Social foraging in vampire bats is predicted by long-term cooperative relationships

Stable social bonds in group-living animals can provide greater access to food. A striking example is that female vampire bats often regurgitate blood to socially bonded kin and nonkin that failed in their nightly hunt. Food-sharing relationships form via preferred associations and social grooming within roosts. However, it remains unclear whether these cooperative relationships extend beyond the roost. To evaluate if long-term cooperative relationships in vampire bats play a role in foraging, we tested if foraging encounters measured by proximity sensors could be explained by wild roosting proximity, kinship, or rates of co-feeding, social grooming, and food sharing during 21 months in captivity. We assessed evidence for 6 hypothetical scenarios of social foraging, ranging from individual to collective hunting. We found that closely bonded female vampire bats departed their roost separately, but often reunited far outside the roost. Repeating foraging encounters were predicted by within-roost association and histories of cooperation in captivity, even when accounting for kinship. Foraging bats demonstrated both affiliative and competitive interactions with different social calls linked to each interaction type. We suggest that social foraging could have implications for social evolution if “local” within-roost cooperation and “global” outside-roost competition enhances fitness interdependence between frequent roostmates.

Introduction frequent roostmates. If so, within-roost and outside-roost networks should be negatively 93 correlated. Finally, if entire roosting groups also forage together, then we expect highly 94 correlated within-roost and outside-roost networks. 95 96 To evaluate evidence for these scenarios, we tested whether nightly foraging departures and 97 encounters were predicted by: kinship, roosting associations based on two levels of proximity 98 (during the previous day or over the whole study), and rates of social grooming, food sharing, 99 and co-feeding in captivity. To document roosting associations and foraging encounters, we 100 analyzed social encounter data from proximity sensors placed on 50 free-ranging vampire 101 bats. As additional predictors for 23 of these bats, we used unpublished data on captive co-102 feeding rates and published long-term rates of social grooming and food sharing (Ripperger 103 et al., 2019). Using simultaneous ultrasonic recording and infrared video, we also describe a 104 distinct new type of vampire bat call only observed during hunting interactions. Our findings 105 illustrate how within-roost cooperative relationships influence foraging in vampire bats and 106 how social networks can vary across contexts. 107

119
Subjects 120 Subjects were common vampire bats (Desmodus rotundus) including 27 wild-caught adult 121 females that were tagged and released, and 23 previously captive females (17 adults and 122 their six subadult captive-born daughters) that had spent the past 22 months in captivity and 123 were then tagged and released back into their wild roost tree (see Carter et al., 2020; 124 Ripperger et al., 2019). See supplement for details. 125 126 Kinship 127 We assumed that known mother-daughter pairs had a kinship of 0.5. To estimate kinship for 128 all other pairs, we genotyped bats at 17 polymorphic microsatellite loci (DNA isolated via a 129 salt-chloroform procedure from 3-4 mm biopsy punch stored in 80 or 95% ethanol), then 130 used the Wang estimator in the R package 'related'. See supplement for details. 131 132 Past cooperative interaction rates in previously captive bats 133 To measure cooperative relationships in the previously captive bats, we used previously 134 published rates of social grooming and food sharing from experimental fasting trials (Carter 135 et al., 2020). See supplement for details. To assess tolerance while feeding, we also 136 analysed previously unpublished data on co-feeding among the same captive vampire bats. 137 Social interactions were observed at blood spout feeders while the bats were in captivity, 138 including 1300 competitive interactions and 277 cases of co-feeding where two bats were 139 observed feeding from the same blood spout at the same time (from 1050 h of observation 140 from 70 nights). We used 201 co-feeding events with identified bats to construct a co-feeding 141 network of the number of dyadic co-feeding events (range = 0 to 6) for each pair.

