Kinship and reproductive condition correlate with affiliation patterns in female southern Australian bottlenose dolphins

Social relationships in female mammals are usually determined by an interplay among genetic, endogenous, social and ecological factors that ultimately affect their lifetime reproductive success. However, few studies have attempted to control for, and integrate these factors, hampering our understanding of drivers underlying female sociality. Here, we used generalized affiliation indices, combined with social networks, reproductive condition, and genetic data to investigate drivers of associations in female southern Australian bottlenose dolphins. Our analysis is based on photo-identification and genetic data collected through systematic boat surveys over a two-year study period. Female dolphins formed preferred associations and social clusters which ranged from overlapping to discrete home ranges. Furthermore, matrilineal kinship and biparental relatedness, as well as reproductive condition, correlated with the strength of female affiliations. In addition, relatedness for both genetic markers was also higher within than between social clusters. The predictability of resources in their embayment environment, and the availability of same-sex relatives in the population, may have favoured the formation of social bonds between genetically related females and those in similar reproductive condition. This study highlights the importance of genetic, endogenous, social and ecological factors in determining female sociality in coastal dolphins.

Female affiliation patterns. We found a significant correlation between home range overlap, gregariousness and the association indices; therefore these structural variables were controlled for when estimating female GAIs. The number of sightings per female dyad did not show a significant correlation with the association indices, and therefore it was not controlled for during GAIs estimation (Table 1). GAIs using deviance residuals ranged from −4.52 to 7.06 (mean = −0.41; SD = 1.53; n = 1485). We detected the presence of non-random companionships among female dolphins using a permutation test (Observed SD = 0.1; Random SD = 0.08; p < 0.01), and using GAIs deviance residuals, 117 preferred, 1331 casual and 37 avoided female pairs were identified ( Table 2). The Newman's modularity clustering technique revealed that the Coffin Bay female population was subdivided into seven social clusters (Qmax = 0.45; Fig. 1; Table 3), which ranged in size from two to twelve individuals (mean = 7.9; SD = 3.63). Social clusters showed a mixture of ranging patterns (Fig. 2), with some clusters showing overlapping areas of usage and others using discrete areas in the Coffin Bay embayment. As anticipated, social clusters that were closer in the social network also showed similar areas of spatial use within the bay.
We found a positive correlation between female reproductive condition and affiliation indices (r = 0.08; p < 0.05), and higher values of affiliation among females in similar reproductive state ( Fig. 3; same reproductive state median GAIs = 0.76; different reproductive state median GAIs = −0.91; p < 0.001). Pairs of females in similar reproductive condition associated more often than those in different states; thus, females with calves associated more often with other females with calves, and females without calves associated more often with other females without calves. www.nature.com/scientificreports www.nature.com/scientificreports/ Affiliation patterns and kinship. MtDNA control region haplotype and microsatellite data were gathered for 44 and 49 female dolphins considered in the social analysis, respectively. We only found two haplotypes, both with similar frequencies (A = 23; B = 21) in the female population. There was a significant correlation between the affiliation indices and both pairwise mtDNA haplotype sharing (r = 0.1; p < 0.05) and genetic relatedness based on microsatellites (r = 0.1; p < 0.05; Fig. 4).
Comparing different affiliation classes (preferred, casual and avoided), we found that preferred female affiliates had higher mean pairwise genetic relatedness than casual and avoided pairs (p < 0.05; Table 2). However, we did not detect a significant difference in the frequency of shared haplotypes for the different affiliation classes (p = 0.16; Table 2).
At social cluster level, we found a higher frequency of mtDNA sharing ( Fig. 5; Table 3; p < 0.05) and higher mean genetic relatedness (Table 3; p < 0.05) within than between social clusters. We did not find a significant correlation between genetic relatedness (r = 0.02; p = 0.8) or mtDNA haplotype sharing (r = 0.02; p = 0.3) and the home range overlap of the individuals.

