Introduction

Many animals form groups with conspecifics throughout their lives or during key life history stages (Krause et al. 2002). Systematic group organization arises when the benefits of close associations with conspecifics, such as access to limited resources or participation in other mutualistic and cooperative behavior, exceed the costs, such as parasitism or predation (van der Post et al. 2015). For example, some animals aggregate when mothers are caring for dependent young to increase chances of survival or reproductive success and disperse once their offspring are independent (McComb et al. 2001; Silk et al. 2003). By studying social organization, even for temporally ephemeral groups, we can better understand the tradeoffs individuals make when they choose to live communally or independently.

Social network analysis provides powerful tools to assess individual associations within a group and evaluate group social behavior. Some animal interactions are defined using “Gambit of the Group” theory where, in the absence of observable behavior indicating a relationship, it is assumed that animals are interacting with proximal neighbors (Croft et al. 2008; Franks et al. 2010; Razik et al. 2022). Animal relationships are often characterized using association indices, which calculate the estimated proportion of time a dyad spends together, which can be used to map patterns in individuals’ relationships in a social network (Cairns and Schwager 1987; Whitehead 2008; Johnson et al. 2013).

Bats are interesting candidates for social network analysis—they are long-lived animals, and, in many species, individuals live communally for part or all of their lives (Altringham 2011). Some temperate bats form colonies during specific reproductive periods, called maternity colonies (Nagorsen and Brigham 1993; Lewis 1995). Female myotis in a single colony may switch between several maternity roosts but, via fission–fusion dynamics, not as a cohesive group (Patriquin and Ratcliffe 2016). Reproductive needs are often critical drivers of fission–fusion dynamics observed in bat maternity organization (Patriquin et al. 2010). Myotis group size typically peaks around parturition, when the unfurred pups and energetically stressed mothers depend on roost microclimates and social thermoregulation for survival (Sedgeley 2001; Olson and Barclay 2013). Similarly, northern long-eared myotis (Myotis septentrionalis) and Leisler’s bat (Nyctalus leisleri) have strong social bonds during pregnancy and lactation that may reflect bats seeking mutual warming and decreased competition for high-quality roosts (Patriquin et al. 2010; Nad’o et al. 2017). Colony organization often dissolves once pups are volant and capable of foraging because the benefits of coloniality no longer outweigh costs, such as predation or parasitism (Reckardt and Kerth 2007).

Yuma myotis, Myotis yumanensis, and little brown myotis, M. lucifugus, form fission–fusion maternity colonies in North America, but much about their social organization remains a mystery (Neilson and Fenton 1994; Olson and Barclay 2013). Both are small insectivorous bats that forage around water and each female produces one pup per year on average (Fenton and Bell 1979). Breeding female Yuma myotis and little brown myotis frequently select artificial structures for maternity colony roosts (Nagorsen and Brigham 1993; Mering and Chambers 2014). Bats of both species may roost together in the same maternity roost (Parkinson 1979). However, heterospecific social dynamics between species are poorly understood.

Using social network analysis to investigate Yuma myotis and little brown myotis maternity colony organization is challenging because some artificially roosting populations are large. The ideal social network analysis requires accurate identification of all group-members at many roosting locations to determine relationships over time and space (Whitehead 2008). Silk et al. (2015) estimates that at least 30% of a group should be tracked daily for accurate social network description. If a colony exceeds several hundred individuals, as has been observed for Yuma myotis in particular, tracking 30% of the group every day is a difficult proposition and will result in an imperfect analysis. While methods exist to correct for missing animals or imperfect identification, these solutions are largely arbitrary (Ginsberg and Young 1992; Nagorsen and Brigham 1993). However, it is still possible to use the tenets of social network analysis to study large colonies by tracking the majority of bats at a few key roosting features. While this approach will not yield a perfect representation of colony relationships, studying many individuals at a few key maternity roosts can still provide details of changes in colony organization due to reproductive status.