143
To assess correlations between the captive co-feeding network and networks of food sharing 144 or social grooming, we used Mantel tests. To test the same correlation while controlling for 145 overlap in individual feeding times, we also used a custom double permutation test (Farine 146 and Carter, 2020). This procedure calculates an adjusted co-feeding rate for each pair as the 147 difference between the observed co-feeding rate and the median expected co-feeding rate 148 from 5000 permutations of the co-feeding bat identities, permuted among the bats seen 149 within each hour. The results of this constrained permutation test and the unconstrained 150 Mantel test were similar and gave the same conclusion, so we report only the results from 151 the double permutation test. To test for preferred captive co-feeding partners, we also used 152 the same within-hour permutations to test if social differentiation in co-feeding (the coefficient 153 of variation in co-feeding rates) was greater than expected from the null model.

155
Association rates in the wild using proximity sensors 156 We placed custom-made proximity sensors on all 50 female common vampire bats (sensor 157 mass: 1.8 g; 4.5-6.9 % of each bat's mass) that automatically documented dyadic 158 associations among all 50 tagged bats when those come within reception range (max. 5-10 159 m). To log encounters, each proximity sensor broadcasted a signal every two seconds to 160 update the duration of each encounter. We used 1 s as the duration of encounters that were 161 shorter than two successive signals (i.e., encounters shorter than two seconds). We collected association data on the free-ranging bats at Tolé, Panama (8°12'03"N 167 81°43'46"W), a rural area that is mainly composed of cattle pastures for meat production.

168
Around 200-250 common vampire bats roosted inside a hollow tree on a cattle pasture that 169 was about 15 ha in size. To create a stable food patch, we corralled ca. 100 heads of cattle 170 at a distance of ca. 300 m from the roost from 6pm until 6am between the evening of 171 September 21 until the morning of September 26, 2017 (days 1 to 5 in our study). Before and 172 after that time period, the cattle were ranging freely. A neighboring, much larger pasture west 173 of the roost had about 1,500 heads of cattle within a distance of 1-2 km ( Figure S1).

175
To construct networks of roosting association rates during each daytime period within the 176 roost, we relied on roosting association data that had been used in a previous study 177 (Ripperger et al., 2019). Based on the same two thresholds of signal strength as before, we 178 defined two categories of proximity: "associations" (within a maximum of ca. 50 cm) and 179 "close contacts" (within ca. 2 cm). Roosting network edges were rates of within-roost 180 association or close contact, i.e. the total time two bats spent in association per unit of time.

181
See supplement for details.

183
To log presence and co-occurrence of foraging bats at points outside the roost during the pasture ( Figure S1). We used the same kind of data to find the return times to the roost for 191 each bat and night.

193
Of the 629 dyadic encounters that occurred one minute after leaving the roost and one 194 minute before arriving at the roost, we excluded 43 encounters from further analysis, 195 because a proximity sensor contacted the roost base station, suggesting that those 196 encounters occurred while bats were roosting at the entrance or on the outside of the roost 197 tree. The remaining 586 encounters occurred farther away, outside the communication range 198 of the roost base station, and we refer to these as "foraging encounters".

200
Observing interactions of foraging vampire bats 201 At Tolé, we only observed two occasions, where two bats stopped at the cattle pasture and 202 were associated (for 3.5 and 4.6 minutes). When releasing the corralled cattle in the 203 morning we observed bite marks but to avoid changing their behavior, we did not get close 204 enough to the cattle at night to record audio or video of bats interacting. To collect direct 205 observations on foraging behavior, we recorded simultaneous audio and video of bat 206 foraging behavior at a different farm near La Chorrera, Panama (8°52'42"N 79°52'05"W) 207 using an infrared (IR) spotlight, IR-sensitive video camera (Sony AX53 4K camcorder) and a 208 Avisoft condenser microphone (CM16, frequency range 1 to 200 kHz) and digitizer (Avisoft 209 USG 116Hbm, 1000 kHz sampling rate, 16-bit resolution) connected to a notebook 210 computer. One observer (SPR) moved with a herd of about 20 grazing cattle without visual 211 light and observed the moving cattle through the viewfinder of the IR camera. To compare 212 social calls made during foraging with calls from inside a roost, we used the same recording 213 equipment to record social calls from a roost only a few hundred meters from the foraging 214 site.

