Effect of social context on behaviour in anemone ﬁ sh hierarchies

Animal social groups can be organized in hierarchies where individual status determines privileges within the group, and stability is maintained through con ﬂ ict (aggression e submission) and cooperation. Aggression, submission and cooperation are not homogeneous among group members and are in ﬂ u-enced by social context and associated trade-offs. However, studies of rank-speci ﬁ c behaviours are rare which limits our understanding of these patterns. Here, we performed rank ascension experiments using 15 groups of Amphiprion clarkii , a relatively mobile anemone ﬁ sh, to assess rank-speci ﬁ c behaviour related to social context. We showed that promoted ranks increased cooperation rates compared to nonpromoted ranks to ful ﬁ l the tasks associated with their new status within the group. Group size had no effect on cooperation rates, although it did in ﬂ uence rates of submission and neutral interactions, and subordinates did not increase cooperation after group size reduction. Thus, subordinates modulate rates of cooperative behaviour according to their status, irrespective of their body size or group size. We provide evidence that behaviour is plastic and highly in ﬂ uenced by social context, which appears to be the key driver of cooperation and within-group agonistic encounters in this species. © 2023 The Authors. Published by Elsevier Ltd on behalf of The Association for the Study of Animal Behaviour. This is an open access article under the CC BY license (http://creativecommons.org/licenses/ by/4.0/).

Group living is a common phenomenon in nature; it is key for the survival of many organisms, and has been observed in many taxa such as mammals, birds, fishes and insects (Krause & Ruxton, 2002;Choe & Crespi, 1997;Koenig & Dickinson, 2016;Solomon & French, 1997;Taborsky, 1994).Hierarchical structures within social groups increase group stability by increasing cooperation and coordination among group members while reducing conflict, and they enhance the survivorship and foraging or reproductive success of individuals (Parrish & Edelstein-Keshet, 1999;Tibbetts et al., 2022).The stability of the social hierarchy is maintained through aggressiveesubmissive interactions between group members (Wong et al., 2007).Aggression in social hierarchies is used to assert dominance over other individuals or punish less cooperative group members (Ang & Manica, 2010a).To mitigate the cost of withingroup aggression, individuals may display submissive signals to appease aggressors and avoid eviction (Bergmüller & Taborsky, 2005).Cooperative behaviours such as territory defence or brood care can also be used to appease aggressors (Naef & Taborsky, 2020a, 2020b;Quiñones et al., 2016).Taborsky, Cant, and Komdeur (2021) defined cooperation as the simultaneous or consecutive performance of the same or different behaviours by two or more individuals to achieve a shared goal.Thus, individual behaviour plays an important role in the cohesion and stability of the group.
Behaviour is not homogenous among group members as it is influenced by social context such as group size or rank status (Tibbetts et al., 2022;Webster & Ward, 2011).For example, in studies of the paper wasp Polistes dominulus, aggression rates were shown to be higher for individuals with a higher rank within the hierarchy (Cant et al., 2006).Events such as a member of the group being evicted or predated on can induce changes in social context, and, by extension, changes in individual behaviour (Dijkstra et al., 2022).The ability of individuals to adapt their behaviour in response to the social information they receive is called social competence (Taborsky & Oliveira, 2012).Social competence is essential for individuals as it maximizes the benefits of their behaviour in changing social conditions (Oliveira, 2009).By studying the behavioural changes that individuals undergo after alterations in group social structure, we can better understand the dynamics that modulate group stability and the importance of behavioural plasticity in changing environments.
Anemonefishes have been a model system for studying social group dynamics and behaviour (reviewed in Rueger et al., 2021), including recent research that has shed light on rank-specific behaviour in the marine realm (Rueger et al., 2022).
Anemonefishes are protandrous hermaphrodites that live in groups of unrelated individuals occupying host sea anemones (Buston et al., 2007;Fautin, 1991;Moyer et al., 1978).Anemonefish groups have strict size-based hierarchies with the largest anemonefish being the dominant female and the second largest being the subdominant male which form a breeding pair, and all other ranks are nonreproductive subordinates which queue for a breeding position (Fricke & Fricke, 1977;Moyer et al., 1978).Individuals can only ascend in rank when a higher rank disappears, and they regulate their growth rate to maintain a specific size ratio between themselves and their adjacent dominant (Buston, 2003a(Buston, , 2003b;;Buston & Cant, 2006).Migration between groups is extremely rare (Laudet & Ravasi, 2022).Anemonefishes evict lower ranks and prevent recruitment which further regulates group size in saturated anemones (Buston, 2003a).Eviction from the group often results in mortality, as the probability of dispersing and successfully joining a different existing group or establishing a new group is low (Elliott et al., 1995;Elliott & Mariscal, 2001;Buston, 2003c).
In this study, we extended the recent research on rank-specific agonistic encounters and cooperative behaviours in anemonefishes, by using Amphiprion clarkii to investigate rank-specific behavioural plasticity.Amphiprion clarkii is larger than many other Amphiprion spp.allowing the formation of larger groups (Cleveland et al., 2011;Hattori, 2002;Verde et al., 2015).They also have high mobility and only spend short periods of time near their anemone.Amphiprion clarkii females only spent 10% of their time within 25 cm of their host anemone compared to the 50% spent by female Amphiprion perideraion (Cleveland et al., 2011).The comparatively greater mobility and body size of A. clarkii may contribute to a different social context for this species which makes it an ideal study system to expand our understanding of how social context influences individual behaviour.In the present study we assessed the effect of rank status, size ratio between ranks, rank ascension and group size on cooperation and agonistic encounters in A. clarkii groups.We tested the following hypotheses.(1) Behaviours are plastic and change with rank ascension.We predicted that after rank ascension agonistic behaviour would be more frequent (Cant et al., 2006;Dey et al., 2013) and cooperative behaviour less frequent (Cant & Field, 2001).(2) Behaviours change with changing group size.We predicted that in smaller groups agonistic behaviours would be less frequent (Cant et al., 2006) and cooperative behaviours more frequent (Field et al., 2006).

