Is physiological stress state reflected in acoustic structure of vocalizations? An experimental test in wild North American red squirrels

Acoustic signaling is an important means by which animals communicate both stable and labile characteristics. Although it is widely appreciated that vocalizations can convey information on labile state, such as fear and aggression, very few studies have experimentally examined the acoustic expression of short-term stress state. The transmission of such information about physiological state could have broad implications, potentially allowing other individuals to modify their behavior or life history traits in response to this public information. North American red squirrels (Tamiasciurus hudsonicus) produce vocalizations known as rattles that advertise territorial ownership. We examined the influence of changes in physiological stress state on rattle acoustic structure through the application of a stressor (trapping and handling the squirrels) and by provisioning squirrels with exogenous glucocorticoids (GCs). We characterized the acoustic structure of rattles emitted by these squirrels by measuring rattle duration, mean frequency, and entropy. Our results provide mixed evidence that rattles show a “stress signature”. When squirrels were trapped and handled, they produced rattles that were longer in duration with a higher frequency and increased entropy. However, squirrels that were administered exogenous GCs had similar rattle duration, frequency, and entropy as squirrels that received control treatments and unmanipulated (unfed) squirrels. Our results indicate that short-term stress does affect the acoustic structure of vocalizations, but elevated circulating GC levels are not solely responsible for such changes.

acoustic structure through the application of a stressor (trapping and handling the squirrels) and 28 by provisioning squirrels with exogenous glucocorticoids (GCs). We characterized the acoustic 29 structure of rattles emitted by these squirrels by measuring rattle duration, mean frequency, and 30 entropy. Our results provide mixed evidence that rattles show a "stress signature". When 31 squirrels were trapped and handled, they produced rattles that were longer in duration with a 32 higher frequency and increased entropy. However, squirrels that were administered exogenous  Vocalizations can also contain information on labile, or short-term, traits, such as short-51 term stress or perhaps the changes in glucocorticoids (GCs) that are released in response to a 52 short-term stressor. Stress is associated with elevated glucocorticoid levels and is known to 53 influence the acoustic structure of vocalizations in a number of species (Manser, 2001 vocalizations tend to be lower in frequency and noisier (highly entropic), and fearful 57 vocalizations tend to be higher in frequency and more tonal (Morton 1977). Although some 58 studies have found empirical support for these rules, others have found inconsistencies. For 59 example, vocalizations associated with fear often fail to consistently conform to these 60 4 motivation-structural rules, and are often highly entropic (Morton 1977; August and Anderson 61 1987). The effects of short-term stress on vocalization structure are thus difficult to generalize. 62 Possible mechanisms by which short-term stress affects the acoustic structure of 63 vocalizations have also been investigated. Short-term stress in known to induce a withdrawal of 64 vagal output from motor neurons in the nucleus ambiguus that causes vocal muscles to tighten, 65 resulting in an increase in vocalization pitch (Porges, 1995). In addition, entropy of vocalizations 66 tends to increase with short-term stress because the oscillation of the laryngeal folds, which 67 govern volume, reaches its maximum amplitude but subglottal pressure continues to increase,  with exogenous GCs, and they found that both types of stress significantly altered vocalization 84 features. Compared to untreated individuals, finches in both treatment groups emitted 85 vocalizations of higher frequency than finches in the control group (Perez et al., 2012). 86 The literature on the influence of stress on vocalizations skews heavily towards group-87 living species and focuses primarily on just a few contexts in which stress occurs; far less is 88 known about the relationship between stress and vocalization structure in solitary species, 89 despite the fact that many regularly produce vocalizations in short term stress inducing situations 90 (Hogstedt, 1982). Furthermore, few studies have experimentally examined this relationship, 91 leaving a gap in our understanding of the mechanism by which stress may influence acoustic 92 structure. We examined how a short-term stress (resulting from trapping and handling) and North American red squirrels (Tamiasciurus hudsonicus). Red squirrels defend discrete 95 territories throughout the year, and produce vocalizations called "rattles" that advertise territorial 96 ownership (Smith, 1968), which deters intruders (Siracusa et al., 2017). At the center of each 97 territory is a "midden," a network of underground tunnels that serves as storage space for white  focal squirrels only differentiated between the rattles of kin and non-kin when the playback 106 6 rattles used were emitted by squirrels that had just been live-trapped and handled (henceforth, 107 "post-trap rattles"), suggesting that post-trap rattles, produced in a high-stress state, are 108 structurally distinct from rattles collected opportunistically (Shonfield et al., 2017). Thus, 109 preliminary evidence suggests that rattles may encode information about stress state.

