Novel behavioural assays reveal sex‐specific behavioural syndromes in anemonefish

Abstract

Individual differences in behaviour and behavioural plasticity have been extensively studied in a variety of animals across the phylogenetic spectrum. Amphiprion species bring distinct insight into the topic because of their unique life history, mating system, and extraordinary degree of behavioural plasticity associated with protandrous (male‐to‐female) sex change. Several laboratory studies have begun characterizing individual differences in behaviour and behavioural plasticity in this species. The goals of this study were to expand the repertoire of behavioural assays available for Amphiprion ocellaris, establish repeatability of individual differences, identify sex differences, and explore whether individual differences in correlated behaviours can be detected consistently across experimental contexts (i.e., whether behavioural syndromes can be detected). We measured 35 behaviours across 7 behavioural assays in 9 reproductively active A. ocellaris pairs under 3 different reproductive contexts. Behaviours were repeatedly measured three separate times (rounds) over repeated spawning cycles. We found that 33 out of 35 behaviours were significantly individually repeatable across reproductive contexts and rounds. We found parental care, large intruder aggression, and female‐oriented aggression assays produced the largest sex differences. Males performed 7‐fold more egg care behaviours than females, whereas females performed significantly more aggressive behaviours toward a large heterospecific intruder (Dascyllus trimaculatus). Further, females displayed significantly more direct aggression toward a stimulus female than males. Three different behavioural syndromes were observed in males and none in females. These results expand our understanding of sex differences in behaviour and the division of labour in the iconic anemonefish. Future studies can use these assays to study the behaviour of fish in the middle of sex change or in the study of behavioural plasticity in this unique species.

1. INTRODUCTION

In nature, vertebrate animals regularly display complex goal‐directed behaviours that involve a multitude of decisions in response to dynamic environmental changes throughout their lives. Genetic and environmental factors interact to bias behaviours toward one version (strategy) or another, making some animals prone to particular combinations of behaviours or personalities (Bell, 2007; Sih et al., 2004). Social animals have the added complexity that their behaviours and decisions often depend on the other animals in the group. For example, when multiple individuals form a reproductive unit, often there is a division of labour with respect to territory defence and care of the offspring (Itzkowitz, 1984; Kreider et al., 2022; Rogers, 1988).

Plasticity, defined herein as the ability of the animal to change its behaviour in response to the environment, is an extraordinary evolutionary adaptation that enables life. At some level plasticity is constrained by the unique personalities that individuals display determined by their prior experiences and genetics. Individual differences in behaviour and behavioural plasticity have been extensively studied in a variety of animals across the phylogenetic spectrum (Giray & Robinson, 1994; Maruska & Fernald, 2010; Milewski et al., 2022; Relyea, 2001). The anemonefish family Amphiprion brings distinct insight into the topic because of its unique life history, mating system, and extraordinary degree of behavioural plasticity. Anemonefish undergo protandrous (male‐to‐female) sex change (Fricke & Fricke, 1977; Liu et al., 2017). In the wild, Amphiprion ocellaris live in small, isolated groups comprising one reproductively active α female, one reproductively active β male, and potentially several smaller non‐reproductive γ subordinates. Should the α female be killed or removed, the β male takes its spot in the dominance hierarchy and undergoes protandrous sex change to become the new α female, whereas the most dominant γ ascends to become the β male, and all subsequent fish move up one position in the dominance hierarchy (Fricke & Fricke, 1977; Moyer & Nakazono, 1978). All members of the group behave in distinct ways based on their position in the dominance hierarchy (Rueger et al., 2022). In particular, there is a division of labour where the reproductively active β male is the primary caretaker of the eggs and the female the primary defender of the territory (DeAngelis & Rhodes, 2016). Behavioural sex differences have added significance for anemonefish because the same individual fish with the same genotype first behaves as a male and then behaves as a female after changing sex. Thus, behavioural sex differences in anemonefish represent extraordinary instances of plasticity. The sequence of events that begins in the brain when the male senses it is in the α position, to the changes in the brain that control gonadal composition and function, and all the way to the changes in behaviour that are typical of males and females is the central topic of the recent investigation in our lab. Thus, one of the goals of this study was to establish sex differences in behaviours that could be used as a tool to measure the progress of sex change in terms of behavioural tendencies of the sex‐changing fish toward the female from the male phenotype (Parker et al., 2022).

Anemonefish display extremely high levels of territorial defence, likely related to their dependence on their host sea anemone for survival (Da Silva & Nedosyko, 2016). Typically, this involves aggression toward interspecific intruders such as Domino damselfish (Dascyllus trimaculatus) that commonly reside near the anemone, hermit crabs (Coenbita spp.), and other small fish, and potential threats to their anemone host, such as crabs or starfish that prey on sea anemones (Annett & Pierotti, 1984; Imafuku et al., 2000). As the largest and behaviourally dominant member of the group, females are the most aggressive member of the pair (Iwata & Manbo, 2013). For example, female anemonefish (not males) have been known to attack much larger foes, including human divers and sharks that stray too close to their host sea anemone (Hoepner et al., 2022; Iwata & Manbo, 2013). One explanation for why females may have evolved to be the larger and more aggressive of the two sexes in anemonefish is that the size of the female limits the number of eggs that can be laid, whereas sperm number is not a limiting factor for male fertility, and there is no male–male competition due to the monogamous mating system (Hattori, 2012). Having the larger member of the pair be the more aggressive one has obvious advantages toward larger interspecific intruders.

Anemonefish's dependence on their sea anemone also likely unintentionally contributed to their suitability for laboratory research. In the wild, anemonefish rarely stray more than a few meters away from the safety of their sea anemone; thus, they are comfortable in the confines of small (~60 L) aquaria (Donelson et al., 2022). The environmental distinction between the lab and the wild is considerable for all marine fish kept in captivity. Laboratory housed anemonefish (typically and in our study) do not host a live anemone, do not normally face non‐conspecifics, are not under predation risk, live under lower water flow conditions, have a different diet, and habituate to humans and regular feeding schedules. Nevertheless, in the lab, anemonefish readily reproduce and display normal parental care and nest behaviours even in the absence of a sea anemone (DeAngelis et al., 2017, 2020; Laudet & Ravasi, 2022; Phillips et al., 2020). Thus, unlike many other marine fish with larger home range sizes, it is possible to obtain meaningful results with laboratory experiments in anemonefish with respect to parental care, territorial defence, and nest behaviours.