Discussion
We showed that female southern Australian bottlenose dolphins in Coffin Bay form moderate to strong affiliations and social clusters in which kinship generally correlated positively with their associations. Furthermore, we found that the reproductive condition of females correlated with the strength of affiliations. The patterns observed conforms to theoretical expectations for the formation of social bonds in female dolphins inhabiting inshore environments 45 , suggesting that similar evolutionary forces may be shaping the social behaviour of dolphin populations from disparate geographic areas and different species.
The formation of non-random associations in female mammals is known for several species, such as Asian elephants (Elephas maximus), giraffes 21 , spotted hyenas 53 , grey kangaroos, (Macropus giganteus 54 ), zebras (Equus grevyi) and onagers (Equus hemionus 55 ). In Coffin Bay, female dolphins form preferred associations and social clusters that are similar to those described in some other coastal populations of bottlenose dolphins (e.g. 36,37,39 ). For example, in the Port Stephens' dolphin population in eastern Australia, females generally associated at low to moderate levels with each other, but more closely with certain females forming social clusters 39 . Moreover, these social clusters used different spatial areas within the embayment, which were characterised by different habitats 39,56 . Female dolphins in our study area showed a similar pattern, establishing non-random associations with certain individuals and forming social clusters that showed spatial preferences for particular sub-embayments of Coffin Bay. The social and spatial divisions observed corresponded well with those reported at the population level in Coffin Bay, however, in clusters C and G some females were assigned as belonging to different communities 49 . This could be potentially explained by these two clusters inhabiting the home range area of overlap between the two communities detected at population level, probably acting as connectors between the communities (e.g. 57 ). The formation of social clusters among females also support the idea of a hierarchically organized dolphin society in Coffin Bay 49 , similar to the one reported in Port Stephens 56 .
The formation of social groups in female bottlenose dolphins has been attributed to benefits leading to increased infant survival, defence against predators and male coercion, and lifetime fitness 11,13,36,39,43 . In Coffin Bay, although apparently low (Diaz-Aguirre, pers. obs.), the risk of predation could be influencing female behaviour (e.g. 58 ). Great white sharks (Carcharodon carcharias) are relatively common in South Australian waters, and have been observed within our study area, with at least one dolphin observed with severe shark injuries likely inflicted by this species (Diaz-Aguirre, pers. obs.). In addition, male dolphins in Coffin Bay form small groups, likely alliances, which could function to increase their access to females 50 . Thus females may also benefit by forming social groups to defend against sequestering or coercion attempts by males (e.g. 13 ).
In societies with fission-fusion dynamics, such as those of bottlenose dolphins (e.g. 36 ), females associations may vary according to their developmental stage and reproductive condition, which in turn is affected by ecological and social pressures (e.g. 59,60 ). In our study area, we observed that females in similar reproductive condition associated more often than those in different states. GAIs were higher between individuals in similar reproductive  Table 3. Mean GAIs and genetic relatedness, and number of pairs sharing haplotypes for each of the seven female social clusters of southern Australian bottlenose dolphins (Tursiops cf. australis) identified in Coffin Bay, South Australia. Social cluster IDs are represented as in Fig. 1. Asterisks denotes mean genetic relatedness that differed from random expectations (p < 0.05).
condition, and we also found a weak but significant correlation between the reproductive state and the strength of the affiliations. Our results are similar to those reported by Möller and Harcourt 43 , where association indices are higher among females in similar reproductive state. However, dissimilar methodologies for estimating associations between the studies prevent direct comparisons. Both studies, however, did not have the age class of the calves, so it is possible that females with calves prefer to associate with others with calves in similar developmental stage. It has been suggested that female bottlenose dolphins in similar reproductive condition may benefit by associating with each other because of similar energetic and protection requirements 37,43 . In Coffin Bay, females with calves may have different food requirements (e.g. 61 ), and be under higher pressure of predation by sharks and www.nature.com/scientificreports www.nature.com/scientificreports/ injuries inflicted by coercing males (e.g. 62,63 ), thus favouring associations among individuals in similar reproductive condition. We suggest that reproductive condition is a an important determinant of associations in our study area, however others factors such as philopatry, calf and/or mother age, social familiarity, maternal experience and kinship might also play a role in female sociality 43 .
