Testosterone induces a conditioned place preference to the nest of a monogamous mouse under field conditions

Transient increases in testosterone (T-pulses) occur after social interactions in males of various vertebrate species, but the functions of T-pulses are poorly understood. Under laboratory conditions, the rewarding nature of T-pulses induces conditioned place preferences (CPPs), but what are the effects in a complex field environment? We present the first evidence that T-pulses administered to males at their nest site in the wild increased time spent at the nest regardless of pup presence in the monogamous, biparental, and territorial California mouse (Peromyscus californicus). Female partners of the T-males, in turn, spent less time at the nest. Independent of treatment, mice produced more ultrasonic vocalizations (USVs) when alone, but T-mice produced more USVs than controls. T-males produced USVs with a smaller bandwidth that likely traveled farther. Our combined results provide compelling evidence that T-pulses can significantly shift the behavioral focus and location of individuals in a complex field setting.

likely traveled farther. Our combined results provide compelling evidence that T-pulses can 31 significantly shift the behavioral focus and location of individuals in a complex field setting. 32

Introduction 33
Animals frequently adjust their allocation of time as they move through various life-34 history stages and meet different social challenges. One mechanism for adjusting approach to a 35 stimulus is through rewarding or reinforcing neural processes (Glickman and Schiff 1967) 36 through the repeated linkage between testosterone (T) release and the presence of a stimulus. T-37 pulses can act as an internal reward (Gleason et al. 2009) or reinforcing stimulus that when 38 released naturally or through an injection can increase approach to different stimuli such as the 39 physical location in which the T-pulse was experienced, at least under laboratory conditions in 40 rodents (e.g. Zhao and Marler 2014). Because male T-pulses are released in response to social 41 conditions such as a male-male challenge or a male-female interaction across a variety of species 42 including humans (Gleason et al. 2009), we speculate that neural reward and reinforcement 43 mechanisms allow adjustment to changing social challenges that can be linked to location (Zhao 44 and Marler 2016), such as different areas within a territory. In the case of a biparental species, T 45 release near the nest may provide a mechanism for increasing a male's attendance at the nest, as 46 suggested by the results of a laboratory study (Zhao and Marler 2014). Reinforcing effects can 47 alter the probability for the successful acquisition of essential resources necessary for survival 48 and reproduction (Tinbergen 1957). One paradigm for testing reinforcing effects is to assess 49 changes in behavioral preference for a location at which the stimulus (i.e. T-pulse in response to 50 a social stimulus; Gleason et al. 2009) occurred in the form of a conditioned placed preference 51 (CPP) (Arnedo et al. 2000;Frye et al. 2001). The reinforcing effects occur via activation of the 52 internal reward system (Bell and Sisk 2013). We explore the hormone T as a stimulus that has 53 Figure 1. Experimental Design. Paired California mouse males with and without pups were randomly assigned to receive three subcutaneous injections over five nights. After the third and last injection, we deployed the remote sensing equipment (automated radio telemetry, audio recording, and thermal imaging) to record individual behaviors for three consecutive nights. Created with biorender.com monogamous reproductive system of the California mouse and their known time management 119 and vocalization behaviors contribute to a compelling system for assessing behavioral responses 120 to T-pulses and the establishment of male T-induced CPP in the field to alter the amount of time 121 that males spend at the nest. 122

Results 123
Time at the Nest 124 Overall, T-males spent 14% more time at the nest than C-males (GLMM Estimate 125 0.14±0.05, p=0.02; Figure 2A). The T-conditioning appeared to be additive in the response to 126 pups where T-males with pups spent 23% more time at the nest than T-males without pups 127 (GLMM Estimate 0.21±0.04, p<0.01; Figure 2B; Supplemental Table 1). 128 . T-males spent 14% more than C-males (GLMM Estimate 0.14±0.05, p=0.02). B) Control male time at the nest by pups (C with pups, n=6; C without pups, n=5). C) Testosterone male time at the nest by pups (T with pups, n=6; T without pups). T-males with pups spent 15% more time at the nest than C-male with pups, and T-males without pups spent 12% more time at the nest than C-males without pups (treatment GLMM Estimate 0.13±0.03, p<0.01; pups GLMM Estimate 0.21±0.03, p<0.01). Circles represent individual data points.