Recent studies have begun to characterize behavioural variation in A. ocellaris in the laboratory. Studies have established sex differences in parental care, aggression toward an intruder male and female, and aggression toward an interspecific egg predator (DeAngelis et al., 2017, 2018, 2020; DeAngelis & Rhodes, 2016; Parker et al., 2022). Although it is well established that anemonefish display a division of labour where males are the primary care takers of the eggs and females the major defender of the nest, there is substantial individual variation; both male and females care for the eggs and defend the territory, it is just a matter of degree (DeAngelis et al., 2017, 2018, 2020; DeAngelis & Rhodes, 2016; Parker et al., 2022). Thus, there remains an important question as to whether an individual fish with their distinct pattern of behaviour along this continuum is complemented by the other member of the reproductive pair. In other words, whether personalities develop together as a pair and work as a reproductive unit rather than two independent personalities. We hypothesize that the behaviours of the fish will be coordinated such that if the female is on the low end of territorial aggression, its male partner will be on the high end, and vice versa for parental care.

As aggression and parental care are energetically costly behaviours, natural selection might have favoured plastic titration of the levels of these behaviours depending on the perceived value of the resource (e.g., territory, brood of eggs) (Trivers, 2017). For example, if the individual does not currently have eggs present in the nest, it might display lower levels of nest‐defence because the value of the nest is lower; that is, animals should expend more energy defending a nest filled with eggs than an empty nest (Montgomerie & Weatherhead, 1988). On the contrary, without eggs, they might have more time and energy for nest defence, and so they might behave in the opposite way and display higher levels of nest defence without eggs.

Similarly, kin selection predicts that animals should provide greater care and nest defence when the eggs are their own compared to another pair's (Eberhard, 1975). However, we have previously shown that anemonefish will care for eggs that are not kin with similar vigour as they would their own eggs (Phillips et al., 2020). This is a relatively unusual feature as most fish will eat eggs that are not their own (Pereira et al., 2017). It is not clear why this unusual behaviour occurs in anemonefish. One possibility is that reproductive pairs of anemonefish in the wild rarely encounter eggs that are not their own and have thus lost the ability to distinguish kin versus non‐kin due to lack of selection though this is mere speculation. Although anemonefish appear to display similar levels of parental care for surrogate eggs as their own, it remains unknown whether they behave differently in other contexts such as defending the territory depending on whether they have their own eggs or surrogate eggs in the nest.

In this study we analysed multiple different behaviours in A. ocellaris in the laboratory under a few different conditions, when the animals were caring for their own eggs, versus no eggs, versus surrogate eggs. The purpose was to evaluate a few different hypotheses related to plasticity, division of labour, and coordination of behaviours in the unique model organism. First, we hypothesized that we would be able to detect repeatable individual differences in correlated suites of behavioural traits related to aggression and parenting across environmental contexts, that is, behavioural syndromes (Bell et al., 2009). Behavioural syndromes have been described for A. ocellaris pairs in the laboratory before (Wong et al., 2013), and for related species of anemonefish (Barbasch & Buston, 2018; Schmiege et al., 2016; Wong et al., 2017). Here, we extend this work for A. ocellaris by broadening the types of behaviours, including new assays to measure territorial defence in response to a large interspecific intruder (Domino damselfish), small intruder (hermit crab), and nest maintenance (clearing debris from the nest), along with parental care and heterospecific aggression measures more common in the literature (DeAngelis et al., 2017, 2020; DeAngelis & Rhodes, 2016; Parker et al., 2022; Yaeger et al., 2014). Second, we hypothesized that levels of parenting and territorial aggression would be negatively correlated within an individual, and that members within a pair would show opposing patterns of these behaviours. Third, we hypothesized that the pair would display greater territorial aggression when eggs were present, but that their behaviours would not be influenced by whether the eggs were their own or surrogate, because of the reasons described above. Our hypotheses were only partially supported by the data. Our study is important not only for documenting the behavioural dynamics of an iconic species of fish but also useful for providing a survey of reliable behavioural assays that capture the division of labour between the sexes, for future use in mechanistic studies (DeAngelis et al., 2018, 2020; DeAngelis & Rhodes, 2016; Iwata & Manbo, 2013; Laudet & Ravasi, 2022).

2. METHODS

2.1. Animals and husbandry

Anemonefish used in this study were either first or second generation bred in‐house from fish originally obtained from Ocean Reefs and Aquariums (Fort Pierce, FL, USA), obtained as a gift from Peter Buston, or purchased from the pet trade. Fish were housed in 20‐gallon tall (24″ × 12″ × 16″) or 25‐gallon cube (18″ × 18″ × 18″) centrally filtered aquariums with a 6″ terra cotta pot serving as a nest site and spawning substrate.

The Domino damselfish (D. trimaculatus) and hermit crabs (Coenbita spp.) used in this study were obtained from the pet trade and were housed together in 10‐gallon (20″ × 10″ × 12″) aquariums consisting of three juvenile damselfish and one hermit crab. Body lengths of the damselfish were 3.5–4.5 cm. Lengths of the hermit crab shells were 2–2.5 cm. Filtration was provided by sponge filters with reef rocks and terra cotta pots serving as cover. Domino damselfish are commonly found around the territories of A. ocellaris in the wild (Hayashi et al., 2020), are potential egg predators, and are continuously chased away from the territory by anemonefish (JSR, personal observation). Similarly, hermit crabs are potential egg predators and are chased away from the territory by anemonefish in the wild.

Artificial seawater was created using Reef crystals salt and reverse osmosis deionized water. In all tanks, salinity was kept at 35 ppt, temperature was kept between 26°and 28°C, pH was kept between 8.0 and 8.4, and lighting was provided by fluorescent and LED lights over tanks on a 12:12 light schedule (lights on at 07:00 h and off at 19:00 h). Each tank was part of a circulating filtration system and contained one spawning pair of A. ocellaris (one male and one female) and one terra cotta pot. Fish were fed twice daily with ReefNutrition TDO Chroma Boost pellets (Reed Mariculture Inc., Campbell, CA, USA).

2.2. Ethical note

None of the species used in this study are protected or endangered. All experimental procedures were approved by the University of Illinois Institutional Animal Care and Use Committee.