216
Acoustic analysis of calls in foraging bats 217 We used Avisoft SASLab Pro (R. Specht, Avisoft Bioacoustics, Glienicke/Nordbahn, 218 Germany; version 5.2.13) to measure acoustic parameters of the social call types. Start and 219 end of calls were determined manually, based on the oscillogram. Subsequent, five acoustic 220 parameters were measured automatically; one temporal (duration) and four spectral 221 parameters (peak frequency at maximum amplitude, minimum and maximum frequency, and 222 bandwidth). Acoustic parameter extraction was restricted to the fundamental frequency. For every bat, we calculated the times over 9 days when it was clearly distant from the roost 234 tree (ESM File 2). To test whether the previously captive bats and never-captive control bats 235 differed in the departure time and duration of their foraging bouts, we fit linear mixed-effect 236 models (LMMs) with type of bats and day as fixed effects and bat as a random effect (p-237 values estimated with Satterthwaite's degrees of freedom method using the R package 238 lmerTest). To compare consistency of onsets and durations, we measure the unadjusted 239 repeatability (ICC or intra-class correlation) for each type of bat. To see how often tagged 240 bats departed together, we inspected cases where departure times were within one minute.

242
Preferred associations during foraging 243 To test if repeated foraging encounters occurred among the same bats more than expected 244 by chance, we used a custom data permutation test that compared observed and expected 245 social differentiation (the coefficient of variation in co-foraging rates, which increases when 246 some pairs have more repeated encounters) while controlling for overlap in foraging times. 247 Since all bats were sampled evenly within each night and most foraging encounters were 248 brief (median = 1 second), we first used a simple and conservative measure of co-foraging 249 rates based on counting the presence or absence of an encounter during each hour outside 250 the roost over 9 days (counts varied from 0 to 15). For instance, if two bats met twice in the 251 same hour bin, this is still one encounter. These binary observations could be swapped in 252 our null model. To generate a null distribution of 5000 social differentiations expected by 253 chance, we permuted one bat in every dyad to a random possible partner that was also 254 present outside the roost in that same day and hour (to control for overlaps in foraging 255 times).

257
Predictors of social foraging 258 To test predictors of social foraging, we constructed foraging encounter networks where 259 edges were based on either duration of total encounter time outside the roost (seconds) or 260 number of days with foraging encounters (0-9). The latter response variable is far more 261 conservative because it only counts repeats across different days. We included the following 262 predictors: kinship, two proximity levels of within-roost association, social-grooming rate, 263 food-sharing rate. We also tested the effect of dyad type (i.e. both bats previously captive, 264 both bats never captive, one bat previously captive, and both bats captive-born juveniles). 265 We did not use number of nights with foraging encounters as a response for tests that only 266 included the previously captive bats, because 9 of them (including all captive-born bats) left 267 the roost during the study period (Ripperger et al., 2019). To measure how much longer 268 foraging encounters were between kin versus nonkin, we fit a linear mixed effects model with 269 log-transformed duration as the response variable, kinship greater than 0.1 as a binary fixed 270 effect, and both bats' identities as random effects, then converted model coefficients into a 271 percentage difference.

273
To test the effect of predictor networks on a response network, we used regression quadratic 274 assignment procedure (QAP) for single predictors, or multiple regression quadratic 275 assignment procedure with double semi-partialling (MRQAP) for two predictors (using the 276 'asnipe' R package (Farine, 2013)). To create null models, we used constrained (within-day) 277 node-label permutations. This approach is necessary for preserving the daily and nightly 278 network structure (e.g. distribution of edges and edge weights) and for controlling for the 279 presence or absence of bats in the roost each day. To control for foraging bout overlap in 280 each pair, we included that measure as a covariate. We also used QAP to test whether the 281 within-roosting association on each day, predicted the subsequent foraging network that 282 night. We then bootstrapped the mean of the slopes across the eight days to test for an 283 overall paired day-night effect. 284