The effect of philopatry, as home range overlap, was controlled during GAIs estimation. Despite the lack of age information for the females under study, our protocol for identifying individuals in the field, where all females included in the analysis were categorized as adults based on size, let us assume that most of the individuals were adults or about to reach sexual maturity. Thus, the potential for immature and adult animals associating together and confounding our results are unlikely. Due to the long-term nature of bottlenose dolphin associations 36 , it is possible that previously close associations may not dissipate completely if two females display   www.nature.com/scientificreports www.nature.com/scientificreports/ different reproductive conditions 43 . Thus, some females close associates may include individuals with whom they previously shared similar reproductive states in the past 43 . Similarly, it has been shown that experienced mothers associate more often with females with calves when compared with primiparous mothers, which tend to associate with nulliparous females, males or females without calves 43,64 . Long-term photo-identification data of this population and comparisons of survival rates among female offspring inhabiting different areas of Coffin Bay, as well as observations of female-male interactions, could provide important information to further test hypotheses about the role of reproductive state on female group associations.
The importance of kinship in the establishment and maintenance of social relationships has been documented in a wide variety of mammals (e.g. lions, Panthera leo 65 , elephants 19 , hyenas 22 ). Moreover, the establishment of social bonds among philopatric female kin has been suggested as one of the initial steps in the evolution of mammal sociality 66 . Similarly, a recent modelling study has suggested that simple specialization for resources among competing groups might promote the formation of stable social groups in which kin structure may develop via fitness benefits or philopatry 35 . In this study, we have demonstrated that association patterns in female southern Australian bottlenose dolphins are correlated positively with both matrilineal and bi-parental relatedness. In addition, we found higher levels of relatedness and haplotype sharing within than between female social clusters. However, when considering different affiliation classes, we only found support for higher bi-parental relatedness among preferred female pairs, not haplotype sharing. This last result may be partly affected by the low-explanatory power of the maternally inherited marker as only two haplotypes where found in the population. Moreover, the lack of correlation between home range overlap and genetic relatedness suggests that the associations are likely driven by kinship, and do not represent a by-product of the correlation between these two variables. This also suggests that some females could be dispersing from their natal areas into adjacent bays, which differs from the general pattern observed in inshore bottlenose dolphins, where females usually remain philopatric (i.e. 36,44,45,67 ). Under this scenario, kin clusters of preferred associates may relocate to different areas or embayment within Coffin Bay in response to social or ecological pressures, providing an explanation for the lack of correlation between home range overlap and genetic relatedness.
We propose that kinship, likely driven by fitness benefits, plays an important role in the formation of social preferences among female dolphins in Coffin Bay. A positive correlation between association patterns and kinship has been previously reported in cetaceans (e.g. sperm whales 20 ) and in other bottlenose dolphins populations inhabiting inshore environments (e.g. 39,40 ). In a review of delphinid socio-genetic structure, Möller 45 proposed that long-term social bonds in female dolphins occurred mainly among kin, while moderate social bonds may emerge between related and non-related individuals. Thus, the associations among non-related individuals observed in our study may reflect short-term preferences on a larger temporal scale given the long-lived nature of these animals. This could be attributed to ecological factors, such as fluctuations in prey resources, which may promote the emigration of individuals into other non-related social groups when competition is high within their core social clusters. Long-term data on the association patterns of these dolphins may provide important information to better understand the interactions between kinship, philopatry and association patterns.
