008). Finally, the lower frequency of terminal buzz calls has the effect of changing the directionality of the emissions from a narrow beam to a wide angle in order to keep the prey in the ensonified volume (Jakobsen and Surlykke, 2010).

The bats in our study reversed this stereotyped pattern by elongating call durations and pulse intervals after hearing moth clicks, resulting in highly abbreviated or absent buzzes (Table 2; Fig. 5). In some cases (avoidances), these changes in echolocation reflected a decision to abandon pursuit of the prey. In others (close-range attacks), the changes in call structure were temporary and the bats continued pursuing the moths with a highly atypical echolocation sequence (Fig. 5B). Others have reported bats elongating pulse intervals in response to moth clicks, albeit for much shorter periods of time than what is reported here (Barber and Conner, 2007; Ratcliffe and Fullard, 2005). These changes do not appear to be in response to the ranging interference caused by the clicks, as bats elongate pulse intervals even when clicks do not overlap echoes (Ratcliffe and Fullard, 2005) and ranging interference alone does not cause bats to change their echolocation (Miller, 1991). Instead, the bats may be compensating for the processing demands of two information streams (Barber et al., 2003).

Mexican free-tailed bats (Tadarida brasiliensis) flying in the laboratory increase the duration, bandwidth and amplitude of sonar emissions in response to broadband noise, an acoustic response typical of many vertebrates (Tressler and Smotherman, 2009). This response is presumably aimed at improving the detection of the signal in noise. The lengthening of call durations that we observed may be an attempt to do the same. Bats also shift call frequency in order to avoid jamming by conspecifics (Gillam et al., 2007; Ratcliffe et al., 2004; Ulanovsky et al., 2004), chorusing insects (Gillam and McCracken, 2007) and synthetic narrowband noise (Bates et al., 2008; Tressler and Smotherman, 2009), but not broadband noise (Bates et al., 2008; Tressler and Smotherman, 2009). We did not observe bats changing the frequency of their emissions in response to moth clicks, only the durations and pulse intervals. The changes in echolocation we observed in close-range attacks appear to be an effort to avoid jamming; however, this comes at the cost of losing the advantages conferred by the specialized attack echolocation that bats more commonly employ. This has the end result of the bats being unable to capture their clicking prey.

The primary objective of this study was to determine the mechanism of the sonar-jamming defense used by high-duty-cycle tiger moths. Our results, combined with evidence from the literature, provide strong support for the ranging interference hypothesis. In addition, the clicks disrupt the bats' stereotyped echolocation attack sequence. The primary requirement for this defense is the ability to produce numerous, densely packed clicks continuously for several hundred milliseconds. By comparing the sounds produced by low-duty-cycle and high-duty-cycle moths (Barber and Conner, 2006; Corcoran et al., 2010), we see that this is accomplished in three ways: (1) the morphological development of many striations on the bilateral tymbal organs in order to produce numerous clicks per modulation cycle; (2) the rapid and repeated behavioral activation of the tymbal organs; and (3) the rhythmic, alternating activation of the two tymbal organs to maximally occupy time with clicks. These observations provide a hypothetical blueprint for the evolution of a sonar-jamming signal. As has been suggested previously (Barber and Conner, 2006; Corcoran et al., 2010), it appears likely that tiger moths originally developed the ability to click in order to advertise their toxic chemistry (Hristov and Conner, 2005a; Hristov and Conner, 2005b), and later other tiger moths took advantage of this association through Batesian and Mullerian mimicry (Barber and Conner, 2007; Barber et al., 2009). A rudimentary form of jamming may have made for a more salient warning cue for acoustic aposematic moths (Ratcliffe and Fullard, 2005), or provided some modest benefit to mimetic animals avoiding capture by potentially discerning predators (Barber et al., 2009). At this point, a clear path was open to the evolution of a jamming signal. Currently, the arctiine phylogeny remains unresolved (Weller et al., 2009). Without this evolutionary framework, we are unable to understand the order in which their various chemical and acoustic defenses were acquired. However, by understanding the mechanism of the sonar-jamming defense, we have demonstrated that in all probability, few barriers existed for tiger moths to adapt their pre-existing clicking defenses for sonar jamming.

 
• 3-D

  three-dimensional

 
• D_BM

  minimum bat–moth distance

 
• DLT

  direct linear transformation

 
• PI_min

  minimum pulse interval

 
• T_click

  time of first moth click

 
• T_Eabort

  time of aborted echolocation attack

 
• T_Fabort

  time of aborted flight

 
• φ

  angle between the bat's flight vector and the bat–moth vector

We are most grateful to Frank Insana, Jeff Paull and the staff of the Southwestern Research Station for assistance collecting B. trigona; Megan Cullen for bat care; Jeff Muday for technical support; and Brad Chadwell for software development. Two anonymous reviewers provided several insightful and constructive comments.