110
To test this directly, we conducted a two-part study to examine the relationship between 111 stress state and rattle acoustic structure. In the first experiment, we recorded rattles of wild red 112 squirrels after they were live-trapped and handled and compared these to rattles recorded 113 opportunistically, without provocation, from squirrels moving freely around their territories.  To identify if elevated circulating GCs are part of the mechanism by which a short term 118 stressor (such as capture and handling) alters rattle acoustic structure, we conducted a second 119 experiment where we treated squirrels with GCs (dissolved in a small amount of food) and 120 compared their rattles to those of squirrels in a control group (provided with the same amount of 121 food but without GCs) and an unmanipulated group (provided with no food or GCs). A previous 122 study showed that in GC-treated squirrels, plasma GCs rose quickly after treatment and then  be higher in frequency. In the second experiment, we predicted that if GCs are the mechanism by 128 which short term stress alters rattle acoustic structure, rattles emitted shortly after treatment with 129 exogenous GCs would exhibit the same structural distinctions as post-trap rattles when compared 130 with rattles produced prior to treatment and rattles produced by control and unmanipulated 131 squirrels over the same period of time. We expected these structural distinctions to be graded, 132 peaking shortly after treatment and then declining as a function of time since consumption of 133 treatment mirroring the peak and decline of circulating GC levels following treatment.  40-20,000 Hz frequency response (± 2.5 dB); super-cardioid polar pattern). To collect 157 opportunistic rattles, we stood on a squirrel's midden at a distance of no greater than 5 m from 158 the squirrel until it produced a rattle. Red squirrels rattle spontaneously and in response to 159 detection of conspecifics (Smith 1978), but we cannot rule out the possibility that the rattles were 160 elicited by the person recording. A post-trap rattle was the first rattle emitted upon the squirrel's 161 release from a handling bag after spending time in a trap (within a minute of release). We did not 162 record the exact amount of time a squirrel spent inside of a trap, but squirrels were in traps for no 163 more than 120 min before they were released and a rattle was collected. As would be expected,  (Table 1). These rattles were part of a long-term 169 dataset of rattles compiled by prior researchers with the Kluane Red Squirrel Project.  Because we sought to simulate short-term stress induced by a rapid elevation of circulating GC 199 levels, these squirrels were excluded from analysis as well. Our final sample size was GC (n = 200 10), control (n = 12), and unmanipulated (n = 23). 201 We recorded rattles using stationary Zoom H2N Audio Recorders (Zoom Corporation, 202 Tokyo, Japan) that were covered with windscreens and attached to 1.5 m stakes in the center of 203 each squirrel's midden. Because they are not weather-proof, we placed an umbrella 30 cm above 204 each audio recorder to protect it from harsh weather conditions. We set the audio recorders in 205 44.1kHz/16bit WAVE format and recorded in 2-channel surround mode. We deployed the audio 206 recorders between 1700 and 2200 h on the day before treatment so that they would collect "pre- model, which was developed during the same ground-truthing experiment described above, 236 labeled each detection as 'focal rattle,' 'neighbor rattle,' or 'non-rattle,' and assigned a probability 237 that the detection was a focal rattle. Third, we used Kaleidoscope to review spectrograms of all 238 detections labeled 'focal rattle' that have an estimated probability of being a focal rattle of at least 239 0.99999. During this step, we removed any non-rattles that were included erroneously as focal 240 rattles.

241
Our final dataset included 714 rattles from 45 focal squirrels (GC-fed = 232 rattles from 242 10 squirrels, control = 367 rattles from 12 squirrels, and unmanipulated = 115 rattles from 23 243 squirrels). Based on a cross-validated assessment of the accuracy of our approach (see details in 244 Siracusa et al, submitted), 52% of all focal rattles should have been identified correctly as focal 245 rattles (i.e., 48% incorrectly classified as coming from a neighbour, and, therefore, excluded; 246 false negative error rate = 48%), and 6% of the rattles labeled as focal rattles (after manually 247 removing the non-rattles) should actually have been neighbor rattles (i.e., false error rate of 6%).

305
If elevated plasma GCs alter rattle acoustic structure, we expected that the effects of the 306 GC treatment on rattle acoustic structure would be strongest shortly after treatment consumption,  Results: 314 Effects of capture-induced stress on rattle acoustic structure 315 Capture-induced stress caused pronounced differences in rattle acoustic structure: post-316 trap rattles were longer, higher in frequency, and noisier than rattles collected opportunistically.