Our results corroborate predictions from previous studies that in inshore habitats where resources are relatively predictable 44 , female dolphins may benefit by forming social bonds with kin and other females in similar reproductive condition, while maintaining moderate and loose social bonds with some same sex individuals 41,43,45 . The inner area of Coffin Bay is a complex inshore environment, where females exhibit a high degree of site fidelity, and range over relatively small areas, likely due to predictable food resources within the small sub-embayments 48 . Our results also provide insights into the mechanism promoting the formation of male associations based on kinship relationships in Coffin Bay 50 . Considering that the population is relatively large 47 and organised into two social communities 49 , and that females prefer to associate with kin and with others in similar reproductive condition, young male calves may encounter opportunities to associate and develop social bonds with other males that are genetically related. As males grow older these associations may become stronger and form the basis for the formation of social preferences observed in the adult male population 50 .
In summary, our findings demonstrate that kinship and reproductive condition are important factors influencing association patterns of female southern Australian bottlenose dolphins in Coffin Bay, South Australia. Long-term behavioural observations for estimating calving success and how this relates with female age, maternal experience, social network metrics, kinship relationships and areas of usage would provide important information into the functional mechanisms promoting long-lasting female associations and the formation of social clusters. Our results add to the growing knowledge which demonstrates the interplay among social, genetic, endogenous and ecological factors shaping dolphins and other complex mammalian societies.

Ethics statement/approval. This study was carried out under Flinders University Animal Welfare
Committee approval number E310 and under permits to undertake scientific research: E26171-1, E26171-2, E26171-3 and MR00056-1 from the Department of Environment, Water and Natural Resources (DEWNR), South Australia, and under S115 ministerial exemptions (MEs: 9902601, 9902660, 9902714 and 9902779) from Primary Industries Resources South Australia (PIRSA) 49 . Researchers in the field attempted to minimize disturbance to dolphin groups by following the Australian guidelines for whale and dolphin watching 68 . Briefly, dolphins were approached using a constant speed and following a parallel movement to their direction of travel. For obtaining biopsy samples from animals, we used two methods that produce minimum short-and long-term impacts on dolphins following the guidelines reported in 69  www.nature.com/scientificreports www.nature.com/scientificreports/ Study area and data collection. Periodic boat surveys were conducted in Coffin Bay's inner area (123 Km 2 ), South Australia, between March 2013 and October 2015 (Fig. 2). The study area is characterised as a shallow reverse estuary (depths ranging from 1 to 12 meters) and is composed of three different bays that differ in depth and habitat types. Mt. Dutton and Kellidie Bay are mainly dominated by shallow seagrass habitats, whereas Pt. Douglas is deeper and contains a diverse array of habitats, including tidal sandflats, temperate reefs, seagrass meadows and deeper waters with sandy bottoms 71,72 . Boat surveys were conducted in calm sea conditions (Beaufort < 3) and were designed to cover all different habitat types and seasons within the study area 47 . More information on survey design and transects followed during the study are reported in 47 .
Dolphin groups sighted were approached to record their location, composition and size, and to obtain photographs of their dorsal fins using DSLR cameras equipped with 100-300 mm and 100-400 mm zoom lenses. Individual dolphins were characterized using natural marks on their dorsal fins 73 and the best images of each individual in a group were classified in Discovery v1.2 74 . A detailed description of the protocols used for classifying photographs can be found in 47 . We classified individuals >1.5 m in length as non-calves and those ≤1.5 m and closely accompanied by a non-calf individual as calves 50 . In addition, we collected biopsy samples from photo-identified non-calf individuals using the biopsy pole system for bow-riding dolphins 69 , or the PAXARMS remote biopsy system for dolphins 70 . Samples were stored in a 20% DMSO salt saturated solution and frozen at −20 °C following recommendations 75 .