2.3. Behavioural assays

We performed seven behavioural assays to study the behaviours of nine reproductively active pairs (n = 18, 9 males and 9 females; see Supplementary Excel File 1 for weights and lengths of animals). All assays were recorded between 15:00 and 17:00 h to control for possible effects of the circadian rhythm on the behaviours. For the purposes of this study, the term focal animal signifies the fish whose behaviour was recorded, and the term stimulus signifies the animal or object that was introduced into the tank, which varied across assays (see below). Both members of the pair were recorded together for each assay, and each individual's behaviours were counted separately. After recording began, the researcher left the area to avoid influencing the behaviour of the animals. We chose to use the same assay order across all conditions to keep possible order effects consistent in our analyses. Across all egg conditions and rounds (see below), the behavioural assays were performed in the following order:

1. Parental Care: We used a variant of the parental care assay from Phillips et al. (2020), Parker et al. (2022), DeAngelis et al. (2020), DeAngelis et al. (2018), and DeAngelis and Rhodes (2016). Males and females were recorded for 5 minutes with eggs in the nest, and for each individual, the following parameters were measured: nips, pectoral fans, tail fans, nest entrances, and time in the nest (see Tables 1 and 2 for descriptions of behaviours). In the no eggs condition, this assay was omitted.

2. Small Intruder Aggression: The day following the parental care assay, we tested the focal animals' reaction to a small intruder by placing a hermit crab near the nest. A total of four hermit crabs were used in this assay and rotated randomly across the various trials. To control movement around the tank, we fashioned a ‘leash’ out of a centrifuge tube glued to the crab's shell with a small stretch of fishing line connecting it to another centrifuge tube that was placed under the terra cotta pot. This prevented the crab from moving more than approximately a 10 cm radius away from the nest. We measured the latency to the first recorded behaviour, the number of bites, and strikes over the 10‐minute period.

3. Large Intruder Aggression: The day after the small intruder aggression assay, we tested the focal animals' reaction to a large intruder in the tank. Domino damselfish (D. trimaculatus) were selected as they are commonly found in close proximity with A. ocellaris in nature and are often chased away by the anemonefish pair (Hayashi et al., 2020). This assay was modified from the assay used in DeAngelis et al. (2020). Here, we used only one damselfish as opposed to the three used in the cited study to allow for more targeted aggression. The stimulus animal was placed into the tank for 10 minutes, and the following behaviours were recorded: latency to first recorded behaviour; the number of bites, strikes, displays; and chases/charges. A total of nine individual damselfish were sourced for this assay. For each trial, one damselfish was randomly selected from a stock tank containing three fish. The stock tanks were rotated among the three available tanks for each trial to promote balanced use. Damselfish were 3.5 to 4.5 cm in length.

4. Male‐Oriented Aggression: The day following the large intruder assay, we tested the focal animals' reaction to a novel conspecific male. All males used as stimulus fish in this assay were reproductively active males. This assay, along with the female‐oriented aggression assay, was modified from the assays used in Parker et al. (2022) in two relevant ways. First, in this study, the focal pair was kept together, and second, the assay took place in their home tank as opposed to a novel tank. Due to high levels of aggression from some fish, the male was placed in the tank only for 2 minutes as placing it any longer was deemed to be a risk to the stimulus animal's health. The following behaviours were recorded: latency to first recorded behaviour; number of bites, strikes, displays; and chases/charges. A total of 29 different stimulus males were used in this assay and rotated randomly across the different trials (see Supplementary Excel File 1 for weights and lengths of males).

5. Female‐Oriented Aggression: The day after the male‐oriented aggression assay, we tested the focal animals' reaction to a conspecific female. Only females that had previously spawned were used, and we recorded the stimulus females' relative size to the focal female to use as a covariate. For similar reasons as the male‐oriented aggression assay, the stimulus animal was placed in the tank only for 2 minutes. The following behaviours were recorded: latency to the first recorded behaviour; the number of bites, strikes, displays; chases/charges. A total of 28 different stimulus females were used in this assay and rotated randomly across the different trials (see Supplementary Excel File 1 for weights and lengths of females).

6. Immediate Reaction to a Threat: The day after the female‐oriented aggression assay, we tested the focal animals' immediate reaction to a large threat, a proxy for boldness. To test this, a researcher placed their hand in the aquarium, and behaviours were recorded for 10 seconds. The following behaviours were recorded: latency to first recorded behaviour; number of bites, strikes, displays; and chases/charges. Note, due to the rarity of bites and strikes (8 instances for each behaviour across 81 videos), we omitted these behaviours from further analysis.

7. Nest Maintenance: The same day as the immediate reaction to a threat assay (at least 10 minutes after), we tested how the focal pair would respond to a nest contaminant. We placed a small lava rock (~ 4 cm) at the back of the terra cotta pot and recorded the following behaviours for 5 min: latency to the first recorded behaviour and number of bites and strikes.

TABLE 1.

Descriptions of parental behaviours.

Behavior	Description
Time in nest	More than half of the focal animal's body is inside the terra cotta pot
Entrance	The fish swims into the terra cotta pot.
Pectoral fan	The focal animal waves its pectoral fins at the eggs.
Tail fan	The focal animal ‘wiggles’ its tail fin at the eggs.
Nip	The focal animal mouths/ bites at the eggs.

TABLE 2.

Descriptions of aggressive behaviours.

Behavior	Description
Latency	The time from the start of the video to the first recorded behaviour. If no behaviour was observed, the length of the assay was recorded as the latency.
Chase/charge	The focal animal swims quickly at the stimulus animal.
Display	The focal animal shows its side to the stimulus animal. This may include floating at an angle, tilting its body, or flexing its body.
Strike	The focal animal hits the stimulus animal with its tail or rams the stimulus animal with its head.
Bite	The focal animal bites the stimulus animal.

