Male alliance behaviour and mating access varies with habitat in a dolphin social network

Within-species variation in social structure has attracted interest recently because of the potential to explore phenotypic plasticity and, specifically, how demographic and ecological variation influence social structure. Populations of bottlenose dolphins (Tursiops spp.) vary in male alliance formation, from no alliances to simple pairs to, in Shark Bay, Western Australia, the most complex nested alliances known outside of humans. Examination of ecological contributions to this variation is complicated by differences among populations in other potentially explanatory traits, such as phylogenetic distance, as well as female reproductive schedules, sexual size dimorphism, and body size. Here, we report our discovery of systematic spatial variation in alliance structure, seasonal movements and access to mates within a single continuous social network in the Shark Bay population. Participation in male trios (versus pairs), the sizes of seasonal range shifts and consortship rates all decrease from north to south along the 50 km length of the study area. The southern habitat, characterised by shallow banks and channels, may be marginal relative to the open northern habitat. The discovery of variation in alliance behaviour along a spatial axis within a single population is unprecedented and demonstrates that alliance complexity has an ecological component.


Half-weight association coefficient.
At sea, we are more likely to see one of a pair of dolphins when they are apart, in two different groups, than together in one group. Under these conditions the halfweight association coefficient performs well [6], so many dolphin researchers have adopted this measure (perhaps more so now to enable comparison with earlier studies). The half-weight coefficient is defined as 2Nab/(Na + Nb), where Nab is the number of groups in which A and B are found together and Na and Nb are the total number of group sightings for A and B, respectively. This equation yields association coefficient values ranging from 0 (for two individuals that are never sighted together in groups) to 1 (for individuals that are always sighted together).

Alliance stability index.
Connor et al. [7] calculated an 'alliance stability' index for individual males within a single, large 14-member second-order alliance as follows: [1 -(number of different 1 st -order alliances/number consortships)]. This value will be zero for a male that has a different set of 1 st -order alliance partners for each consortship and will approach one (with increasing consortship sample size) for males that keep the same partners. This index correlated with individual consortship rates within the 14-member alliance and was later found to correlate with consortship rates in 12 2 nd -order-alliances ranging in size from 6-14 males [2].

Alliance complexity.
Alliances and coalitions within a group or social network are more complex than simple 'us against them' between-group conflicts because they include competition for alliance partners based on affiliative interactions, and context dependent interactions, where friends in one social context may become foes in another [8][9][10][11].
However, even within-group alliances need not be complex if, for example, they are based entirely on close kinship [11]. Indeed, one study failed to find a relationship between brain size and coalitions in primates [12]. In stable Shark Bay trios, one pair of males more often associates and synchronises their swimming than either does with the third member or 'odd male out', but on a given day the odd male out may shift [13][14]. Such complexity is not possible in pairs. Context dependent and affiliative interactions have been observed in all three levels of Shark Bay male dolphin alliances [2]. Additional alliance levels should increase the complexity of individual decisions because conflicts at one level can impact alliances at another level. For example, a dolphin trio that evicts one of their members to reduce the costs of sharing consorted females should be more vulnerable in 2 nd -order alliance contests [2,[9][10].

Non-mating season
We found that NW-SE axis distance still significantly predicted proportion of trios, consortship rate, and adjusted consortship rate during the non-mating season (July-August) ( Table S1b). Consortship rate and adjusted consortship rate both tended to decrease for 2 nd -order alliances in the non-mating season ( Fig. S2) with mean differences of 0.26 and 0.19 respectively (n = 10 in both cases). In contrast to consortship rate, proportion of trios in consortships did not show a consistent trend between the seasons (Fig. S2) with a mean difference of 0.03 (n = 12).

Low consortship rate variation along the penninsula
In the main text we report a significant relationship between NW-SE axis position and both the maximum individual and maximum consorting dyad consorthip rates for each 2 nd -order alliance. We also compared the minimum individual consortship rates for each 2 nd -order alliance and found a similar relationship with the NW-SE axis position significantly predicting consortship rate ( Fig. S3; GLM, n = 10, z = -2.16, p = 0.031). Since the lowest consortship rate in three groups were outliers (Fig. S3), we also calculated consortship rates for the 2 nd lowest consorting individual ( Fig.   S3; GLM, n = 10, z = -3.00, p = 0.027) and the 2 nd and 3 rd lowest individuals together ( Fig. S3; GLM, n = 10, z = -3.62, p < 0.001) and found similar trends. Together, the analyses of individuals with high and low consortship rates in each 2 nd -order alliance demonstrate that the NW-SE decline in consortship rate is robust and not due to a few extreme values in northern or southern groups.

