Social and spatial conflict drive resident aggression towards outsiders in a group-living fish

Field work and social group selection Field work took place off the southern shore of Chikonde Village, Mutondwe Island, Zambia (8°42'49.0"S 31°07'22.9"E) in October and November 2018. This field site contains a large breeding population of N. multifasciatus, located on a shell bed at a depth of 9 – 11 m. Groups typically contain 1-3 males and 0-5 females, along with numerous juveniles (Jordan et al 2016), and while there is no pronounced sexual dimorphism in coloration, males are larger than females (males 24.5 mm median standard length; females 19.0 mm median standard length; Jordan et al in review), and males are typically more aggressive than females (Jordan et al 2016). Relatedness structure within and among groups is unclear, but it has been suggested that females are the dispersing sex and males may inherit their natal territories (Kohler 1998). Ten social groups were selected while SCUBA diving, each consisting of one adult male, two adult females, and several juveniles. Top-mounted video cameras (GoPro Hero 6) were installed 55 cm above each group (Figure 1 A). After cameras were set up, an observer (JG) remained motionless from a distance of approximately two meters away from the group and made a count of the number of visible gastropod shells in each group’s territory and the home shell of each individual (the shell into which it retreated when threatened). The sex composition of each group was also determined based on their social behaviour and relative body sizes, an approach that was confirmed by dissecting fish after similar field observations in a parallel study conducted concurrently in the same population (AB & AJ personal observations). The standard length (SL) of each resident fish and total territory area (cm2) were subsequently measured in Adobe Photoshop CC from still frames of the video recordings taken by the cameras in which a ruler was placed for reference. Competition and resident response experiments We applied three experimental treatments to the ten selected social groups, using a within-groups repeated-measures design. In the “shell addition” treatment, the number of shells in the focal territory was increased by as close to 20% as possible by taking empty, available shells from the wider shell bed environment (these shells were removed again immediately after the trial). In the “shell subtraction” treatment, ~20% of the shells in the territory were removed and temporarily placed 2 m away from the focal group (these shells were returned to their original locations in the territory immediately after the trial). In the control treatment, ~20% of the visible shells in the territory were taken away and then immediately returned to their original places. These 20% shell manipulations were spatially concentrated in areas of the males’ territories where i) there was no fish’s home shell, such that the home shell of a resident fish was never disturbed during the handling process, and ii) there was a sufficient number of shells present to be taken or supplemented. Previous studies (Jordan et al. 2016) suggest that manipulation of more than this ratio of shells increases risk of territory takeover by larger heterospecifics, so a ratio of ~20% was the maximum manipulation we considered reasonable for this study. The groups were all given 24 hours between each treatment and the following observation recording. Immediately after each observation, the groups were given their next treatment and again allowed 24 hours before their subsequent observation. In each trial, a conspecific from a distant territory (at least 20 m away) was taken, along with its home shell, and placed in a transparent plexiglass cylinder (8 cm diameter). The cylinder was placed on the edge of the focal territory, within 2 cm of one of the peripheral shells, and interactions among the resident and presented fish were recorded (Figure 1 B). Both males and females were chosen to be presented, and each focal N. multifasciatus group received all three shell manipulation treatments in the presence of a presented female and also a presented