In this study, we quantified associations of mixed little brown myotis and Yuma myotis maternity colonies in artificial structures near Vancouver BC, Canada, to test the hypothesis that species identity and reproductive phase drive patterns in associations. Specifically, we predict that (1) we will find that bats preferentially associate with conspecifics over heterospecifics and (2) associations between colony members will be strongest during pregnancy and lactation, when pups need the most care to survive, and weakest during post-lactation, when pups are independent.

Methods

In 2018 and 2019, we used PIT-tag monitors at 2 maternity colony sites in the lower mainland of British Columbia. Both colonies were located in human-built structures in forested areas near Vancouver. Our first site was located near Stave Lake, BC. The bats primarily roosted in the walls of a wooden building, Stave Building 1. Stave Bat Box 1 was on the west-facing outside wall of another building ~ 1 km from Stave Building 1. Bats also roosted in two identical bat boxes (Stave bat boxes 2 and 3) placed on the southwest corner of Stave Building 1, approximately 4 m from the exit of the Stave Building 1 roost. All boxes were 15.2 × 43.2 × 10.2 cm and had 4 chambers each. The date of installation of the bat boxes is unknown, but all roosts have been monitored since 2018. In midsummer, we discovered a second building roost that also contained a mixed Yuma myotis and little brown myotis maternity colony, located approximately 0.5 km from Stave Bat Box 1 and 1.5 km from Stave Building 1. We did not install PIT-tag equipment at Stave Building 2.

The second colony was located in several buildings near Alice Lake in Squamish, a predominantly Yuma myotis maternity colony resided in corrugated tin roofs of several buildings and in three bat boxes (15.2 × 43.2 × 10.2 cm; 4 chambers each) attached to two of the buildings. One bat box faced east (Alice Bat Box 1) and two were adjacent to each other facing south (Alice Bat Boxes 2 and 3). All bat boxes were installed in 2016. We also discovered a mixed little brown and Yuma myotis maternity colony roosting in the roof of one of the buildings in 2019, called Alice Building, Alice Bat Box 1 was 23 m from Alice Building and 38 m from Alice Bat Boxes 2 and 3. Alice Building was 57 m from Alice Bat Boxes 2 and 3.

For the buildings in our study, we were not able to state with confidence whether the bats segregated by species within these larger and more diffuse roosts, because many bats were not visible. The conclusions we have derived from bat box roosts are most likely accurate as there were only 4 chambers. We observed that the animals formed one group when the box was sparsely occupied and bats used all chambers when it was densely occupied. Given the small size of the bat boxes, we inferred that heterospecifics were not segregating by species within the structure.

We captured bats in mist nets and homemade harp traps at both sites. For all captures, we met the standards of the Canadian Council on Animal Care, permitted by the government of British Columbia in accordance with the Wildlife Act under Dr. Cori Lausen (Permit # MRCB15-163,558) and approved by the University of British Columbia Animal Care Committee (Certificate # 718–0031). We were not able to record data blind because our study focused on wild animals in their natural environment. We implanted passive RFID chips (Biomark, Boise, ID; HPT8 and HPT9) under the skin between the scapulae of healthy, adult females who weighed > 4.0 g and sealed the incision sites with Vetbond. We banded adult females, males, and juveniles weighing 3.5–4.0 g with aluminum bands engraved with a unique identifier.

We searched in June for unknown alternative roosts by radiotracking 2–3 bats at each site. We glued small radio transmitters (< 5% body mass, Aldridge and Brigham, 1988) on the backs of bats to follow their movements for 1–2 weeks. We trapped newly located roosts at least once to establish species identification and to look for previously tagged bats.

Following the recommendations of Weller et al. (2007), we classified bats to species by inputting individuals’ forearm length and acoustic call frequencies into an algorithm that determines the probability that a bat was either Yuma myotis or little brown myotis. Bat species identity with probability ≥ 90% were used and all others were assigned the ambiguous identity of YULU—either Yuma myotis or little brown myotis. All juveniles were assigned YULU. Hybridization has been recorded in some myotis species, particularly in Europe (Bachanek and Postawa 2010). Although hybridization between Yuma myotis and little brown myotis is possible, the only genetic analysis to date found these are two distinct species (Herd and Fenton 1983).