Genetic analyses. DNA extraction was performed using proteinase K digestion followed by a salting-out protocol 50,76 . Extracted DNA was amplified at 11 microsatellite loci following the conditions reported in 77 . Samples were genotyped on an Applied Biosystems 3130 DNA Analyser and allele size fragments were scored using GENEMAPPER v4.1 (Applied Biosystems). The software MICROCHECKER 78 was used to test for the potential presence of null alleles and allelic dropout. GENEPOP v4.2 79 was used to assess Hardy-Weinberg equilibrium (HWE) and linkage disequilibrium using 1,000 iterations. There is no evidence of genetic subdivision within the inner area of Coffin Bay for the southern Australian bottlenose dolphin population 50,77 .
We amplified a region of 450 base pairs (bp) of the mitochondrial DNA (mtDNA) control region using primers Dlp-5 (5′-CCA TCG WGA TGT CTT ATT TAA GRG GAA-3′) and Dlp-1.5 (5′-TCA CCC AAA GCT GRA RTT CTA-3′) 80 , following conditions detailed in 81 . PCR products were sequenced on an Applied Biosystems 3130xl genetic analyser. SEQUENCHER v5.2.4 (Gene Codes Corporation, Ann Arbor, MI, USA) was used to align the sequences resulting in a high-quality 437 bp fragment. To genetically determine the sex of each biopsied dolphin, we amplified a fragment of the ZFX and SRY genes using the protocols described in 82 .

Defining female associations and estimating generalized affiliation indices.
In this study we considered a group of dolphins as all individuals within a 100 m radius and participating in similar behavioural activities 37,50 . For subsequent analyses we only considered those groups in which a minimum of 75% of the individuals were photo-identified based on the estimated group size for each sighting 50 . We only included the first sighting of an individual in each survey day, and subsequent sightings of identical groups in the same day were excluded. All dolphins identified in the same group were considered associated, and if new individuals joined a group during a sighting these were included in the same group 50 . To reduce the potential for false null associations among individuals with low number of sightings, we only considered dolphins observed on more than 11 occasions (equals to the median number of sightings for the population) 50 . In addition, we controlled for the cumulative number of sightings of each given dolphin pair during the affiliation index estimation (see below) 50 . In the subsequent analyses, we only included non-calf individuals positively identified as females either through the molecular sexing analysis, or by visual observation of a closely associated calf on more than 10 separate days.
We used generalized affiliation indices (GAIs 51 ) to estimate the strength of associations between pairs of females. Using this method, we corrected for the correlation between association indices and structural variables known to affect social analyses (e.g. 51 ). Using the half-weight index (HWI 83 ), we constructed a matrix of associations and tested the above using controlled multiple regression quadratic assignment procedures (MRQAP): cumulative number of sightings for each dolphin pair, gregariousness (number of associates of an individual 84 ), and home range overlap 50 . Variables that showed a significant correlation with the HWI were retained and used for GAIs estimation. The controlled generalized affiliation indices were used for subsequent social analyses. We did not include reproductive condition or genetic relatedness as structural variables within the GAIs because we were interested in evaluating the influence of these variables on the strength of the affiliations (as suggested by Whitehead 51 ).
We used the package AdehabitatHR 85 in R v3.2.3 86 to estimate individual dolphins 95% utilization distributions. We used the href function to estimate the smoothing parameter (h) and subsequently visually explored individual ranges to determine the value that best adjusted to the data set. A value of h = 550 was chosen as this provided the best representation for the data 50 . Home range overlap between dolphin pairs was subsequently estimated using the kernel-based utilization distribution overlap index method 87 in AdehabitatHR 85 . SOCPROG 2.7 88 was used to calculate gregariousness, to carry MRQAP and to estimate GAIs.