Study Site
The study was conducted from 30 September to 13 October 2018 and from 5 May to 16 May 2019 in Kimbe Bay, Papua New Guinea (5 30 0 S, 150 05 0 E).A total of 15 groups of A. clarkii were used.Group sizes ranged from four to eight fish and were associated with the anemones Stichodactyla mertensii and Heteractis crispa.During the study, only one treatment group was breeding.However, the group was on day 6 of their breeding (anemonefish breeding takes 7 days from laying to hatching; Laudet & Ravasi, 2022) and this only affected their behaviour on day 1 of the study, before experimental removal of the subordinate.

Behavioural Analysis
All groups were recorded using GoPro Hero5 action cameras which were placed in front of groups on stationary tripods.For this component of the study, we filmed each group once and recorded 12 min of footage for each group.Video footage of all groups was watched to score individual rank behaviours using BORIS (v.7.12), an open-source software (Friard & Gamba, 2016) For each rank, behaviour was scored separately using 10 min of continuous footage from the video.The minutes at the beginning and end of each video were discarded due to disturbance by divers when setting up and removing the cameras.The behaviours scored were divided into four categories: (1) aggression; (2) submission; (3) cooperation; and (4) neutral interactions (see Appendix Table A1).For aggressive and submissive behaviours, the target individual receiving that behaviour was also recorded as in such agonistic encounters there tended to be a clear winner and loser from the interaction.Cooperative behaviours were divided into defence (aggressive displays towards egg predators such as Labridae and anemone predators such as Chaetodontidae), competition (aggressive displays towards food competitors such as other Pomacentridae), cleaning (interactions with the anemone such as cleaning, massaging, biting) and parental (interactions towards the egg clutch such as fanning or mouthing the brood) (Rueger et al., 2022).Neutral interactions are defined as nonagonistic interactions between individuals that are within one body length of each other, such as swimming together, meeting, following or touching (Rueger et al., 2022).Individuals were recognized by natural markings (the unique patterns of the colour bands), which vary by individual in this genus, and size differences between ranks due to the size ratio between ranks, which allows easy identification of individuals in the hierarchy.Visual recognition and identification of individuals is widely used in research on this genus (Laudet & Ravasi, 2022).All behaviours were analysed per minute of observation (time the individual was within frame) to account for the differences between ranks in the amount of time spent in and out of frame.