Reproductive females were classified by palpating the abdomen and examining teats, and assigned as pregnant (P), lactating (LAC), or post-lactating (PL) (Racey 1974). Females with some hair missing from around the nipples but no indication of recent nursing were classified as parous (PAR) meaning they had reproduced in the past, but did not have a pup in the current summer. If phalangeal epiphyses of a bat were not yet fused together, then the bat was classed as juvenile (Racey 1973). All juvenile females were classed as nulliparous (NULL) and all juvenile males were classed as non-reproductive (NR). Adult females with difficult-to-find nipples (no worn hair around them) were classed as nulliparous (Racey 1974).

During June 2019, we installed Biomark PIT-tag readers to monitor bats tagged with RFID chips; we recorded bat roosting behavior until late August 2019. We placed three PIT-tag systems at four known roosts at Stave Lake: one system at the exit to Stave Building 1, one system to cover Stave Bat Boxes 2 and 3, and the final system for Stave Bat Box 1. We placed two PIT-tag systems at Alice Lake: one system monitored Alice Bat Box 1, and the other monitored Alice Bat Boxes 2 and 3 so that any bats entering either box would be recorded. Alice Bat Box 3 was rarely occupied, and only by a few individuals. With the exception of the Stave Building 1 exit, which used a 25-ft cable antenna powered using deep cycle batteries, all PIT-tag systems used a single or double bar antenna to read RFID chips in bats approaching within 3 inches of the antenna as they exited or entered the roost. We monitored all systems weekly for reliability (Supplementary Material A).

We calculated the probability a bat would leave a roost, displayed in Fig. 1, by counting the number of times an individual bat left a roost after detection the previous day. We then divided this number by the total number decisions to stay or leave. To show how this probability changes over time, we calculated this probability for each reproductive period.

Fig. 1
figure 1

The daily probability of a bat leaving a roost based on reproductive period during 2019 at Alice Lake and Stave Lake in British Columbia, Canada. Values are medians, quartiles, and range. Dots represent outliers, such as bats that only appeared a few times during the reproductive period but always switched

Full size image

Network construction

We created a series of time-aggregated social networks to test for differences in bat associations during pregnancy, lactation, and post-lactation. Stave Lake bats were at each stage 2–3 weeks earlier than Alice Lake, likely due to the latter site’s higher elevation (Table 1). All networks were created in R Studio version 1.3.1093, using the R package ‘igraph’ and network edges were weighted using R package ‘asnipe’ (Csardi and Nepusz 2006; Farine 2013). Raw PIT-tag data used to calculate all networks can be found in Supplementary Material C.

Table 1 2019 dates used to construct each seasonal network by site. Data collection ceased prior to the seasonal disbanding of the colonies
Full size table

We monitored the roosts for 89 days at Alice Lake but the PIT-tag reader at the Stave Building 1 ceased working on August 15, so we analyzed Stave Lake bats for only 72 days. As parturition was earlier at Stave Lake than Alice Lake (Table 1), we assume that this error did not affect our results. We used the automated PIT-tag reader data of daily individual PIT-tag records to weight the edges of our networks (Croft et al. 2011) using the Simple Ratio Index (SRI) (Ginsberg and Young 1992). SRI-calculated associations of dyads range between 0 (never observed together) to 1 (always observed together). We excluded bats that were observed ≤ 4 times to remove animals that died or were only peripherally associated with the colony (Whitehead 2008).

Network analysis

To assess whether the observed SRI values between bats were non-random, indicating specific colony structure, we compared all observed association matrices to randomly permuted networks in the R package ‘asnipe’ (Farine 2013). We constructed 1000 randomized networks for comparison against the observed network by swapping group membership within observed association matrices (Bejder et al. 1998; Whitehead et al. 2005; Manly 2006). The p value was calculated from the proportion of random networks with a coefficient of variation larger than the observed network (Manly 2006).