male. The male presentations and the female presentations each took place in short succession to one another during the observation phases of each experimental treatment (counterbalancing for order). Thus, every social group experienced three shell manipulation treatments (one treatment per day, in randomized order), and the ensuing behavioural interactions between residents and presented fish were observed for each treatment. Trials took place between 9:00 and 14:00. The presented fish were returned to their home territories after completion of their trials and not used again in any further experimental trials. Recording was started after the placement of the cylinder containing the presentation fish. The presented fish emerged from its shell while within the cylinder 85 ± 87 seconds (mean ± s.d., range = 15 - 423) after placing it on the territory edge. These presentations elicited appreciable levels of aggression towards the presented fish but also aggression amongst the resident fish themselves, which had previously showed little or no intra-group aggression. All aggressive interactions were scored for a 10-minute period. Behaviour was scored manually using the software BORIS (Friard and Gamba 2016). Because manipulations were visually apparent, the scorer (JG) could not be blind to treatment. Behaviours were scored using the ethogram presented in Table 1 and pooled into one count of aggression. Note that frontal displays were rare in our observations, and sometimes difficult to accurately assess from the top-down field video footage, and were therefore not included in our counts of aggression. Although we scored all aggressive acts occurring amongst the resident fish, aggression by resident females towards resident males was also exceptionally rare; across all of our 10-minute trials, resident females aggressed against their males a total of 19 times, a sample insufficient to draw statistical inferences from. Furthermore, aggression between resident females was also rare, occurring only 35 times and only in seven field videos. Our statistical comparisons of within-group aggression across experimental treatments therefore focus on resident male versus female aggression. Lastly, we measured the distances between each resident fish’s home shell and the presented cylinder for each trial using the Adobe Photoshop CC. Statistical analysis All statistical analyses were conducted in R (v. 3.6.2, R Core Team 2019). To test whether shell manipulations influenced the aggression by the resident fish towards the presented fish, we fit a generalized linear mixed effects model (GLMM) assuming a quasi-Poisson error distribution with a log link function (using the ‘nbinom1’ family from the glmmTMB R package, Brooks et al. 2017). We included the counts of aggressive acts by each resident fish towards the presented fish as the response variable, as well as treatment (3-level categorical variable: control, shell addition, shell subtraction), sex of the presented fish (2-level categorical variable: male, female), and sex of the resident fish as predictor variables, along with each of their pairwise interaction terms. In addition, we included the distance between the resident fish’s home shell and the presented fish (cm, but scaled so that mean = 0, s.d. = 1) as another predictor variable along with its interaction with sex of the resident fish. Finally, we also included the order in which the shell manipulation treatments were given to account for potential order effects. We included a random intercept of fish ID nested within territory ID to account for non-independence of responses (because multiple N. multifasciatus individuals per group were repeatedly tested across treatments). As a model offset term, we included the cumulative time durations over which both the resident fish and the presented fish were outside their shells and thus had the opportunity to interact (log-transformed). We tested whether inclusion of the interaction terms significantly improved model fit based on a likelihood ratio test (LRT), and if not, we omitted them. We used the ‘emmeans’ R package (Lenth 2020) to make further comparisons using the Tukey