142
T-females without pups spent 19.4% less time at the nest than T-females with pups, 143 whereas C-females without pups spent 11.6% less time at the nest than C-females with pups 144 (Supplemental Table 2). T-and C-females spent more time at the nest on night three of recording 145 compared to night one of recording (night three GLMM Estimate 0.10±0.04, p<0.02; 146 Supplemental Table 2). T-females spent 13% more time on night three than night one and C-147 females spent 6% more time on night three than night one (Supplemental Table 2). Females 148 spent less time in the nest during season one than season two, independent of treatment (season 149 one GLMM Estimate -0.15±0.06, p=0.02; Supplemental Table 2). T-females spent 15.6% less 150 time at nest during season one than season two (Supplemental Table 2). While C-male time at 151  Table 3). T-females spent 5% less time at the nest than their mates, whereas, C-158 females spent 18% more time at the nest than their mates (Supplemental Table 3). 159 160

USVs at the Nest 161
We recorded a total of 549 USVs across the 26 nest sites (T USVs=368, C USVs=181). 162 We assigned context to 385 USVs from video; 157 USVs were produced when a mouse was 163 alone (T USVs=101, C USVs=56), 119 USVs were produced when the mouse was <1m away 164 from another mouse (T USVs=94, C USVs=25), and 109 USVs were produced when the mouse 165 was >1m away from another mouse (T USVs=76, C USVs=33). T-pairs produced twice as many 166 total USVs at the nest than C-pairs (GLMM Estimate 0.87±.40, p=0.04; Figure 4A; 167 Supplemental Table 4). Independent of treatment, pairs also produced more USVs on night one 168 than night three after the last injection, (night two GLMM Estimate -0.33±0.26, p=0.15; night 169 three GLMM Estimate -0.76±0.26, p=0.01; Figure 4B and 4C; Supplemental Table 4). The effect 170 Supplemental Table 3. Descriptive Statistics on California Mouse Pair Time at the Nest. Each female was paired with a male who received three testosterone (T=10) or saline (C=8) injection at the nest, after the final inject, we recorded time spent at the nest for three consecutive nights.   Table 4). 181 We have evidence that T treatment influences specific USVs directly. Regardless of 182 distance between mates, T-pairs produced proportionately more 4SVs at the nest than control 183 pairs (W=43, p=0.03; Supplemental Table 5). All call types (1-, 2-, 3-, 4-, 5-, 6SV, and barks) 184 were recorded for the male and the female at both C-and T-nests. There was no significant 185 difference between treatments in any other call type produced (1-, 2-, 3-, 5-, or 6SV; p>0.137). 186 T-mice were more likely to call when the mate (or any other individuals besides the potential 187 presence of pups) was not at the nest (Treatment GLM Estimate 0.72±0.11 p<0.01; mouse alone 188 GLM Estimate 0.52±0.12 p<0.01) and there was a nonsignificant trend for more USVs produced 189 when the mouse was far (GLM Estimate 0.22±0.13, p = 0.09). When alone (regardless of pup 190 presence), T-mice were more likely to produce 1-, 2-, and 4SVs (1SVχ2=9.95, df=2, p<0.01; 191 2SVχ2 = 9.59, df=2, p<0.01; 4SVχ2 = 9.48, df=2, p<0.01). In C-mice there was no significant 192 difference in the type of call produced (1-5SVs) when alone, or when they were close or far from 193 There was a direct treatment effect on call bandwidth, T-males produced calls with a 196 11.25% smaller bandwidth than C-males (GLM Estimate -580.22±182.47, p<0.01; Figure 2; 197 Supplemental Table 6)  Supplemental Table 6. Descriptive statistics on spectral characteristics of male calls. The first call in the sequence for 1-, 2-, 3-and 4SVs produced by males (Testosterone = 86 and Control=31).
multiple T-pulses experienced in a location can increase the amount of time that males spend at 212 that location. Importantly, this change in time allocation occurred in the naturally complex 213 environment at the Hastings Natural History Reservation. We found evidence supporting the 214 concept that the weak conditioning effects of T pulses via CPPs increased time allocation by a 215 mammal to a location, the nest, within a territory in the wild. in California mouse reproduction is water availability (Nelson et al. 1995) and therefore, in 303 nature, California mice breed during the cold rainy season and cease breeding during the dry 304 summer months (Nelson et al. 1995). When reproduction occurs during harsh environmental 305 conditions and offspring require constant care, there must be a balance in the time invested 306 towards offspring maintenance and time spent towards foraging and resource defense. To 307 achieve balance, biparental care is essential for facilitating offspring survival and maximizing 308 reproductive success. We, therefore, propose that in some biparental species, T-induced CPPs 309 could be a mechanism for keeping the male at the nest to care for the young while the female 310 forages or conducts other behaviors related to territory maintenance. Another selection pressure 311 for T-induced paternal behavior may be increased protectiveness of pups to prevent the high 312 levels of conspecific infanticide found in rodents (Agrell et al. 1998  Interestingly, changes in male allocation to tasks closer to the nest resulted in females 318 spending more time away from the nest. This could occur through female preference/choice or 319 because males aggressively pushed females out of the nest. We again do not think that there was 320 an increase in aggression between the mated pair because of the absence of increased barks at the 321 nest as has been shown in mated pairs that are stressed (such as through a fidelity challenge; 322 Pultorak et al. 2018). An alternative possibility is that females change their spatial preference to 323 be away from the nest to compensate for the T-induced changes in male spatial preferences. We 324 also observed plasticity in female but not male time at the nest in different seasons, suggesting 325 plasticity in maternal behavior in response to environmental factors. In this study we speculate 326 that plasticity in the males is influenced by T from social stimuli, whereas the plasticity we see in 327 the females may be influenced more directly by the physical environment. In species that form 328 pair-bonds where both members of a pair are engaged in offspring care and territory defense, the 329 delegation of tasks is beneficial. In a wider variety of taxonomic groups, including insects, birds, USVs produced on night three. One difference between the studies is that Timonin et al (2018)  354 found that T-pairs produced proportionately more 1-, 4-and 5SVs, whereas we only found a 355 difference in 4SVs; we also found, however, that T-pairs were more likely to produce 1-, 2-, and 356 4SVs when alone. The difference between the studies could be attributed to year, population 357 densities, or a higher sample size in the current study. Anecdotally, densities were lower in the 358 current study which could alter social interactions. The current study also reveals that the 359 increased time apart in T-pairs may indirectly drive the greater number of USVs produced by the 360 T-pairs. However, while pairs call more when separated, T significantly increased calling rate 361 when a member of a pair was alone, with a nonsignificant trend when pairs were far apart (p = 362 0.09). 363 We further discovered that the increase in SV production was associated with a decrease 364 in bandwidth. An intriguing speculation is that the narrower bandwidth USVs produced are less R4500S DCC receiver/datalogger and a Yagi antenna (ATS), we located the pair the following 433 day at the nest (described below). All 33 putative pairs were confirmed as pairs when the signals 434 from both the male and female transmitters were emitted from the same nest. We ensured that 435 the tracked nest location was the primary nest and not one of the satellite locations by monitoring 436 nest occupancy for up to three days. A total of 28 pairs were in the nest for up to three days post-437 tracking, and we ensured that the nest was in a suitable location for setting-up our remote sensing 438 equipment (described below). We placed 15-20 Longworth traps (14 x 6.5 x 8.5cm, NHBS, 439 Totnes, Devon, UK) within a 2m radius surrounding the nest to trap the male and administer 440 injections (described below). 441