We measured modularity to assess subgroup partitioning (Newman and Girvan 2004; Newman 2006). Highly modular networks, with Q values > 0.03, are characterized by dense connections within subgroup nodes but are less tightly connected between subgroups. We used the infomap algorithm in R package ‘igraph’ (Rosvall and Bergstrom 2007) for community detection and modularity calculation. While not included in this manuscript, we calculated degree of centrality for all bats and looked for characteristics in highly centralized individuals (Supplementary Material B).

As both colonies contained little brown myotis and Yuma myotis, we tested for assortative mixing by species in the colonies to see if bats were more likely to roost with conspecifics over a single reproductive period (Newman 2002; Shizuka and Farine 2016). We calculated Newman’s assortative coefficient, which ranges from − 1 (avoidance of conspecifics) to 1 (preference of conspecifics), for each reproductive period at each site. We used the ‘assortnet’ package in R (Shizuka and Farine 2016) to calculate the weighted assortative coefficient for bats identified to species. We compared the observed assortative coefficient against the expected in random networks created by our previously described permutation method. We then compared assortative coefficient values with significance greater than 0.05 between reproductive sampling periods (Newman 2002). Because many bats could not be conclusively identified to species, we only included animals with 90% or higher species identity certainty in this analysis.

Results

We captured 1108 bats from the two sites over 16 capture sessions (Table 2) with 623 bats captured at Stave Lake and 485 bats at Alice Lake. Of the 481 adult female bats we PIT-tagged at Stave Lake, 39.3% were little brown myotis, 22.0% were Yuma myotis, and 38.7% were unidentifiable as either species. At Alice Lake, we captured and PIT-tagged 338 adult female bats; 7.2% were little brown myotis, 70.8% were Yuma myotis, and 22.0% were unidentifiable. Ninety-nine tagged bats at Alice Lake and 163 at Stave Lake (20–25% of tagged animals) were never recorded again after their initial capture. The majority of these missing bats were captured early in the season or at roosts that did not have PIT-tag scanners attached to monitor bat behavior.

Table 2 Summary of capture efforts at Stave Lake and Alice Lake in 2019. Only tagged bats with ≥ 4 observations were included in the social network analysis
Full size table

At Stave Lake, we recorded roost occupancy for 342 bats over 72 days. Of the original 342 bats, 228 were recorded more than 4 times and included in our analysis, resulting in 4636 bat-day-roost observations. The average Stave Lake bat was recorded 18.1 ± 11.5 (mean ± SD) times throughout the season, and an average of 18.7% of the tagged bats were observed daily at one of the 3 monitored roosts. PIT-tag roost monitoring ended on August 15, 2019, when the Stave Building 1 reader failed. Using radio telemetry, we located the previously unknown Stave Building 2 roost site by radiotracking 3 bats in late June; we then captured 89 bats at the roost in late July. Based on an early July colony count, > 200 adults occupied Stave Building 2.

At Alice Lake, we monitored 3 bat boxes for 89 consecutive days, obtaining 3274 bat locations. We recorded 242 individual bats, 167 of which were recorded more than 4 times and included in the analysis. Alice Lake bats were recorded on average 14.7 ± 8.6 times in the boxes throughout the season and, on average, 13.2% of the tagged bats were recorded in a monitored roost each day of the study. We discovered bats roosting in the tin roof of the Alice Building after radiotracking two bats in June, and we subsequently trapped bats emerging from two locations in the roof.

Associations during pregnancy and lactation

At both Stave Lake and Alice Lake, there were strong individual associations between colony members during pregnancy and lactation; network density was above 60% at both sites. Average SRI values were 0.23 ± 0.18 (mean ± SD) at Stave Lake and 0.15 ± 0.16 at Alice Lake during pregnancy, and 0.20 ± 0.19 at Stave Lake and 0.16 ± 0.19 at Alice Lake during lactation, indicating that most bats had strong associations with their colony members (Table 3; Fig. 2). There was little ingress or egress of individuals within the colony during these periods. At Stave Lake, 181 of 189 bats were recorded in both of the pregnancy and lactation social network periods. Similarly at Alice Lake, 90 of 96 bats were recorded during these same periods.