Consistency of individual social traits 285
To test whether bats that are more socially connected within the roost are also more 286 connected in foraging networks, we tested if the nodes' degree centrality was correlated 287 between roosting and foraging networks. We measured degree centrality independently 288 within each day or night network when the bat was present and then took the mean for each 289 bat. Bats with no encounters in that day or night were considered missing for that day (i.e. 290 not counted as zero degree). We fit general linear mixed effect models with foraging network 291 centrality as the response variable, roosting network centrality (either association and close 292 contact) as fixed effect, and bat as random effect. P-values were calculated from 5000 293 permutations of the bat's foraging centralities within each night. These constrained node-294 label permutations (within night) are necessary to control for the fact that foraging and 295 roosting network centralities could be correlated simply by some bats being present at the 296 site longer. between day and night on day 4 (Table S1). 307 308

310
Sampled bats did not depart together 311 The never-captive control bats departed from the roost 8.3 hours after sunset and returned 312 2.5 h later, on average (ESM 2). The previously captive bats foraged earlier and less 313 predictably (see below). We observed only five cases where two bats departed within five 314 seconds of each other and none of these cases was followed by a foraging encounter. For 315 the cases where pairs did have a foraging encounter, the shortest differences in departure 316 times were 8, 21, and 28 s.

318
Previous captivity influenced departures and foraging 319 Compared to the wild control bats, the previously captive bats departed the roost on average 320 1.6 hours earlier (t = -4.55, p<0.0001), but they did not forage consistently longer (LMM; 321 t=1.29, p=0.2, ESM 2). The captive-born bats departed 2 hours earlier (t = -3.15, p = 0.002) 322 and also did not forage longer (t = -0.41, df = 47.8, p = 0.7) than control bats. All these 323 models control for departure times being on average 14 minutes later each day (t = 6.6, p < 324 0.0001, ESM 2), perhaps due to moonset times being 20-40 min later each day during the 325 study period. The total duration of foraging encounters did not clearly differ between types of 326 pairs ( Figure S3A), but pairs of control bats had significantly more nights with foraging 327 encounters ( Figure S3B) compared to other types of pairs, possibly due to control bats 328 having more consistent foraging times. Departure times were more consistent across days 329 within each control bat (intraclass correlation coefficient (ICC) = 0.58) compared to within 330 each previously captive bat (ICC = 0.21) or captive-born bat (ICC = 0). The duration of the 331 longest foraging bout was also more consistent in wild control bats (ICC = 0.54) than the 332 previously captive bats (ICC = 0.35) or captive-born bats (ICC = 0.15).

334 Preferred associations in foraging encounter networks 335
Foraging encounters were orders of magnitude shorter in duration than within-roost 336 encounters, their median duration was 1 s, and they never exceeded 30 minutes (ESM 1, 337 Figure S2). Of 151 pairs with a foraging encounter, 45 did this repeatedly across 9 nights. 338 The variation in number of hours in which two bats reunited was greater than expected from 339 our null model that simulated random encounters among bats that were outside the roost in 340 the same hour (observed coefficient of variation = 4.36; p < 0.001; 95% of expected values: -341 2.2 to 2.4). Most of these foraging encounters occurred at locations outside our sampled 342 areas, but 10 (among eight pairs) occurred near the other base stations on the surrounding 343 cattle pastures (ESM 1, Figure S1), and only three foraging encounters (among three pairs) 344 occurred at the corral that we created as a stable food patch about 300 m from the roost (two 345 encounters on days one and three while the cattle were present and one encounter on day 346 seven). encounter for close kin (r > 0.1) was 9 s, which was 135% longer in duration relative to the 353 duration of foraging encounters between nonkin (r < 0.1; median duration = 1s; β =0.85, 354 df=175, p=0.001).