Analysis of female affiliation patterns. We used two different approaches for testing preferred and avoided companionships among female dolphins in Coffin Bay. First we used the permutation method implemented by Bejder et al. 89 and modified by Whitehead 88 , using daily sampling periods to avoid demographic effects 90 . For this procedure we randomly permuted the affiliation matrices until P values stabilized using the standard deviations of the affiliation indices as test statistic 50 . In addition, we converted GAIs raw residuals into deviance residuals for identifying pairs of dolphins that demonstrated preferred, casual or avoided affiliations. These was done following recommendations by Whitehead and James 51 , who considered pairs with values above 2.5 as preferred, between 2.5 and −2.5 as casual, and below −2.5 as avoided companionships 50  www.nature.com/scientificreports www.nature.com/scientificreports/ Furthermore, we tested if females in similar reproductive condition associated more often than those in different states. We constructed a similarity matrix for females according to two categories: (1) observed with a dependent calf for >18 months and (2) observed without a calf or observed with a calf for <6 months. The time frame selected allowed us to differentiate females that were observed with calves during most of the study period from those that did not have calves or had them but for only a short period of time, either because the calf died or was weaned soon after the start of the study. Thus, (i) two females in category 1 (or two in category (2) were assigned a similarity value of 1; and (ii) two females, one in category 1 and the other in category 2 were assigned a similarity value of 0. In category 1 we included 24 females that were observed accompanied by a calf for at least 18 months. In category 2 we included two females that had calves for 4 and 5 months, and 29 females that were never observed with a calf. We tested for a correlation between the similarity matrix and the affiliation indices using a Mantel test with 10,000 permutations in SOCPROG 2.7 88 . In addition, we tested for differences in the affiliation index values with respect to the females' reproductive condition (same or different) using a Mann-Whitney test.
To examine potential social divisions among female dolphins in Coffin Bay, we used Newman's modularity matrix clustering technique [91][92][93] implemented in SOCPROG 2.7 88 . This technique attempts to divide the population into social modules that have higher affiliation indices between members of the same social group than expected by chance using an eigenvector-based method 92,93 . Values of modularity above 0.3 are considered to provide a good description of the data 92 . Social network diagrams were subsequently constructed using the spring-embedded method implemented in NETDRAW v2.1.5.5 94 to display female social groups and affiliations. In addition, to characterize the spatial distribution of the social clusters identified, we allocated the positions of the group sightings to the different social modules, and calculated representative (95%) and core (50%) ranges using the utilization distribution method implemented in AdehabitatHR 85 .

Kinship relationships and genetic relatedness.
To assess the role that kinship plays on female affiliations, we evaluated the degree of genetic relatedness at three different levels of associations. First, we tested for a correlation between bi-parentally and maternally inherited DNA markers and the strength of the affiliations (GAIs). We constructed a matrix of pairwise haplotype sharing for the mtDNA dataset and a matrix of pairwise genetic relatedness for the microsatellite dataset. Individuals with identical or different haplotypes were assigned a value of 1 or 0, respectively. Pairwise microsatellite genetic relatedness was estimated using the triadic likelihood estimator (TrioML 95 ) in COANCESTRY v 1.0.1.5 96 . To evaluate the correlation between the strength of affiliations (GAIs) and both matrices, we used Mantel tests (with 10,000 permutations) in SOCPROG 2.7 50,88 .
Second, we compared the frequencies of shared haplotypes among preferred, casual and avoided affiliates using a randomization chi-square test with 10,000 Monte Carlo permutations. In addition, mean pairwise genetic relatedness was compared to random expectations among preferred, casual and avoided affiliates using 10,000 permutations in PERM 97 .
Third, we investigated whether females identified in the same social cluster had higher mean pairwise relatedness or higher frequencies of shared haplotypes than those of different clusters. We compared mean pairwise genetic relatedness and frequencies of haplotype sharing within and between social clusters using 10,000 permutations in COANCESTRY v1.0.1.5 96 , and a randomization chi-square test with 10,000 permutations, respectively. Finally, to discard the possible association between genetic relatedness and philopatry, which could confound our kinship analysis, we tested for a correlation between the individuals' home range overlap and their genetic relatedness (maternal and bi-parental separately) using Mantel tests with 10,000 permutations in SOCPROG 2.7 88 .

Data availability
Data made available to all interested researchers upon reasonable request to Fernando Diaz-Aguirre (fernandobiologist@gmail.com).