2.4. Egg conditions

To determine how the various behaviours measured in the behavioural assays were impacted by the presence of eggs in the nest, we repeated all the behavioural assays across three egg conditions. Assays were recorded when the focal pair had their own eggs (Own), surrogate eggs of a similar age from a different pair (Surrogate), and no eggs (None/No). In the no eggs condition, the parental care assay was omitted. To control for hormonal changes across the breeding cycle, surrogate eggs were exchanged with the pair's own eggs the day after spawning such that the pair would be in the same position in its breeding cycle in both conditions (DeAngelis & Rhodes, 2016). When exchanging surrogate eggs for the pair's own eggs, we exchanged the terra cotta pots containing the pair's own eggs with the surrogate eggs and allowed 24 h for acclimation to the novel terra cotta pot before behavioural recordings. In conditions with eggs in the nest, we recorded the number of eggs present in the terra cotta pot prior to the parental care assay (2–3 days post fertilization), and following the immediate reaction to a threat assay (7–8 days post fertilization). The no‐eggs condition began the day after all eggs had naturally hatched or died, and if the pair spawned again before all assays could be completed, the assays were completed in the next inter‐spawn period.

2.5. Rounds

To measure the repeatability of the behavioural assays, we repeated all the assays in each egg condition thrice during three successive spawning cycles. Within a round, the own eggs condition was typically performed first when the pair was caring for a batch of their own eggs, followed by the no‐eggs condition in between spawning cycles, followed by the surrogate eggs condition during the subsequent spawn by the pair when their eggs were switched with eggs from a different pair as described above. The average interval between rounds was 48.5 days ± 21.6 SD for the nine pairs used in this study over the course of the study.

2.6. Behavioural analysis

All assays were recorded using Samsung HMX‐W300 cameras. Twenty‐four hours prior to the first assay until the end of the last behavioural assay, the camera and tripod were set up in front of the tank to acclimatize the pairs to recording. Videos were analysed using BORIS (version 8.22) to record and timestamp the behaviours as the video progressed (Friard & Gamba, 2016). The timer for each analysis began after the researcher's hand left the tank or the stimulus was added and ended after the allotted amount of time had passed. Behaviours were only recorded in this time window. The terms sum of parental behaviours, sum of aggressive behaviours, or sum of behaviours indicate the sum of measured behaviours in each assay not including latency. In the parental care assay, nest entrances and time in nest do not contribute toward this sum.

2.7. Statistical analyses

R (4.2.3) was used for statistical analyses with a p < 0.05 considered statistically significant (R Core Team, 2024).

2.7.1. Linear regression

The lmer() function from the package ‘lme4’ was used to fit linear mixed‐effects models to the behaviours in each assay (Bates et al., 2024). We modelled both the sum of behaviours recorded in the assay and the count of each individual behaviour within each assay. The experimental design determined the general framework of the models. Fixed effects included round, egg condition, and sex, with sex by round and sex by egg condition interactions. Note that round and egg condition are within‐subjects fixed effects, whereas sex is between‐subjects. The fish ID was entered as a random effect to account for repeated measures. In several analyses we included other covariates as predictor variables in the models based on the nature of the assay and hypothesized interactions. For example, in the female‐oriented aggression assay, we included the difference in size between the focal female and the stimulus female, reasoning that the size differential could affect the behavioural response of the focal fish. We evaluated a similar covariate for the male‐oriented aggression, the difference in size between the focal male and the stimulus male, but because it was not significant in any of the analyses, it was omitted from the final analyses. In analyses where we compared own eggs to surrogate eggs groups, we included the age of the eggs and the number of eggs in the nest as covariates, to remove the effect of this possible confounding variable. Residuals were modelled using a Gaussian distribution. In instances where residuals were not normally distributed, we transformed data with an appropriate transformation so that skewness of residuals was between −1 and 1 and kurtosis between −2 and 2 (see Supplementary Excel Files). In cases where we log transformed data, we added 1 to all values to avoid taking the log of zero, which is undefined. Whenever we refer to fold, times, or percentage changes throughout the manuscript, the calculation is done on raw untransformed values of recorded behaviours. To extract p‐values for individual factors with multiple levels (e.g., round, egg condition) from the mixed models, we used the Anova() function from the package ‘car’ with the type = ‘III’ option specified (Fox et al., 2023). Pair‐wise comparisons were computed using the emmeans() functions from the package ‘emmeans’ with the Tukey p‐value adjustment specified (Lenth et al., 2024). Partial eta squared (${\mathrm{\eta }}_{\mathrm{p}}^2$) for mixed effects models and eta squared (η²) for linear regression models were calculated using the eta_squared() function from the package ‘effectsize’ (Ben‐Shachar et al., 2020).

2.7.2. Aggression ratio

As in Parker et al. (2022), a female‐to‐male aggression ratio was calculated. The sum of aggressive behaviours in the female‐oriented aggression assay was divided by the sum of aggressive behaviours in the male‐oriented aggression. We added a value of 1 to both of these values to avoid zeros in the denominator. The female/male aggression ratio was then log10 transformed due to non‐normal distribution of residuals. This results in a measure centred around 0 where positive values indicate a female aggression bias, and negative values indicate a male aggression bias.

2.7.3. Repeatability

To estimate the repeatability of each behavioural measure, we calculated the intra‐class correlation (ICC). To do this, we divided the estimate of the individual variance from the mixed effects linear model (that included round, egg condition, sex, sex by round, and sex by egg condition interactions as factors; see above) by the sum of residual variance + individual variance. To calculate the repeatability within sex, we calculated the ICC in the same manner as described above from mixed effects linear models that were fit only to male or only to female behaviours and thus did not have a term for sex or any interactions. To calculate the significance of repeatability estimates, we used the function ranova() from the package ‘lmerTest’.

2.7.4. Principal component analysis

The princomp() function was used to perform principal component analyses (PCA). Two types of analyses were performed. The first analysed the behaviour of each fish across all seven assays by using the sum of behaviours collapsed across the egg conditions and rounds for each behavioural assay as input variables. The second analysed the behavioural composition within each assay by using each of the recorded behaviours as input variables. The cor = TRUE option was specified to correct for differences in the scale of the measurements within and between assays. For example, in the analysis that included all behavioural assays, some of the assays were sums of behaviours collected over 10 s (e.g., immediate reaction to a threat), whereas others occurred over 5 min (e.g., nest maintenance). Sex differences in PC1 and PC2 were evaluated using unpaired t‐tests. We also compared the Euclidean distance between the mean male and mean female for principal components 1 and 2 across assays to identify assays that show the largest sex differences.