356
Within-roost association rates predicted foraging encounters 357 Bats that spent more time near each other within the tree during the day, also spent more  (Table S1), 365 but the paired relationships between day and night networks tended to be greater than zero 366 overall (associations: mean β = 0.026, 95% CI = 0.004 to 0.051; close-contact associations: 367 mean β = 0.018, 95% CI = 0.003 to 0.04).

374
Cooperative relationships in captivity predict foraging encounters in the field 375 In the previously captive bats, foraging encounter time was predicted by social grooming independent predictors of social foraging. 381 382

Co-feeding among familiar captive bats was not limited to cooperative relationships. 383
In contrast to the measures of social foraging in the field, we detected only weak evidence for 384 preferred associations during co-feeding in captivity (social differentiation = 2.10, p = 0.047 385 when controlling for hour, p = 0.041 when not controlling for hour), and we found no 386 correlation between captive co-feeding and social grooming (r=0.008, p=0.36), food sharing 387 (r=0.015, p=0.28) or social foraging time in the wild (r=0.003, n=20, p=0.42).

389
Behavioral interactions during foraging 390 To record a sample of bat interactions during foraging encounters, we recorded infrared 391 video and ultrasonic audio of 14 interactions between foraging vampire bats (Tables S2 and  392 S3). Social calls during foraging had three general spectral shapes ( Figures S5 and S6 calls" were noisy without clear tonal structure and occurred during antagonistic interactions.

396
We observed "n-shaped calls" produced by bats interacting while near cattle ( Figure S6), and 397 to our knowledge this call type has never been seen in wild roosts, from confrontations at the 398 feeders in captivity (Sailler and Schmidt, 1978), or from individually isolated bats in captivity 399 (Carter et al., 2012, Carter, unpublished data). 400 401 402 Discussion 403 404 Long-term cooperative relationships predicted repeated foraging encounters 405 Tagged female vampire bats departed the roost individually, but often re-united far from the 406 roost during foraging bouts. The rates of these foraging encounters were consistently higher 407 than expected in specific pairs and predicted by roosting associations, kinship, and by the 408 history of social grooming and food sharing in captivity, even when controlling kinship.

409
Previous experiments with female vampire bats suggest that these measures-roosting Wilkinson , 1985). In this study, we knew the cooperation histories among the previously 413 captive bats, and that these individuals had no interactions with the control bats for at least 414 the previous 22 months. We could therefore infer that relationships that are typically defined 415 by associations and cooperative interactions within roosts, also extend beyond the roost and 416 may provide benefits during foraging. In addition to consistent social relationships across 417 context (from captivity to roosting to foraging), we found that bats that encountered more 418 associates in the roost during the days also encountered more associates while foraging 419 during the nights, suggesting consistent individual variation in social traits. 420 Although some foraging encounters may have occurred before or after foraging, most of 421 these encounters were likely to have occurred during foraging for several reasons. First, 422 foraging encounters were brief, whereas associations among non-moving bats should be 423 much longer in duration ( Figure S2). Second, foraging is likely to take up a substantial 424 amount of the limited time outside the roost (mean = 2.4 h). After commuting, searching, and 425 selecting a host, a vampire bat can take up to 30 minutes to select a wound site, 10-40 426 minutes to prepare the wound site, and 9-40 minutes to feed (Greenhall, 1972;Greenhall et 427 al., 1971). Third, we used infrared video to observe several interactions on or near cattle that 428 were consistent with the short durations of foraging encounters in the proximity data (e.g., 429 Videos S1, S2, S4). Fourth, foraging encounters among close female kin had a median 430 duration of 9 s and were longer than among non-kin (median duration of 1s), which is 431 consistent with observations that affiliative interactions last longer.