Rank Ascension Experiment
We experimentally removed one nonbreeding subordinate from each group to test behavioural differences between ranks and changes in behaviour when nonbreeders ascended in rank (following Rueger et al., 2022).In the treatment groups (N ¼ 8), rank 3 or rank 4 (a large subordinate) was removed which provided an opportunity for the ranks below to ascend in rank.In the control groups (N ¼ 7), the lowest rank was removed to account for the effect of group size reduction (see Appendix Table A2).Four videos (12 min each) were recorded for each group: before removal of the subordinate (preremoval), immediately after subordinate removal (0 h), 24 h after subordinate removal and 48 h after subordinate removal.The subordinates were caught using two hand nets, carried to the surface in sealable plastic bags and placed in buckets filled with sea water for transport via boat to the laboratory facilities (Fig. 1aec).
Once at the laboratory, removed fish were held in aerated sea water tanks (51 Â 25 cm and 30 cm high) with shelter (aquarium plants, rocks, etc.).They were fed daily with dried brine shrimp (Omega One) and fish pellets (New Life Spectrum).Removed fish were reintroduced into their original groups at the end of the experiment, 2 days after their removal.All fish except one were accepted back into their groups and suffered minimal aggression from the other fish in the group.One individual was not accepted by the group and was chased out of the anemone; however, it was caught and introduced to an empty anemone on the same reef.After the observations, we measured all fish standard lengths (mm, SL) using callipers.The size ratios between adjacent ranks (focal individual SL/individual above in the hierarchy SL) were calculated (mean ratio ± SE ¼ 0.781 ± 0.015).

Data Analysis
All statistical analyses were performed in R v. 4.1.3(R Core Team, 2021).We used the lme4 package to fit generalized linear mixed models (GLMM) to the data (Bates, Maechler & Bolker, 2012).For the rank-specific behaviour component of the study, the behaviour of all ranks (from 1 to 7) in the group was measured and the rate of each behaviour was compared across the hierarchy.One group had a rank 8 but due to its small body size and the lack of other eightmember groups, rank 8 behaviour was not measured or analysed.For the analysis of aggression and submission rates related to size ratio between ranks, we analysed the data from rank 1 to rank 5, as only three of the 15 groups used had more than five ranks and ranks 6 and 7 displayed lower behaviour rates than higher ranks.For the manipulation experiment we analysed the data of ranks 3, 4 and 5 from the control and treatment groups, as they were the ranks that ascended in rank after the manipulation.We used a negative binomial distribution to account for zero-inflation in the data.Group ID was used as a random variable to account for nonindependence of individual behaviours within the same group.We used Akaike's information criterion (AIC; Sakamoto et al., 1986) for model selection and used a log likelihood test when DAIC was below 2 to select the best-fit model.For significance testing of the GLMMs, we performed log likelihood tests of the full model with the effect, against the model without the effect.The performance package was used to calculate the marginal R 2 (R 2 m ; provides the variance of the model explained by the fixed effect) and conditional R 2 (R 2 c ; provides the variance explained by both fixed and random effects) using Nakagawa's R 2 (Nakagawa et al., 2017).There was singularity in some of our models as the random effects did not explain any of the variance and thus the conditional R 2 was not applicable (R 2 c ¼ NA); however, we maintained the random effects in all models to facilitate comparison between models.We controlled for repeated measures using random factors.

Ethical Note
This study was carried out with approval from the University of Wollongong Animal Ethics Committee, permit number AE18/06.

Rank-specific Behaviour
There were significant differences between ranks in aggression, submission, cooperation and neutral interactions (Fig. 2; aggression: c 2

Rank Cooperation Levels
Interactions with the anemone ('cleaning') made up most of rank 2 (approximately 63%) and rank 3 (approximately 71%) cooperative behaviour, while rank 4 cooperative behaviour mainly consisted of defence against competitors ('competition'; approximately 57%; see Fig. 3).Rank 1 cooperative behaviour was almost equally divided between anemone care and defence against competitors (approximately 49% for both categories).