2.7.5. Pair dynamics

To determine whether the male–female pairs coordinated their behaviours toward the introduced stimuli, we analysed the male and female data separately using a similar linear mixed effects models described above (that included round and egg condition), except for the key difference that we also included the opposite sex's behaviour as a covariate in the analyses. For example, analysis of female displays had male displays as a covariate toward a large intruder. Likewise, analysis of male displays had female displays as a covariate. We interpreted a significant effect of the covariate to indicate that the behaviours of the pairs were correlated, with the sign of the effect indicating whether the behaviours were positively or negatively correlated.

2.7.6. Behavioural syndromes

To assess the presence of behavioural syndromes, we fit Pearson's correlations in a pair‐wise manner between all seven behavioural assays. For this analysis, we used the mean of the sum of behaviours collapsed across round and egg condition to represent individual values for each assay. This resulted in a total of 21 comparisons. To correct for multiple comparisons we considered the correlation to be significant if it had a q‐value less than 0.05 using the false discovery method (Benjamini & Hochberg, 1995). Because we observed significant effects of sex when analysing the composite behaviours separately (see above), we assessed behavioural syndromes for males and females separately.

3. RESULTS

3.1. Parental care: Males contribute significantly more time and effort toward parental care than females

We began our experiment by using an assay that is well known to produce large and significant differences between males and females: a parental care assay (DeAngelis et al., 2017, 2018; DeAngelis & Rhodes, 2016; Parker et al., 2022; Phillips et al., 2020). As expected, the sum of parental behaviours was highly sexually dimorphic (square root transformed, χ² ₁ = 18.060, p < 0.001, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.725; Figure 1a), as well as time spent in the nest (no transformation, χ² ₁ = 9.770, p < 0.001, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.619; Figure 1b). Males spent over three times as much time in the nest and produced over seven times as many parental care behaviours compared to females. Considering each behaviour individually, males performed 4.5‐fold more nips (square root transformed, χ² ₁ = 8.319, p = 0.004, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.588; supplementary Figure 1A), and 10.3‐fold more pectoral fans (square root transformed, χ² ₁ = 19.943, p < 0.001, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.737; supplementary Figure 1B), than females; however, the sexes did not differ in tail fans (p > 0.05), likely due to their rarity. As in Phillips et al. (2020), we found that pairs behaved similarly regardless of whether they were caring for their own eggs or surrogate eggs (all p > 0.05; see Supplementary Excel File 2 for ANOVA tables, and 3 for means and standard errors).

FIGURE 1.

Males are the primary caregivers of the eggs. All data points, including repeated measures across the two egg conditions and three rounds, are shown. (a) The square root of the sum of parental behaviours by sex. Males performed approximately 7‐fold more parental behaviours than females. (b) The amount of time spent in the nest by sex. Males spent approximately 3‐fold more time in the nest than females. (c) The square root of the number of nest entrances by sex. Males entered the nest slightly more than females, but not significantly. (d) The square root of entrances plotted against time spent in nest, split by sex. There was a significant entrances‐by‐time interaction (p < 0.001); males show a negative correlation, whereas females show a positive correlation. Lines represent linear regression estimates, and ribbons represent the ±95% CI from the mixed model. In (a‐c), error bars show the standard error estimate from the linear mixed model that represents individual variance centred around the mean estimate from the model, ***p < 0.001, and ‘NS’ indicates not significant.

In addition to parental care behaviours and time in the nest, we measured the number of nest entrances. Although the sexes did not differ in the number of nest entrances (p > 0.05, Figure 1c), the relationship between time spent in the nest and number of nest entrances was opposite for males compared to females (i.e., the interaction between sex and time in the nest for square root transformed number of entrances was significant; square root transformed, ${\mathrm{\chi }}_2^2$ = 14.564, p < 0.001, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.200; Figure 1d). Males spent less time in the nest as nest entrances increased, whereas females spent more time in the nest as nest entrances increased. This reverse pattern is likely a reflection of the division of labour regarding parental care and territorial defence. Males spend most of their time caring for eggs and only rarely leave the nest to defend the territory or explore the environment, whereas females spend the majority of their time out of the nest exploring the environment. Thus, in males, an increase in nest entrances indicates increased time out of the nest relative to other males, whereas in females, an increase in nest entrances indicates increased time in the nest compared to other females (see Supplementary Excel File 2 for ANOVA tables, 3 for means and standard errors, and 4 for pair‐wise comparisons). When the sexes were considered together, all recorded parental care behaviours were significantly repeatable (see Table 3 for ICC and p‐values). Interestingly, when the repeatability of males and females were considered separately, males were significantly repeatable in all six recorded behaviours, whereas females were only significantly repeatable in three out of the six recorded behaviours (see Supplementary Excel File 5 for values calculated separately by sex). This is likely a reflection of the high degree of variability in parental care within females.

TABLE 3.

ICC and p‐value for the random effect for all measured behaviours.

Assay	Behavioural measures	ICC	p‐Value
Parental care	Sum of parental behaviours	0.366	<0.001
Parental care	Nips	0.407	<0.001
Parental care	Pectoral fans	0.319	<0.001
Parental care	Tail fans	0.27	<0.001
Parental care	Time in the nest	0.651	<0.001
Parental care	Nest entrances	0.484	<0.001
Small intruder	Sum of aggressive behaviours	0.664	<0.001
Small intruder	Latency	0.458	<0.001
Small intruder	Strikes	0.662	<0.001
Small intruder	Bites	0.462	<0.001
Large intruder	Sum of aggressive behaviours	0.605	<0.001
Large intruder	Latency	0.211	<0.001
Large intruder	Displays	0.808	<0.001
Large intruder	Chases/charges	0.24	<0.001
Large intruder	Strikes	0.181	<0.001
Large intruder	Bites	0.136	0.008
Female‐oriented aggression	Sum of aggressive behaviours	0.22	<0.001
Female‐oriented aggression	Latency	0.094	0.06
Female‐oriented aggression	Displays	0.331	<0.001
Female‐oriented aggression	Chases/charges	0.275	<0.001
Female‐oriented aggression	Strikes	0.3	<0.001
Female‐oriented aggression	Bites	0.169	0.001
Male‐oriented aggression	Sum of aggressive behaviours	0.403	<0.001
Male‐oriented aggression	Latency	0.069	0.137
Male‐oriented aggression	Displays	0.273	<0.001
Male‐oriented aggression	Chases/charges	0.238	<0.001
Male‐oriented aggression	Strikes	0.25	<0.001
Male‐oriented aggression	Bites	0.339	<0.001
Immediate reaction to a threat	Sum of aggressive behaviours	0.471	<0.001
Immediate reaction to a threat	Latency	0.322	<0.001
Immediate reaction to a threat	Displays	0.385	<0.001
Immediate reaction to a threat	Chases/charges	0.251	<0.001
Nest maintenance	Sum of behaviours	0.34	<0.001
Nest maintenance	Latency	0.375	<0.001
Nest maintenance	Strikes	0.5	<0.001

Note: Bold indicates statistical significance value (p < 0.05).