method. Next, we tested whether resident male-to-resident female aggression varied with the sex of the presented fish. To do this, we fit a GLMM assuming a quasi-Poisson error distribution (‘nbinom1’ from glmmTMB). We included the counts of aggressive actions by the resident male towards resident females as the response variable. Treatment and sex of the presented fish were included as predictor variables, and we tested whether to include their interaction term based on a LRT (as above). We included a random intercept of female ID nested within male ID and also a model offset term to account for differing time windows when both the male and each resident female were out of their shells and thus had the opportunity to interact. Finally, we focused only on the scenario when the presented fish was female, and we tested whether resident male aggression towards his resident females was disproportionately directed towards the resident females that were currently closer to the presented female. Here, we fit a binomial GLMM using the ‘logit’ link function. We included a binary response variable indicating whether or not the attacked resident female was the closer of the two females. We also included treatment as a predictor variable as well as a random intercept of female ID nested within male ID.野外工作与社群选择 野外工作于2018年10月至11月在赞比亚穆通德韦岛奇孔德村南侧近岸海域（南纬8°42'49.0"，东经31°07'22.9"）开展。该野外调查区域的9~11米水深的贝壳床中，栖息着庞大的N. multifasciatus繁殖种群。其社群通常由1~3条雄性、0~5条雌性以及大量幼鱼组成（Jordan等，2016）；该物种体色无显著两性异形，但雄性体型大于雌性（雄性标准体长中位数为24.5 mm，雌性为19.0 mm；Jordan等，待刊），且雄性通常较雌性更具攻击性（Jordan等，2016）。社群内部及社群间的亲缘关系尚不明确，但有研究提出雌性为扩散性别，雄性可继承其出生领地（Kohler，1998）。 本研究通过水肺潜水选取10个社群，每个社群均包含1条成年雄性、2条成年雌性及若干幼鱼。在每个社群上方55 cm处安装顶置式摄像机（GoPro Hero 6）（图1A）。摄像机架设完成后，观察者（JG）在距社群约2 m处保持静止，统计每个社群领地内可见的腹足类贝壳数量，以及每只个体的庇护所贝壳（即受威胁时退避栖息的贝壳）。同时，通过社会行为与相对体型确定每个社群的性别组成，该方法的有效性已通过同期在同一种群开展的平行研究中，对野外观察后的鱼类进行解剖得到验证（AB & AJ，个人通信）。随后，借助Adobe Photoshop CC，从摄像机录制的视频帧中测量每条定居鱼类的标准体长（SL）以及总领地面积（cm²），测量时以放置的标尺作为参照。 竞争与定居者响应实验 本研究对选取的10个社群采用组内重复测量设计，设置三种实验处理： 1. **贝壳添加处理**：从更广范围的贝壳床环境中选取空贝壳，尽可能将焦点领地内的贝壳数量提升约20%，实验结束后立即将添加的贝壳移除； 2. **贝壳移除处理**：移除焦点领地内约20%的贝壳，并临时放置在距焦点社群2 m处，实验结束后立即将贝壳放回原位置； 3. **对照处理**：移除领地内约20%的可见贝壳后，立即将其放回原位置。 上述20%的贝壳操作均集中在雄性领地内满足以下两个条件的区域：① 无定居鱼类的庇护所贝壳，确保操作过程中不会干扰任何个体的庇护所；② 存在足够数量的贝壳以供移除或添加。既往研究（Jordan等，2016）表明，若贝壳操作比例超过该阈值，会增加领地被更大异种生物接管的风险，因此本研究将贝壳操作比例设定为约20%，为合理的最大操作幅度。 每个处理与后续观察记录之间间隔24小时。每次观察结束后，立即为社群施加下一种处理，且再次间隔24小时后开展后续观察。每次实验中，选取来自至少20 m外远距领地的同种个体（连同其庇护所贝壳），放入直径8 cm的透明有机玻璃圆筒中。将圆筒放置在焦点领地的边缘，距周边某一贝壳的距离不超过2 cm，随后记录定居鱼类与展示鱼类之间的互动行为（图1B）。展示个体同时包含雄性与雌性，每个焦点N. multifasciatus社群在展示雌性与展示雄性的情况下，均接受全部三种贝壳操作处理。在每种实验处理的观察阶段中，雄性展示与雌性展示依次进行，以平衡实验顺序。因此，每个社群均经历三种贝壳操作处理（每日一种，随机顺序），并记录每种处理下定居鱼类与展示鱼类的行为互动。实验均在9:00至14:00之间开展。展示鱼类在实验结束后被放回其原领地，且不再用于后续任何实验。 录像于放置载有展示鱼类的圆筒后开始。展示鱼类在圆筒被放置于领地边缘后的85±87秒（平均值±标准差，范围15~423秒）时从贝壳中钻出。此类展示会引发定居鱼类对展示鱼类的明显攻击行为，同时也会引发定居鱼类之间的攻击行为——而此前定居鱼类几乎未表现出社群内攻击行为。所有攻击互动均在10分钟内进行计分。行为计分采用BORIS软件（Friard & Gamba，2016）手动完成。由于贝壳操作视觉上可被观测到，计分者（JG）无法对处理组设盲。行为计分依据表1中的行为谱，将各类行为汇总为攻击行为计数。需注意的是，本研究观察中正面展示行为较为罕见，且从顶置野外视频中有时难以准确评估，因此未将其纳入攻击行为计数。尽管我们对定居鱼类之间的所有攻击行为进行了计分，但定居雌性对雄性的攻击行为同样极为罕见：在所有10分钟的实验中，定居雌性攻击雄性的总次数仅为19次，样本量不足以开展统计推断。此外，定居雌性之间的攻击行为也较为稀少，仅出现35次，且仅在7段野外视频中被记录。因此，本研究针对实验处理间社群内攻击行为的统计比较，聚焦于定居雄性与雌性之间的攻击行为。最后，我们通过Adobe Photoshop CC测量了每次实验中，每条定居鱼类的庇护所贝壳与展示圆筒之间的距离。 统计分析 所有统计分析均在R（v.3.6.2，R核心团队，2019）中完成。为检验贝壳操作是否影响定居鱼类对展示鱼类的攻击行为，我们构建了广义线性混合模型（GLMM），假设其服从准泊松误差分布并采用对数连接函数（使用glmmTMB R包中的"nbinom1"族，Brooks等，2017）。以每条定居鱼类对展示鱼类的攻击次数作为响应变量，以处理类型（3水平分类变量：对照、贝壳添加、贝壳移除）、展示鱼类的性别（2水平分类变量：雄性、雌性）、定居鱼类的性别作为预测变量，并纳入所有两两交互项。此外，我们将定居鱼类庇护所贝壳与展示鱼类之间的距离（cm，经标准化处理，均值为0，标准差为1）作为另一预测变量，并纳入其与定居鱼类性别的交互项。最后，我们还纳入了贝壳操作处理的实施顺序，以控制潜在的顺序效应。为校正因每个社群内多条N. multifasciatus个体在不同处理下被重复测试导致的响应变量非独立性，我们纳入了嵌套于领地ID内的鱼类ID的随机截距项。作为模型偏移项，我们纳入了定居鱼类与展示鱼类均脱离其贝壳、从而具备互动机会的累计持续时长（经对数转换）。我们基于似然比检验（LRT）检验交互项的加入是否显著提升模型拟合度，若未显著提升则将其剔除。我们使用"emmeans" R包（Lenth，2020），通过Tukey法开展后续多重比较。 接下来，我们检验定居雄性对定居雌性的攻击行为是否随展示鱼类的性别变化而变化。为此，我们构建了服从准泊松误差分布的GLMM（使用glmmTMB包中的"nbinom1"族），以定居雄性对定居雌性的攻击次数作为响应变量，以处理类型与展示鱼类的性别作为预测变量，并基于似然比检验决定是否纳入二者的交互项。我们纳入了嵌套于雄性ID内的雌性ID的随机截距项，并纳入模型偏移项以控制雄性与各定居雌性均脱离贝壳、从而具备互动机会的不同时间窗口。 最后，我们仅聚焦于展示鱼类为雌性的场景，检验定居雄性对其定居雌性的攻击行为是否不成比例地指向更靠近展示雌性的定居雌性。为此，我们构建了采用logit连接函数的二项式GLMM，以二进制响应变量（被攻击的定居雌性是否为两个雌性中距离展示雌性更近的个体）作为因变量，同时纳入处理类型作为预测变量，并纳入嵌套于雄性ID内的雌性ID的随机截距项。