Rank Ascension
All promoted individuals modified their behaviour to accommodate to their new status within the group after removal of the individual above them in the hierarchy.
When we compared promoted rank 4 (which ascended to rank 3) with control rank 4 (which did not ascend in rank), we found that promoted rank 4 showed a significant increase in cooperation rates (c 2 1 ¼ 9.195, P ¼ 0.047, estimate ± SE ¼ 1.974 ± 0.726 cooperation behaviours/min; Fig. 4).The fixed effects explained 11% of the variance (R 2 c ¼ 0.257, R 2 m ¼ 0.11).No significant differences were found between treatment and control for rates of aggression, submission or neutral interactions (see Appendix Table A4).No significant temporal treatment effect of the rank ascension experiment was found (see Appendix Fig. A1).
When we compared promoted ranks (initially ranks 4 and 5) and the equivalent ranks in control groups (ranks 3 and 4, respectively), we found no significant differences in the rates of aggression, submission, cooperation or neutral interactions between individuals.We also calculated the size ratio between promoted and removed individuals; however, we found no effect of the size ratio in the model for any of the behaviours observed (see Appendix Table A4).The pairwise comparison of behaviour between subordinates (ranks 3e7) did not show significant differences (see Appendix Table A5).

Group Size Comparison
We found a significant effect of group size on the rates of subordinate submissive and neutral interactions; however, no effect was found on aggression or cooperation rates.For both treatment and control, combined submission rates in ranks 3 and 4 declined significantly with larger group sizes ( b) submission rates (behaviours/min) for A. clarkii control and treatment groups at each stage of the rank ascension experiment: before removal of the subordinate (preremoval), immediately after subordinate removal (0 h), 24 h after subordinate removal (24 h) and 48 h after subordinate removal (48 h).Points: median; lines: lower and upper quartiles.Outliers are not included for visual simplicity.

DISCUSSION
Our results demonstrate that social context influences rankspecific behaviours and that individuals can modify their behaviour according to changes in group social structure.We showed that after rank promotion, individual behaviour changed immediately to accommodate the new tasks and roles associated with that rank within the group, despite there still being a size difference between the promoted individual and the individual that had been removed.These findings contribute to our understanding of complex rankspecific behaviours in social hierarchies and support our initial prediction that subordinates regulate their cooperation rates based on the future fitness benefits associated with their status; however, this regulation is independent of body size and group size.
Aggression, submission and cooperation rates varied between individuals according to their social status, with aggression and cooperation rates decreasing down the hierarchy and submission rates increasing.Our results further support the findings of Rueger et al. (2022) in Amphiprion percula and A. perideraion.Subordinates mainly interacted with the anemone or displayed aggression towards food competitors which might confer indirect benefits for dominants.This might explain dominants' tolerance of them despite the lack of direct benefits of subordinate presence, as suggested by the 'pay-to-stay' hypothesis (Buston, 2004;Bergmüller & Taborsky, 2005).
Experimental rank promotion induced a behavioural change in A. clarkii individuals.After rank ascension, individuals increased their cooperation rates to the same level as equivalent ranks in the control group, despite being smaller, consistent with the findings of Rueger et al. (2022).This result suggests that behaviour is heavily influenced by social status within the hierarchy, which aligns with previous findings in other taxa (Dijkstra et al., 2022;Maruska & Fernald, 2010;Piefke et al., 2021;Williamson et al., 2017).Also, the similarities in cooperation rates between promoted ranks and their equivalent rank in the control in the short time span of the experiment, which did not allow promoted ranks to make up the difference in body size, indicates that, in higher ranked subordinates, body size is not a limiting factor in cooperation.The promoted ranks had higher cooperation rates than the ranks that were not promoted.This suggests that even if their physiological characteristics allow them to help more, lower ranks are less likely to increase cooperation rates than higher ranks as the gain from cooperation may be lower further down the hierarchy.Studies suggest that changes in social context can act as a trigger for changes in hormonal expression and, by extension, individual behaviour (Maruska & Fernald, 2010;Oliveira, 2009;Williamson et al., 2017).Thus, social context is an important factor in determining individual behaviour.
Individual removal and, consequently, group size reduction decreased overall group aggression and submission rates but had no effect on group cooperation rates.Also, no effect of group reduction was found for the average rates of aggression, submission or cooperation per number of group members.This result suggests that the removal had no effect on individual behaviour for the remaining members of the group and that the observed reduction in overall aggression and submission rates was the result of fewer encounters between individuals after fish removal.Consistent with Brouwer et al.'s (2005) study on cooperative breeding cichlids but contrary to our prediction, group size reduction did not increase cooperative behaviour of the group which suggests that subordinates did not increase cooperation rates to compensate for the removed fish.However, a significant effect of group size was found on submission rates and neutral interaction rates, the first being lower in larger groups and the latter being higher in larger groups.Studies suggest that in larger fish groups behavioural regulation enforced by dominants is less effective, allowing individuals to avoid punishment more easily than in smaller groups (Fischer et al., 2014).Ang and Manica (2010b) showed that in larger groups of angelfish, Centropyge bicolor, dominants were less effective at regulating subordinate foraging behaviour, size differences between ranks were smaller, and there was a greater spatial segregation between ranks.This might explain the significant reduction in subordinate submission rates observed in larger groups as subordinates may be able to avoid punishment and do not require submissive behaviours to appease dominants; however, further research needs to be conducted.We hypothesize that neutral interactions are more frequent in larger groups due to size competition.Neutral interactions involve individuals being within one body length of each other, sometimes even touching (see Rueger et al., 2022 andappendix in Wong et al., 2013).Mechanosensory cues, along with other sensory cues, are used by anemonefish to evaluate the size of other ranks and, therefore, are necessary to engage in growth regulation (Desrochers et al., 2020).Given that in larger groups size differences between ranks can be reduced (Ang & Manica, 2010b), these interactions might be important in monitoring size competitors and ensuring hierarchical status within the group.Thus, in anemonefish, cooperation rates are not influenced by group size and might only be defined by the likelihood of inheriting a breeding position and the cost of cooperation.Our results suggest that despite the removal of an individual triggering behavioural changes for the individuals promoted in rank, it has no effect on a group level which might contribute to maintaining group stability.