3.2. Small intruder aggression: Females are more aggressive than males toward a small intruder

A hermit crab was placed near the nest to measure the pairs' response to a small intruder. Females produced 1.5‐fold more strikes than males (square root transformed, χ² ₁ = 4.879, p = 0.027, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.149; Figure 2a; see Supplementary Excel file 6 for ANOVA tables and 7 for means and standard errors). No significant effects of egg condition, round, or any interactions were detected. None of the other measured behaviours in this assay showed significant effects of sex, egg condition, round, or interactions (All p > 0.05). All recorded small intruder behaviours were significantly repeatable (see Table 3 for ICC and p‐values, and Supplementary Excel File 5 for values calculated separately by sex).

FIGURE 2.

Females are more aggressive toward interspecific intruders than males. All data points, including repeated measures, are shown. (a) The square root of strikes in the small intruder aggression assay is shown by sex. Females struck the hermit crab approximately 1.5 times more than males. (b) The square root of the sum of aggressive behaviours toward the stimulus animal in the large intruder aggression assay is shown by sex. Females displayed approximately 3.5‐fold more aggressive behaviours toward the damselfish intruder than males. (c) The square root of the sum of aggressive behaviours in the large intruder aggression assay is shown by round and sex. Pairs got more aggressive as rounds progressed. (d) The log of latency to address the stimulus animal in the large intruder aggression by round and sex. Pairs addressed the intruder more quickly as rounds progressed. Error bars show the standard error estimate from the linear mixed model that represents individual variance centred around the mean estimate from the model, ***p < 0.001; *p < 0.05; and ‘NS’ indicates not significant.

3.3. Large intruder aggression: Females are more aggressive than males toward a large intruder

To model the fish's response to a large intruder, we introduced a Domino damselfish (D. trimaculatus) into the home tank of the focal pair and measured their behavioural responses. We found a significant effect of sex on the sum of aggressive behaviours (square root transformed, χ² ₁ = 4.968, p = 0.026, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.397; Figure 2b); females produced more than 3.5 times the number of aggressive behaviours than males (see Supplementary Excel File 8 for ANOVA tables, and 9 for means and standard errors). No significant effects of egg condition were detected for the sum of aggressive behaviours or for any of the behaviours individually (all p > 0.05).

We did, however, find a significant effect of round on the sum of aggressive behaviours (square root transformed, ${\mathrm{\chi }}_2^2$ = 33.070, p < 0.001, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.253; Figure 2c), chases and charges (log transformed, ${\mathrm{\chi }}_2^2$ = 12.477, p = 0.002, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.135; supplementary Figure 1C), displays (square root transformation, ${\mathrm{\chi }}_2^2$ = 24.236, p < 0.001, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.200; supplementary Figure 1D), and latency (log transformed, ${\mathrm{\chi }}_2^2$ = 9.879, p = 0.007, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.122; Figure 2d). In all of these cases, pairs increased in aggression as rounds progressed as indicated by a 2‐fold increase in the number of aggressive behaviours including a 2.5‐fold increase in chases/charges, a 1.8‐fold increase in displays, and a 77% reduction in latency between rounds 1 and 3 (see Supplementary Excel File 4 for pair‐wise comparisons, 8 for ANOVA tables, and 9 for means and standard errors). When males and females were considered together, all recorded large intruder behaviours were significantly repeatable (see Table 3 for ICC and p‐values). Similar to the phenomenon in parental care, however, when the sexes were considered separately, males were significantly repeatable in five out of six recorded behaviours, whereas females were only significantly repeatable in two out of the six recorded behaviours (see Supplementary Excel File 5 for values calculated separately by sex). This is likely a reflection of the high degree of variability in aggression in females, whereas males were consistently passive.

3.4. Male‐oriented aggression: Both sexes display similar progressive aggression toward male intruders

For the male‐oriented aggression assay, we introduced a novel stimulus male into the home tank of the focal pair and measured their behavioural responses. Both focal males and females displayed similar levels of aggression toward the stimulus male (all p > 0.05). A significant effect of egg condition on latency was detected (log transformed, ${\mathrm{\chi }}_2^2$ = 16.626, p < 0.001, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.095; Figure 3a). Pairs addressed the stimulus male approximately 52% faster in the no egg or surrogate eggs condition compared to the own eggs condition. No difference in latency was detected between the no‐eggs condition versus surrogate eggs condition (see Supplementary Excel File 4 for pair‐wise comparisons, 10 for ANOVA tables, and 11 for means and standard errors).

FIGURE 3.

Females are more aggressive toward female intruders than males, but both sexes are similarly aggressive toward male intruders. All data points, including repeated measures, are shown. (a) The log of latency to address the stimulus male in male‐oriented aggression is shown by egg condition and sex. Pairs responded to the stimulus male approximately 50% more slowly when they had their own eggs in the nest compared to when they had surrogate eggs or no eggs. No sex differences were detected. (b) The square root of the sum of aggressive behaviours toward the stimulus male in male‐oriented aggression assay is shown by round and sex. Pairs displayed approximately 10% more aggressive acts in rounds 2 and 3 compared to round 1. No sex differences were detected. (c) The square root of bites toward the stimulus female in female‐oriented aggression is shown by sex. Females bit the female stimulus animal approximately 6.5‐fold more times than males. (d) The square root of displays toward the stimulus female in female‐oriented aggression is shown by sex. Males displayed toward the stimulus female approximately 1.9‐fold more than females. (e) The square root of bites in female‐oriented aggression is shown by the difference in length between the focal female and stimulus female, and sex. As the difference in length between the focal female and stimulus female increases, the pairs get significantly less aggressive (p = 0.008), particularly females. Lines represent linear regression estimates, and ribbons represent the ±95% CI from the mixed model. (f) The square root of the sum of aggressive behaviours toward the stimulus female in female‐oriented aggression is shown by eggs and sex. Pairs produced approximately 1.3‐fold more aggressive behaviours in the none or surrogate egg conditions relative to the own condition. Error bars show the standard error estimate from the linear mixed model that represents individual variance centred around the mean estimate from the model, ***p < 0.001; **p < 0.01; *p < 0.05; and ‘NS’ indicates not significant.