433
No clear evidence for highly coordinated collective movements 434 For animals with fluid social structures (e.g. high fission-fusion dynamics), it is important to 435 clarify the ambiguous meaning of a "social group", and similarly, one must distinguish 436 between different possible forms of "social foraging" (Lang and Farine, 2017). In bats, the 437 relative degree of social coordination during foraging can be difficult to assess and compare 438 due to differing limitations in the observational methods and the lack of knowledge of 439 differentiated social relationships within the colony. In this study, we took advantage of well 440 described within-roost relationships to assess evidence for several alternative scenarios of 441 foraging behavior (Figure 1). Kinship and rates of association and cooperation led to longer 442 and more frequent foraging encounters, but we did not observe highly coordinated joint 443 departures or collective movements (Figure 1). This fluid pattern, of not moving in 444 coordinated stable groups yet repeatedly encountering preferred associates during foraging, 445 is also reflected in co-roosting networks where individuals form roosting groups that 446 frequently change composition, yet maintain preferred relationships over time (Wilkinson, 447 1985). Given the many unsampled bats inside the same tree (~200), it is possible that bats 448 departed with other unobserved roostmates, but we did not see departures of large groups 449 (while catching bats outside the roost) nor did we see evidence for coordination between 450 roosting and departing in the tagged bats.

452
The ways that specific bats reunited with preferred associates therefore remains unknown, 453 but the downward sweeping calls that we recorded in foraging bats ( Figure S6), are similar to 454 contact calls that captive vampire bats can use to find and recognize preferred partners 455 (Carter and Wilkinson, 2016). The role of calls, in particular a possibly foraging-specific call 456 type ("n-shaped call" Figure S6, Figure S7) 1978; Schmidt, 1978;Wilkinson, 1985), but it remains unclear how often these competitive 471 interactions occur among familiar versus unfamiliar vampire bats. In our study, we observed 472 foraging vampire bats engaging in both affiliative and competitive interactions (see Table S3, 473 Videos S1-5). The competitive interactions were far more aggressive than what we observed 474 among familiar captive bats feeding from an accessible and unlimited source of blood. This 475 observation and our results above are consistent with the hypothesis that competitive 476 interactions are more likely between less familiar bats.

478
Implications for social dominance 479 The fluid nature of foraging encounters has potential implications for social dominance. 480 Dominance hierarchies should be common when animals move together in groups, because 481 the same frequent groupmates will also be primary competitors for first access to food. Such forms of social foraging in vampire bats may have implications for the spatial scale of 500 competition-a key factor shaping social evolution in humans (West et al., 2006) and other 501 group-living animals (Radford et al., 2016). In female vampire bats, cooperation occurs 502 'locally' with specific frequent roostmates, and competition over food might occur more 503 'globally' with members of the much larger population. If so, a more 'global' scale of 504 competition could reduce conflict and increase interdependence among highly associated 505 females. To test this idea, it would thus be useful to determine if sampled groups of vampire 506 bats consistently feed on the same or different prey individuals, and if vampire bats are more 507 likely to approach or avoid the social calls of foraging bats that are frequent roostmates 508 versus unfamiliar conspecifics. 509 510 Implications for describing social structure 511 A major advantage of proximity sensors was the ability to continuously track associations 512 among multiple individual bats both inside and outside their roost, which allows for the 513 construction of dynamic and multi-layer networks. Studies on social foraging and other social 514 behaviors in bats and other small highly mobile vertebrates have historically been limited by 515 the available tracking technology (Ripperger et al., 2020b). Radiotelemetry has poor spatial 516 resolution and continuously tracking many individuals is difficult. Current GPS-tags for bats 517 have rather short runtimes and the tags need to be recovered to download the data. On-518 board ultrasound recorders (e.g. Egert-Berg et al., 2018) do not reveal the identity of 519 encountered individuals. A major downside to proximity sensors was that many foraging 520 encounters occurred at unknown locations. However, placing proximity sensors or antennas 521 at more locations and on livestock would allow a better reconstruction of foraging behavior. A 522 combination of biologging approaches can also help to overcome existing challenges (e.g. 523 Leoni et al., 2020; Ripperger et al., 2020a). Standardized high-throughput methods for 524 measuring social network structure across bats and other diverse groups allow for 525 comparative studies that assess the relative ecological and evolutionary drivers of social 526 traits and social complexity across species. 527 528