Individual foraging tactics in Shark Bay
Whereas foraging cultures may vary across groups and populations of primates and cetaceans [15][16][17], in marine mammals it is also common to find different foraging tactics among individuals that occupy the same area. Some female and male dolphins in Shark Bay with extensively overlapping ranges employ different foraging tactics. Some of these individual foraging tactics are quite striking, including 'sponge-carrying', 'beaching', 'shelling' and 'kerplunking' [18][19][20][21][22][23].

Supplementary Methods
The study subjects and site.
In Shark Bay, over 1500 individually identified dolphins have been observed since the mid-1980s [reviewed in 2]. The dolphins exhibit a highly dynamic fission-fusion grouping pattern in an open social network: a mosaic of overlapping male and female ranges extends along the 50km length of the study area, but dolphins in the northern and southernmost parts of the study area do not overlap (Fig. S1). Females typically conceive first at age 11 and give birth at age 12 to a single infant that is dependent for 3-5 years. Most males are members of 2 nd -order alliances and begin consorting adult females by age 13-14. Some males and females live into their 40s.

Documenting Consortships.
Males cooperate in pairs and trios to sequester individual females for periods of minutes to weeks. Defining criteria are outlined in Connor et al. [7], including an association lasting one or more hours, or on multiple surveys spanning one or more hours, a capture of the female by the males, the female attempting to escape (bolting), or aggression by the males toward the female (head jerks, charge or the males producing the 'pop' vocalization). Aggression is observed in about half of consortships. Males almost always consort females with members of their 2 nd -order alliance; in about 3% (15 of 496 consorthips), 1 st -order alliances contained members of two 2 nd -order alliances [1]). By definition, membership is stable for the duration of the consortship. If we find the same trio with a different female the next day or even later on the same day, or there is a change in trio composition, a new consortship is recorded.

Non-mating season vs mating season consortship rates.
Our analysis of consortship rates and trio formation focused on the mating season (September-November) because that is the period during which the majority of births and conceptions occur [24]. Consortships are longer during the mating season [25] and, in this study, consortship rates were higher in the mating compared to the non-mating season. Females that are consorted and conceive during the mating season (based on birth records) are often consorted for short periods during the preceding non-mating season [25]. Thus, females consorted prior to the mating season might not be contested as much as females during the mating season. For this analysis, we decided, a priori, to focus on mating season consortship rates, as that should be the period of greatest intensity of competition for females and where a difference in consortship rates would be of greatest consequence.
Combining the periods could potentially obscure important mating season phenomena. This concern was motivated by results from other taxa showing that male mating access may change based on the proximity to female conception. For example, during the period of maximal estrus swelling period in chimpanzees (Pan troglodytes), mating and mate guarding increase [26][27][28], with the greatest degree of dominant possessive behaviour occuring in the few days of maximal fertility, so the extent to which dominant males monopolize females would be obscured by observing male mating rates over a longer period.

Supplementary Figures
Supplementary Figure S1. Ranges (Minimum convex polygon, MCP) of the 12 2 ndorder alliances and 5 'lone trios' studied from 2001-2006 (see text). Reprinted from [28], the map was created in ArcGIS. Lone trios (size 3) were excluded from analyses in this paper. Figure S2. Comparison of 2 nd -order alliance consortship rate (a), proportion of trios (b), and adjusted consortship rate (c) between the mating and non-mating seasons. Figure S3. For each of the ten 2 nd -order alliances, the (a) lowest consortship rate (CR) for an individual, (b) the 2 nd lowest consortship rate (CR), and (b) the summed consortship rate of the 2 nd and 3 rd lowest consorting individuals are plotted along the NW-SE axis. Grey lines show 95% confidence intervals.

Supplementary Tables
Supplementary Table S1a. Regression summaries for binomial generalized linear models of the proportion trios, consortship rate (CR), and adjusted consortship rate during the mating season (September to Novemeber).