Conclusion
Our study demonstrates that social context influences individual behaviour within fish social groups and builds on recent research showing that the behaviour of individuals within social hierarchies can depend greatly on their rank status.By quantifying how behaviours change with social context, we highlight the plasticity of behaviour within individuals and how modifications to behaviour may arise according to future fitness benefits, such as the likelihood of inheriting a breeding position or a territory.This study highlights the importance of considering species-specific characteristics when examining behaviour, as slightly different ecological and social conditions may contribute to complex and nuanced behavioural variations within social groups, and across taxa.

Figure 1 .
Figure 1.Schematic of the experimental removal of a subordinate: (a) the unmanipulated group; (b) the removal process where the subordinate is caught using hand nets; (c) the lower rank subordinates' promotion within the group social hierarchy.

Figure 5 .
Figure5.(a) Aggression and (b) submission rates (behaviours/min) for A. clarkii control and treatment groups at each stage of the rank ascension experiment: before removal of the subordinate (preremoval), immediately after subordinate removal (0 h), 24 h after subordinate removal (24 h) and 48 h after subordinate removal (48 h).Points: median; lines: lower and upper quartiles.Outliers are not included for visual simplicity.

Figure 6 .
Figure 6.Comparison of (a) submission and (b) neutral interactions (behaviours/min) between subordinates (ranks 3, 4 and 5) of A. clarkii in groups of four to eight individuals.Points show the individual data per behaviour and rank and the black lines show the linear regressions.

Figure A2 .
Figure A2.Total number of (a) aggressive and (b) submissive behaviours of each rank in relation to their size ratio with the targeted rank.Ratios closer to 1 indicate that both ranks are similar in size while ratios closer to 0 indicate a large size difference between the ranks.Points show the individual data per behaviour and rank.The colour lines show the trend and the grey area shows the confidence intervals (CI).
No significant effects of group size, treatment or interactions on the behaviour of dominants were found.

Table A2
Group sizes and individuals removed form control and treatment groups

Table A4
Promoted and control rank comparison, testing the effects of treatment and size ratio between treatment and control individuals The models show: predictor variable ~fixed effect (random effect).