To evaluate whether the significant difference in latency to address the stimulus animal between the own and surrogate eggs conditions might be related to variation in the age or number of eggs present in the nest between the two conditions, we analysed latency including these variables as covariates in the linear mixed model. The average age of the eggs in the own condition was 0.44 days younger than that in the surrogate conditions (mean ± SE, Own: 2.365 ± 0.076, Surrogate ±2.815 ± 0.084; p = 0.020, t[50] = −2.410). There were also slightly fewer eggs in the own condition than in the surrogate condition, though this difference was not significant (mean ± SE, Own: 740.375 ± 36.189, Surrogate: 745.308 ± 48.339; p > 0.05). When including both terms as covariates, the effect of egg condition was no longer significant (p > 0.05; see Supplementary Excel File 10 for the ANOVA table including these covariates). This indicates differences between own and surrogate conditions for latency to address the stimulus male may be explained by variation in the number and age of the eggs used between the tests, rather than the fact the eggs were surrogate versus own.

Similar to the large intruder aggression assay, we found significant effects of round on a few behavioural measures toward the male stimulus. The sum of aggressive behaviours (square root transformed, ${\mathrm{\chi }}_2^2$ = 13.509, p = 0.001, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.082; Figure 3b), latency (log transformed, ${\mathrm{\chi }}_2^2$ = 26.594, p < 0.001, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.205; supplementary Figure 1E) and displays (no transformation, ${\mathrm{\chi }}_2^2$ = 13.765, p = 0.001, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.125; supplementary Figure 1F) changed as rounds progressed. Between rounds 1 and 3, pairs produced approximately 1.2 times more total aggressive behaviours. This included 1.5 times more displays and a 76% reduction in the latency to address the stimulus male (see Supplementary Excel File 4 for pair‐wise comparisons, 10 for ANOVA tables, and 11 for means and standard errors). When the sexes were considered together, all recorded male‐oriented behaviours were significantly repeatable except for latency (see Table 3 for ICC and p‐values). When the sexes were considered separately, males were significantly repeatable in four out of six behaviours, whereas females were significantly repeatable in three out of six behaviours (see Supplementary Excel File 5 for values calculated separately by sex).

3.5. Female‐oriented aggression: Females are more aggressive toward female intruders than males

For the female‐oriented aggression assay, we introduced a stimulus female into the home tank of the focal pair and measured their behavioural responses. We found that in several instances, the size of the stimulus female relative to the focal female had a significant effect on our models. Thus, in all analyses of behaviours in female‐oriented aggression, we added a covariate ‘length difference,’ which was the length of the focal female subtracted by the length of the stimulus female (see Supplementary Excel File 1 for lengths of animals). We also included a length difference‐by‐sex interaction.

Although we did not find a significant effect of sex on the sum of aggressive behaviours (p > 0.05), sex differences emerged when we analysed the behaviours separately. We observed a significant effect of sex on the number of bites (square root transformed, χ² ₁ = 7.846, p = 0.005, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.582; Figure 3c), strikes (no transformation, χ² ₁ = 6.259, p = 0.012, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.517; supplementary Figure 2A), chases/charges (square root transformed, χ² ₁ = 4.523, p = 0.033, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.252; supplementary Figure 2B), and displays (square root transformed, χ² ₁ = 4.161, p = 0.041, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.410; Figure 3d). Females bit the stimulus animal over 6.5 times more than males, struck the stimulus animal 3.2 times more than males, and chased the stimulus animal 4.1 times more than males, whereas males produced nearly double the number of displays. Overall, these data suggest males preferred indirect aggression (displays), whereas females preferred direct aggression (bites, strikes, chases/charges) (see Supplementary Excel File 12 for ANOVA tables, and 13 for means and standard errors).

The difference in length between the focal female and stimulus female had a significant effect on bites (square root transformed, χ² ₁ = 5.684, p = 0.017, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.022; Figure 3e), and strikes (no transformation, χ² ₁ = 4.253, p = 0.039, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.013; supplementary Figure 2C). In both cases, pairs were less aggressive toward the stimulus female as the size of the stimulus female increased relative to the focal female. Specifically, pairs produced approximately 0.9 fewer bites and strikes per centimetre length difference (Figure 3e and supplementary Figure 2C). The interaction between length difference and sex was not significant (p > 0.05) (see Supplementary Excel File 12 for ANOVA table).

We also found a significant effect of egg condition on the sum of aggressive behaviours (square root transformed, ${\mathrm{\chi }}_2^2$ = 25.446, p < 0.001, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.141; Figure 3f). Evaluating behaviours individually, this effect was present for chases/charges (square root transformed, ${\mathrm{\chi }}_2^2$ = 14.743, p < 0.001, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.101; supplementary Figure 2D), displays (square root transformed, ${\mathrm{\chi }}_2^2$ = 9.682, p = 0.008, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.057; supplementary Figure 2E), and strikes (no transformation, ${\mathrm{\chi }}_2^2$ = 10.462, p = 0.005, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.066; supplementary Figure 2F). Similar to male‐oriented aggression, pairs tended to be more aggressive toward the stimulus female in the none or surrogate conditions compared to the own condition. Relative to the own condition, pairs in the none or surrogate conditions displayed 1.3‐fold more aggressive behaviours, 2.2‐fold more chases/charges, 1.5‐fold more strikes, and 1.1‐fold more displays. For all the above measures, no significant difference was detected between the none and surrogate eggs conditions (see supplementary Excel File 4 for pair‐wise comparisons, 12 for ANOVA tables, and 13 for means and standard errors).