Table 3 Comparison of social network metrics by site and reproductive period. Average SRI for all dyads were calculated as a measure of association strength across the colony. Each bat has a degree-centrality, indicating how many other bats they associated with during the time period. Density and modularity are network level measurements
Full size table
Fig. 2
figure 2

Bat social network visualization by season and location. Stave Lake (A) pregnancy, (B) lactation and (C) post-lactation networks and Alice Lake: (D) pregnancy, (E) lactation and (F) post-lactation networks. Nodes represent individual bats; edges represent simple ratio index associations. Bats that are spatially closer together have stronger associations with each other than bats that are spatially separate. Species identity is denoted by node color: white for little brown myotis, black for Yuma myotis and gray for bats we could not identify to species

Full size image

Associations during post-lactation

In contrast, bat associations within the maternity colony weakened and colony organization fragmented during post-lactation, a trend reflected in the decrease in some measures of network connectedness at both sites (Table 3; Fig. 2). Bats were less likely to roost with colony members at Stave Lake, as the SRI values (0.07 ± 0.13), mean degree-centrality (60 ± 28) and overall network density (0.36) dropped from previous reproductive periods. At Stave Lake, the range of roost switching probabilities of individual bats increased from previous periods (Fig. 1). Fifty tagged bats captured early in the season were not detected in monitored roosts until post-lactation. These bats were tagged at the beginning of the season or at Stave Building 2 in July.

During post-lactation at Alice Lake, bats began to roost independently of colony members—marked by a drop in the average SRI values (0.08 ± 0.09)— while the mean degree-centrality increased (101 ± 32) and the network density (0.63) remained the same as previous periods. The probability of roost switching increased during post-lactation at Alice Lake (Fig. 1). We observed 32 previously tagged bats that had not been recorded prior to this period that suddenly arrived at the monitored roosts. These new bats were tagged in May and early June before the PIT-tag systems were installed at Alice Lake, or during captures at the Alice Building. Bats switched from Bat Box 2 to Bat Box 1 during this time but the group remained together during this move.

Heterospecific associations

At Stave Lake, 295 (95%) tagged bats roosted almost exclusively in the Stave Building 1 or other, unknown roosts during pregnancy and lactation. During post-lactation, the colony split between Stave Bat Box 1 and the Stave Building 1, contributing to the formation of two distinct communities within the Stave Lake colony. The network modularity score, which had been 0 during pregnancy and lactation, rose to 0.39 during post-lactation. The Infomap algorithm assigned bats in this period to 3 communities (Fig. 3), although one community consisted of only 2 bats that were observed only observed roosting with the group once or twice. Excluding bats with indeterminate species identity (approximately 40%), Community A was predominately Yuma myotis while community B was predominately little brown myotis but still mixed with Yuma myotis (Fig. 3). We also observed that little brown myotis left the study site first at Stave Lake with few detected after the first week of August. Yuma myotis moved to the Stave Bat Box 1 around August 8th and persisted until September.

Fig. 3
figure 3

Network visualization of post-lactation period at Stave Lake by community assignment by Infomap, which is designated by shape, and by species identity, which is designated by color

Full size image

Bats of both species at Stave Lake roosted together during pregnancy and lactation, then segregated by species during post-lactation as they split into different communities. While the assortative coefficient for 129 bats identified to species was very low during pregnancy (0.02) and lactation (0.00), there was an increase during post lactation (0.06); this value suggests that bats had higher affinity for conspecific bats as young became volant. All assortative coefficients were significantly different from randomly permuted networks (p < 0.01).

In contrast, assortative species mixing did not occur at Alice Lake for 142 bats identified to species: assortative coefficient values for all reproductive periods did not differ from a random expectation (p > 0.10), indicating no species segregation. Our analysis accounts for species distribution because the species ratio is preserved during randomization, so the species mixing is not just an artifact of low little brown myotis numbers.