Treatment 442
We randomly assigned 28 males to receive either T (n=15) or saline ( included body mass as an independent variable in our statistical analysis. All three injections 451 were administered within five days, with only one injection per day. One male was excluded 452 because he did not receive all three injections within five days. We refer to females whose mate 453 received T as "T-females" and the nests as "T-nests". Females whose mate received saline are 454 referred to as "C-females" and the nests as "C-nests". We also recorded total number of nights 455 needed to administer all three injections (three or four nights), therefore, we included total nights 456 as an independent variable in our statistical analysis. After the third and last injection, we 457 deployed the remote sensing equipment (automated radio telemetry, audio recording, and 458 thermal imaging; described below) to record for three consecutive nights ("recordings nights" 1-459 3). We treated data collected by the remote sensing equipment over one night as a sample unit duration (100ms or less), that pass through multiple high to low and low to high frequencies with 497 a peak frequency of around 100kHz (Pultorak et al. 2015). Similar to the SVs, the barks, simple 498 sweeps, and complex sweeps occur as a single call or bout of calls. 499 We used ultrasonic microphones (Emkay FG Series from Avisoft Bioacoustics, Berlin, 500 Germany) to assess the number and type of USVs produced at the nest. We set up two 501 microphones; one next to the nest entrance and a second 2m away directly from the nest 502 entrance. Microphones recorded as described in Timonin et al. 2018. When possible, we 503 assigned USVs to individuals by matching the radio telemetry data with the time of the mouse 504 USV. By examining telemetry data within one minute of USV production and based on the 505 transmitter signal strength (Briggs and Kalcounis-Rueppell 2011), we determined if the male or 506 the female produced the USV. We were not able to assign 51% of the USVs to one individual 507 because both the male and the female were at the nest with strong transmitter signal strengths 508 and therefore, we only used the assigned data to test the treatment effect on the spectral and 509 temporal characteristics of USVs. The acoustic recording system was set-up at 27 nest sites 510 (T=15, C=12). Due to equipment failure, we did not record data at one T-nest. Our final dataset 511 consisted of 78 recording nights from 26 nest sites (T=14, C=12). Mouse USVs were counted 512 and classified into one of the following types: 1SV, 2SV, 3SV, 4SV, 5SV, 6SVs or barks ( 513 Kalcounis-Rueppell et al. 2018). We counted USV numbers recorded from sunset to sunrise and 514 refer to the value as "total USVs". Lastly, we determined if the proportion of a specific type of 515 USV (1-, 2-, 3-, 4-, 5-, 6SVs and barks) differed between treatments by totaling each USV type 516 per nest site and dividing by the total number of USVs produced at that nest. 517 Using SAS Lab Pro, we extracted spectral and temporal characteristics (duration, 518 bandwidth [number of frequencies a call passes through], and five frequency variables [peak, 519 minimum, maximum, start, and end]) from USVs recorded at the nest. Each spectrogram was 520 generated with a 512 FFT (Fast Fourier Transform), and a 100-frame size with Hamming 521 window. For each call, we measured duration, bandwidth, and five frequency parameters (start, 522 end, minimum, maximum, and frequency at maximum amplitude). 523 Thermal Imaging: 524 We used a thermal imaging lens (Photon 320 14.25 mm; Flir/Core By Indigo) to assign 525 social context to USVs. The thermal imaging lens was suspended to capture the full view of the 526 nest and a circular area with a 2m radius surrounding the nest. The lens was connected to a JVC 527 Everio HDD camcorder which recorded continuously throughout the night. We watched the 528 video footage in three-minute increments, (1-minute before, 1-minute during and 1-minute after 529 call production) to determine behavior and number of mice on the screen. If there was only one 530 member of a pair present at a time, the behavior was assigned as "alone". If both mates were 531 present, we determined the proximity of mice to each other by using a 1m scale that was overlaid 532 in the video for each site. If mice were less than 1m apart, we assigned them as "<1m", and if the 533 mice were more than 1m apart, we marked them as ">1m". We assessed the types of USVs (1-, 534 2-, 3-, 4-, 5-, 6SVs and barks) produced by context (alone, <1m or >1m) and treatment type. 535

