How does individual variation in sociality influence fitness in prairie voles?

Comparative studies aid in our understanding of specific conditions favoring the initial evolution of different types of social behaviors, yet there is much unexplained intraspecific variation in the expression of social behavior that comparative studies have not yet addressed. The proximate causes of this individual variation in social behavior within a species have been examined in some species but its fitness consequences have been less frequently investigated. In this study, we quantified the fitness consequences of variation in the sociality of prairie voles (Microtus ochrogaster). We characterized sociality of voles in semi-natural enclosures using an automated behavioral tracking system paired with social network analyses to quantify the degree of spatial and temporal co-occurrence of different voles. We then assessed the relationship between sociality with mating success (number of different conspecifics with which an individual produced offspring) and reproductive success (total number of offspring surviving to first capture). We measured the number of social connections each individual had with all voles and only with opposite-sex voles by calculating unweighted degree through social network analyses. Both female and male voles varied in the number of social connections they had with all conspecifics and with opposite-sex conspecifics. Voles with an intermediate number of social connections with voles of both sexes had higher mating success overall. In our analyses that considered all social connections with voles of both sexes, voles with an intermediate number of social connections produced more offspring. Males with a very high or low number of social connections also had the lowest average body mass. Overall, our results suggest some limit on the fitness benefits of sociality. Although there was substantial individual-variation in our measure of vole social behavior, intermediate levels of social connections may be most favorable.

To determine parentage, we used Cervus 3.0.7 parentage analyses with known sexes, which 227 calculates a likelihood ratio for each potential mother and father in order to determine the most 228 likely biological parents in the population for each offspring . We were able to 229 determine both parents (trio confidence level) with a 95% confidence level for 33/41 (80.5%) 230 offspring, so only these 33 offspring were included in the analyses of mating and reproductive 231 success. 232

Density and Body Mass 244
Population density was calculated based on the number of unique individuals caught 245 within each two-week period (over each two-week period both enclosures were trapped with 246 equal effort except for occasional cancellations due to weather). To investigate population 247 density over the course of the field season, we used a linear model with density (log-248 transformed with base 10 to improve normality of residuals) with the fixed effects of enclosure 249 and weeks in the study and the interaction of these terms. Sex ratio was calculated by dividing 250 the number of adult males by the number of total adults for each two-week period. We used a 251 binomial generalized linear model to investigate sex ratio with the fixed effects of enclosure and 252 weeks in the study and the interaction of these terms. VIFs for all non-interaction terms were all 253 < 3.57. 254

Social Network Analyses 255
We measured the number of social connections (unweighted degree, hereafter degree) 256 between same-sex or opposite-sex voles based on co-occurrence data from the RFID 257 antennas. Individuals with a high degree would have had instances of spatial and temporal co-258 occurrence with many other voles whereas those with a low degree had few. We conducted all 259 social network analyses using the R package asnipe version 1.1.4 (Farine, 2017b). In order to 260 generate our social networks, we took the PIT tag readings from the RFID antennas and ran 261 them through a Gaussian Mixture Model with each day labeled separately (Psorakis et al., 262 2012). This model goes through the raw data of the PIT tag readings and creates groups based 263 on when tag readings at the same antenna are clumped throughout time. Therefore, there is not 264 a uniform time period used to create these groups, they are based on how our data were 265 distributed over time. This model uses clusters of tag readings as "centres of mass" where data 266 are concentrated and then determines the groups based on the amount and distribution in time 267 of tag readings in each cluster to determine where to split groups (Psorakis et al., 2012). The 268 duration of these group events ranged from 0 seconds (so voles were both at the antenna at the 269 same time) to 66,161 seconds with an average of 655.2 ± 3,352.8 seconds. This then creates a 270 group by individual matrix where being in the same spatial and temporal "group" counts as an 271 association between individuals. As we were only interested in the number of connections each 272 individual had (not the strength of these connections), we used a binary, unweighted 273 measurement of degree where any non-zero association was counted as a "1". Thus, anytime 274 we refer to the number of social connections in this paper, we calculated this using the 275 unweighted degree. For more details about the construction of the social networks see Sabol et For all models including the number of mates (mating success) or the number of 279 offspring produced that survived to emergence from the natal nest (reproductive success) as the 280 response variable, we used Poisson generalized linear models. For each response variable we 281 ran two models using social network data, one including all social connections in order to 282 investigate sociality overall and one including only the opposite-sex social connections. These 283 models all had fixed effects of the number of social connections, the interaction of this with sex, 284 the number of social connections squared to assess non-linear effects of social connections on 285 mating success or reproductive success (one model including all connections and another for 286 each response variable including only opposite-sex connections), the interaction of this with sex, 287 enclosure, and survival (calculated as the proportion of the field season the individual survived 288 based on last detection). To test if mating success and reproductive success were related, we 289 ran a separate model with the number of offspring produced as the response variable and fixed 290 effects of the number of mates with which individuals produced offspring, the interaction of this 291 with sex, and enclosure. None of the GLMs were over-dispersed as all the dispersion 292 parameters were <1, which we tested using R package AER version 1.2.5 (Kleiber and Zeil, 293 2008). VIFs were all < 3.5 except interaction and squared terms, which were predictably high. 294