325
Effects of administration of glucocorticoids on rattle acoustic structure 326 Administration of exogenous GCs did not produce the same effects on rattle acoustic 327 structure as capture-induced stress; the rattle acoustic features of GC treated squirrels did not 328 follow the predicted pattern of peaking after treatment and then declining as a function of time 329 since treatment (Tables S1-S3, Fig. 2). There was, however, a significant linear interaction 330 between treatment and the amount of time elapsed since treatment consumption on rattle 331 duration (F 2, 677.4 = 3.78, P = 0.02). This effect was largely driven by the increases in rattle 332 duration observed in unmanipulated squirrels ( Fig. 2A): rattles from unmanipulated squirrels 333 increased in length throughout the day compared to those treated with GCs (b = 0.33, t = 2.67, P 334 = 0.01, Table S1, Fig. 2A). Rattle durations of squirrels treated with GCs did not change 335 differentially over the course of the day when compared with rattle durations of squirrels fed 336 peanut butter only (b = 0.07, t = 0.73, P = 0.47, Table S1, Fig. 2A).

337
There were no treatment effects on rattle mean frequency (F 2,2 = 0.60, p = 0.63, Table S2) 338 or entropy (F 2,56 = 0.47, p = 0.63, Table S3) and the effects of the treatments on rattle mean 339 frequency or entropy did not depend upon the amount of time that had elapsed since treatment  Discussion: 349 Our study shows that short-term stress, in this case induced by live-capture and handling, 350 significantly influences the acoustic structure of territorial vocalizations in red squirrels.

351
Squirrels experiencing capture-induced stress produced rattles that were longer in duration, 352 higher in frequency, and noisier (higher entropy) than rattles produced by control squirrels. 353 However, we found little support for our second hypothesis that the changes in rattle acoustic 354 structure induced by trapping and handling were caused primarily by increases in circulating 355 GCs, despite the fact that GCs increase in response to trapping and handling. In our second 356 experiment, the rattles of squirrels treated with GCs did not exhibit the expected structural 357 distinctions from the rattles of control or unmanipulated squirrels over the treatment period.

358
The effects of short-term stress (trapping and handling) on rattle acoustic structure that 359 we observed (longer duration, higher mean frequency, and higher entropy) are congruent with 360 such trends in acoustic structure in relation to stress in many species. Chimpanzee screams, for 361 example, increase in duration with the severity of an agonistic encounter (Slocombe et al., 2009).  Because treatment with exogenous GCs did not induce the same changes to vocalization 400 structure as trapping, it is possible that these changes were produced by an effect of trapping 401 besides stress. Because rattles function to advertise territorial ownership, it is possible that a 402 squirrel that has been in a trap and unable to defend its territory for up to two hours, upon 403 release, compensates by producing rattles that are longer and noisier. This hypothesis, however, 404 would require explicit tests.

405
Our findings constitute further evidence that territorial vocalizations such as rattles 406 contain more information than territorial ownership. In red squirrels, not only do rattles have the

413
There are several hypotheses on the functional significance of these tendencies in 414 vocalizations associated with high-stress contexts. In social species, the unpredictability 415 hypothesis states that calls that contain more non-linearities are more difficult to habituate to, 416 and thus noisy alarm calls are more likely to capture the attention of a conspecific in the event of 417 a predatory or otherwise dangerous event (Blumstein and Recapet, 2009). Another hypothesis 418 holds that screams produced when an animal is in imminent danger of predation serve to either 419 startle and distract the predator, or solicit intervention from another animal, either a social group 420 member, or a "pirate" predator that may attempt to steal the prey and unintentionally free it 421 (Hogstedt, 1982). In the case of red squirrels, one hypothesis that can be envisaged is that 422 honestly communicating stress to neighbors may advertise a willingness to aggressively defend 423 one's territory. Another possibility is that instead of honestly depicting a willingness to defend a 424 territory, vocal cues of stress might inadvertently reveal that the caller faces some other 425 challenge and might, therefore, be less capable of defending their territory. These two 426 20 hypotheses, however, would need to be tested directly -for example, a playback study could test 427 whether the rattles of stressed squirrels are more or less likely to deter territorial intrusions from 428 neighboring squirrels than rattles of unstressed squirrels. If stress-influenced rattles are more 429 likely to deter intruders, and if their production predicts an attack or further escalation by the  We thank the Champagne and Aishihik First Nations for providing access to the land on 457 which the study sites for this project were located, in particular Agnes MacDonald and her 458 family for long-term access to her trapline. We also thank Zach Fogel and Noah Israel, the 459 diligent field technicians whose work on the GC-induced experiment was crucial for its success.

715
Squirrels fed supplemental food, exogenous GCs, or those that were unmanipulated had similar acoustic structure 716 except that unmanipulated squirrels had significantly longer rattles than GC-treated squirrels as the time since 717 treatment consumption increased (Table 1B) .