Statistical Analyses 536
Time at the nest for both the male and the female was normally distributed and therefore 537 we fitted a Gaussian distribution. Pair time at the nest and total USVs were in violation of 538 normality and variances and could not be normalized and therefore we used either a 539 Quasibinomial or Poisson distribution. We used General Linear Models (GLM) with time at the 540 nest, pair time at the nest and total USVs as the dependent variables and we included individual 541 identification code independent of treatment type to account for individual differences. Using the 542 package lme4 (Bates et al. 2015), we fitted Generalized Linear Mixed Models (GLMM) with the 543 individual identification code as a random term and treatment as the fixed term. 544 In addition to treatment type, we also considered the following covariates: presence of 545 pups at the nest, season, male and female body mass, total nights needed to administer all three 546 injections, and recording night. Due to our small sample size, when modeling covariates we 547 included a maximum of two fixed terms in one GLMM model (treatment type and one 548 covariate). We first modeled the interaction term between treatment type and the one covariate. 549 If the interaction term was not significant, the term was dropped. We also used the non-550 parametric Wilcoxon Rank Sum test for our comparison of USV types. We compared the median 551 of the proportion of each USV type by treatment. We used a GLM to examine the relationship 552 between USV types by context and treatment. 553 For the analysis of the spectral and temporal characteristics, we used factor analysis to 554 extract principal component (PC) scores for the frequency parameters (as in Kalcounis-Rueppell 555 et al. 2010). We only analyzed calls assigned to an individual male or female and the calls were 556 analyzed separately. We generated a single PC score that represented the frequency variables 557 using the first call in the 1-, 2-, 3-and 4SVs sequence. We did not include 5SVs, 6SVs, and 558 barks due to a small sample size (<4). PC1 accounted for 67% of the variation in acoustic 559 variables for male calls and 71% variation for female calls. We fitted GLMM with ID as a 560 random term and USV type and treatment as the fixed terms. For both male and female calls, 561 duration and bandwidth variables were in violation of normality and variances. We, therefore, 562 fitted our models using a Poisson family distribution. PC scores were normally distributed, and 563 we used a Gaussian distribution in our models. All data are represented using box plots. We used 564 an alpha level of p<0.05 for the rejection criterion. All data were analyzed using R software 565 (Version 3.2.2.) 566

Acknowledgments 567
We thank A. Campos, J. Caprio, C. Falvo, M. Grupper, C. Kovarik, A. Larsen, J. Neill, 568 and E. Sakonjic for their assistance in data collection and Brian Trainor for providing our C and 569 T injectate. We also thank J. del Valle, V. Voegeli and Hastings Natural History Reserve for 570 their support and providing the use of their facilities during our field season. Lastly, we 571 appreciate the input from H. Li on all aspects of the analysis and B. Trainor, R. Bhandari, G. 572 Wasserberg and C. Snowdon for comments on the manuscript. This work was supported by the 573 National Science Foundation (NSF; IOS-1355163) and UNC-Greensboro. 574

Conflict of Interest 575
We declare RP, MCKR, and CCM have no competing interest. 576