Body Mass 295
To investigate body mass, we calculated the average body mass for each male vole for 296 the entire field season (range 1-19 measurements, average 7.75 measurements). Females 297 were not included because we were using body mass as a proxy for body quality, and female 298 mass would be affected by both pregnancy status and body condition. We then used a general individual. We visually assessed the distribution of the data and residuals for normality. VIFs 305 were all < 2.2 except interaction or squared terms and survival. However, when survival was 306 excluded from the model, VIFs for all of the other terms were < 3.5 except interactions and 307 squared terms. Including survival did not alter the statistical significance of any of the results 308 shown below so we left it in. 309

Randomized Models 310
For every model that included unweighted degree (the number of social connections), 311 we used the network permutation method in asnipe (Farine, 2017b). This method is useful 312 because it helps control for the fact that social network data are not independent. This method 313 also allows us to investigate our hypotheses more specifically by allowing us to test if the 314 observed relationships are significantly different from random networks with the same structure 315 was swapped. We also restricted swaps to only voles in the same enclosure that were recorded 320 on the RFID antennas during the same day to control for voles that did not survive the entire 321 season. Further, for the opposite-sex networks we restricted swaps to include only voles of the 322 same sex so that we were only comparing our opposite-sex network to other opposite-sex 323 networks, not all possible combinations. We then compared the regression coefficients from the 324 model for each variable that includes a social network statistic to corresponding b-values from 325 randomized networks and calculated a new P-value based on the number of randomized 326 models that produced a b-value with a higher absolute value than the absolute value of the 327 observed model. Therefore, our P-value shows us whether the relationship we have observed is 328 stronger than the relationship from 10,000 randomizations of our dataset (Farine, 2013). We ran mating success seemed to be largely due to males having slightly more overall social 365 connections than females while female mating success peaked at a lower number of social 366 connections (Fig. 2a). There is also a qualitative difference in the shape of the curve, with 367 female mating success peaking at a lower number of social connections but then dropping off 368 more steeply, while male mating success peaked at a higher number of social connections but 369 declined more gradually (Fig. 2a). 370 The same relationship was true when only opposite-sex connections were considered.  Table 1, Fig. 2b). This 375 latter difference between the sexes seems to be driven by the difference in the number of social 376 connections between the sexes with females tending to have slightly more social connections 377 than males. 378 Overall, both female and male voles in the low-density enclosure had higher mating 379 success than individuals in the higher density enclosure (from model for all social connections: b 380 = 1.89, z = 4.55, P < 0.0001; from model for all opposite-sex social connections: b = 2.03, z = 381 4.72, P < 0.0001, Table 1). Individuals that survived in the enclosures for longer had higher 382 mating success (all social connections: b = 2.07, z = 2.33, P = 0.020; opposite-sex social 383 connections: b = 1.75, z = 2.09, P = 0.037, Table 1 voles, density is quite variable across years (Getz et al., , 2001) and some previous 20 observational studies of prairie voles in field settings suggested that socially monogamous 483 behavior is more common at low densities (McGuire et al.,1990;Solomon et al., 2009; but see 484 Getz and McGuire,1993). There is also some evidence that resource distribution may impact the 485 mating strategy of prairie voles and this effect may be mediated through its influence on density 486 . This suggests the possibility that selection on the social behavior of 487 prairie voles varies among years due to changes in population or female density but additional 488 multi-year studies measuring a broader array of social behaviors in free-living voles are needed 489 to test this prediction. This could reflect the energetic costs associated with having many social connections or living in 499 a large group (e.g., Lutermann et al., 2013), or these could be agonistic interactions with males 500 on neighboring territories, resulting in males investing more time in territory defense than males 501 with fewer neighbors. Why males with very few social connections were also lighter in body 502 mass is not clear but these males may have been of lower phenotypic quality given that they 503 had few social connections, low body mass, and low mating and reproductive success. 504 Alternatively, having fewer social connections could result in a loss of body mass if these males 505 had no assistance in territory defense and thus, expended more energy than males with more 506 social connections (e.g., having a female social partner). Females likely face many of these same tradeoffs, but as we did not test quality in females (due to changes in mass being linked to 508 pregnancy) and so further study is needed to investigate this relationship in females. 509 Our results suggest that it is not advantageous for voles to have social connections with 510 too many opposite-sex conspecifics. One possible explanation is that individuals with an 511  those regression coefficients were compared to those from randomized networks three times to 786 determine if they were consistently significant (see methods). "# of measures" refers to number 787 of times we measured body mass, which were used to generate average mass for each 788 individual. Survival refers to proportion of days the vole was in the enclosure. "LD" is low-density 789 enclosure. week period of the field season. Note that the area of enclosures is equal, so the number of 798 voles in each enclosure can be used to compare relative density between the two, b) The sex 799 ratio in each enclosure, calculated as the proportion of total adult voles trapped during each 800 two-week period of the field season that were males, over time of the study. 801  Table 2. 816 offspring with a greater number of different mates) produced a great number of offspring that 820 survived to emergence from the natal nest. Points for females and males are jittered. Full results 821 shown in Table 3. 822 male voles in their enclosure or b) with just female voles, were significantly heavier over the 826 course of this study. Body mass for males was averaged for the entire duration of this study. 827 The number of times we measured body mass ("N") varied among males so the size of each 828 point is scaled based on the number of recorded mass measurements we have for each 829 individual. Full results shown in Table 4. 830 Figure A1. Layout of the two RFID arrays in the enclosures. The RFID system was kept at each 833 array in each enclosure for three days in the order: array 1 enclosure 1, array 2 enclosure 1, 834 array 1 enclosure 2, and array 2 enclosure 2 and then repeated for the duration of the field 835 season. 836