Discussion

Our results clearly show that Yuma myotis and little brown myotis maternity social organization is influenced by both reproductive phase and species identity. As we predicted, colony associations were strongest during pregnancy and lactation, and weakest during post-lactation. While we predicted that bats in mixed species colonies would preferentially associate with conspecifics, we found that this was only true during post-lactation at Stave Lake.

We found maternity colony organization of little brown myotis and Yuma myotis had marked temporal differences related to reproductive period. During pregnancy and lactation, bats had strong associations with other colony members at monitored roosts; the colonies’ social networks were dense, highly connected, and centralized. During post-lactation, bats formed weaker associations with colony members and were more likely to switch roosts. Seasonal changes in behavior are well established in many bat species, including little brown myotis. For example, the group size of breeding little brown myotis in tree roosts peaks during parturition, when social thermoregulation most benefits the unfurred pups (Olson and Barclay 2013). At the end of summer, juveniles and adult little brown bats use torpor more frequently to aid in fat accumulation and roost in smaller groups or independently (Dzal and Brigham 2013). Weaker post-lactation associations are associated with increased roost switching in noctule bats (Nyctalus noctule) and Leisler’s bats (N. leisleri). While bats switched roosts throughout the summer, averaging every 1.4 days at Alice Lake and every 2.2 days at Stave Lake, patterns of roost switching also changed by reproductive period (Rensel 2021). At both Stave Lake and Alice Lake, Rensel (2021) found a bat was less likely to switch roosts during pregnancy and early lactation, and most likely to switch during late lactation and post-lactation. This contributed to changes in our social network analyses, including the influx of new bats during post-lactation and the weakening colony associations.

Our results suggest that little brown myotis and Yuma myotis can roost with heterospecifics during pregnancy and lactation. Mixed myotis maternity colonies have been recorded as far north as BC, Canada (Kellner 2020) and south to New Mexico, USA (Chung-MacCoubrey 1999). While we are confident that bats observed in bat boxes were roosting with heterospecifics, we were not able to observe if the bats in Stave Building 1 were segregated by species because we did not observe grouping behavior. We speculate that heterospecific roost sharing reflects limited availability of high-quality roosts, linked to increased offspring fitness, so Yuma myotis and little brown myotis roosting together in a single roost should maximize mutualistic benefits (Kerth 2008; Salinas-Ramos et al. 2020). 

We also observed species segregation at Stave Lake during post-lactation. Specifically, little brown myotis roosting in the Stave Building 1 left in mid-August, while Yuma myotis roosting in the Stave Bat Box 1 persisted until early September. Similarly, the predominantly Yuma myotis colony at Alice Lake roosted in the bat boxes until the last week of September. This is not a unique observation: other maternity sites in BC have similar species-specific behavior with little brown bats departing before Yuma bats at maternity sites (Lausen et al. 2021). We cannot say for certain that all the remaining bats after mid-August were Yuma myotis, as many were not conclusively identified as either species.

By focusing our social network analysis on a few key roosts, we found a unique opportunity to directly link specific artificial roosts with patterns of social organization. Similar PIT-tag studies, where colonies with less than 100 individuals were monitored as bats moved across 50 or more artificial roosts, had stronger average dyadic associations than observed in our colonies (Kerth et al. 2011; Webber et al. 2016; Zeus et al. 2018). We observed < 20% of tagged bats on a daily basis, compared to 50% or greater in the studies that focused on bats in natural roosts. We suspect our average association index values may be lower than reality as bats may have associated with other colony members in roosts we did not observe. When we calculated daily detection rates during specific reproductive periods, Stave Lake bats were detected daily during pregnancy and lactation 46.4% ± 21.7% and 41.7 ± 25.0%, respectively. Daily detection rates were consistent at Alice Lake at about 18% across all reproductive periods, but we also point out that there was a major roost approximately 50 m from our bat boxes that could account for this low detection rate. To counteract this possible bias, we chose metrics that relied on network topography (degree-centrality, density, modularity) and focused on comparisons among temporal periods. Our research demonstrates viability of using social network techniques at a few artificial roosts to monitor changes in social structure in large colonies, and highlights that social organization is temporally dependent on reproductive status.