Similar to what was described in male‐oriented aggression, to evaluate whether increased female‐oriented aggression in the surrogate condition relative to the own condition could be related to variation in the age or number of eggs between the tests, we again included these variables as covariates in analyses including only own or surrogate egg conditions. The age of eggs had a significant negative effect on the sum of aggressive behaviours (square root transformed, χ² ₁ = 10.101, p = 0.001, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.113; see supplementary Figure 2G) and displays (square root transformed, χ² ₁ = 5.919, p = 0.015, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.068; see supplementary Figure 3H) but was not significant for strikes or chases/charges (p > 0.05). Pairs exhibited approximately six fewer total aggressive behaviours and two fewer displays as the age of the eggs increased per day. The count of eggs also had a significant negative effect on the sum of aggressive behaviours (square root transformed, χ² ₁ = 3.926, p = 0.048, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.046; supplementary Figure 3I). Pairs exhibited approximately 0.6 fewer aggressive behaviours per 100 eggs in the nest. However, even when including these covariates in the mixed model, the difference between own and surrogate conditions remained significant for all of the aforementioned variables (see Supplementary Excel File 12 for ANOVA tables including these covariates). This suggests that there may have been some subtle differences between the own versus surrogate conditions that cannot be explained by variation in the number or age of the eggs during the tests (see Discussion).

There was also a significant interaction between egg condition and sex for the sum of aggressive behaviours (square root transformed, ${\mathrm{\chi }}_2^2$ = 6.587, p = 0.037, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.047; supplementary Figure 3A). Females displayed 37% fewer aggressive behaviours in the own condition compared to the none and surrogate conditions, whereas males showed no significant difference in aggression based on egg condition. This interaction remained significant when including the age and count of eggs as covariates (see Supplementary Excel File 12 for the ANOVA table including these covariates).

Similar to the large intruder aggression and male‐oriented aggression assays, we again found a significant effect of round on the sum of aggressive behaviours (square root transformed, ${\mathrm{\chi }}_2^2$ = 8.173, p = 0.017, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.120; supplementary Figure 3B), displays (square root transformed, ${\mathrm{\chi }}_2^2$ = 6.125, p = 0.047, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.145; supplementary Figure 3C), strikes (no transformation, ${\mathrm{\chi }}_2^2$ = 7.899, p = 0.019, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.036; supplementary Figure 3D), and latency (log transformed, ${\mathrm{\chi }}_2^2$ = 22.238, p < 0.001, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.187; supplementary Figure 3E). Both males and females got more aggressive and more rapidly addressed the stimulus female as rounds progressed. The total number of aggressive behaviours increased 1.4‐fold between rounds 1 and 3. Broken down by individual behaviours, we observed a 1.6‐fold increase in displays, a 1.1‐fold increase in strikes, and a 60% decrease in latency to address the stimulus female between rounds 1 and 3 (see Supplementary Excel File 4 for pair‐wise comparisons, 12 for ANOVA tables, and 13 for means and standard errors). All measured female‐oriented behaviours were significantly repeatable except latency (see Table 3 for ICC and p‐values, and Supplementary Excel File 5 for values calculated separately by sex).

3.6. Aggression ratio: Females have a higher aggression ratio than males

We replicated the finding in Parker et al. (2022) that females displayed an approximately 2.7‐fold higher aggression ratio than males (sum of aggressive behaviours in the female‐oriented aggression assay divided by the sum of aggressive behaviours in the male‐oriented aggression assay; log10 transformed, χ² ₁ = 6.881, p = 0.009, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.392; Figure 3f). No significant effects of egg condition, round, or any of the interactions were detected (p > 0.05; see Supplementary Excel File 14 for the ANOVA table, and 15 for means and standard errors).

The female–male aggression ratio was not significantly repeatable (log transformed, ICC: 0.058, random effect p = 0.214) suggesting within individual variation was magnified by taking the ratio of behaviours across two assays (Certo et al., 2020; Jasieński & Bazzaz, 1999).

3.7. Immediate reaction to a threat: Females are bolder than males

To measure boldness, we measured the focal animal's immediate response to a threat (a researcher's hand) within a short timeframe (10s). We found a significant effect of sex on the number of chases/charges (log transformed, χ² ₁ = 4.952, p = 0.026, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.244; supplementary Figure 4A). In fact, no males were observed charging, whereas four females performed this behaviour. Females were also 59% faster to respond to the stimulus than males (no transformation, χ² ₁ = 7.782, p = 0.005, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.350; Figure 4a; see Supplementary Excel File 16 for ANOVA tables, and Excel File 17 for means and standard errors).

FIGURE 4.

Females are bolder than males, and nest maintenance increases when eggs are in the nest. All data points, including repeated measures, are shown. (a) Latency to address the stimulus in immediate reaction to a threat shown by sex. Females were 35% faster to respond to the stimulus than males. (b) The sum of behaviours aimed at removing the contaminant in nest maintenance is shown by egg condition and sex. Pairs devoted more effort to remove the contaminant with eggs in the nest compared to without. Error bars show the standard error estimate from the linear mixed model that represents individual variance centred around the mean estimate from the model, *** p < 0.001; ** p < 0.01; and ‘NS’ indicates not significant.

Although there were no significant main effects of sex or egg condition for displays, there was a significant interaction between these factors (no transformation, ${\mathrm{\chi }}_2^2$ = 6.319, p = 0.042, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.044; supplementary Figure 4B), though no pair‐wise contrasts were significant. We also found a significant interaction between sex and egg condition for latency (no transformation, ${\mathrm{\chi }}_2^2$ = 10.063, p = 0.007, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.069; supplementary Figure 4C). Post hoc tests indicated that females did not significantly differ in latency based on egg condition, whereas males addressed the stimulus 33% faster with no or surrogate eggs in the nest relative to own eggs (see Supplementary Excel File 4 for pair‐wise comparisons, 16 for ANOVA tables, and 17 for means and standard errors).

To evaluate whether the difference in latency to address the stimulus in males between the own and surrogate conditions could be related to variation in the age or number of eggs between the tests, we again included these as covariates in an analysis of only these two egg conditions. Neither of these covariates were significant, and the significant difference between own and surrogate egg conditions remained (see Supplementary Excel file 16 for the ANOVA table including these covariates). This suggests the slight delay that males have in addressing an immediate threat when caring for their own versus surrogate eggs cannot be explained by variation in the number or age of the eggs between the tests. All three behaviours measured for the immediate reaction to a threat assay were significantly repeatable (see Table 3 for ICC and p‐values, and Supplementary Excel File 5 for values calculated separately by sex).

3.8. Nest maintenance: Females contribute slightly more toward nest maintenance than males

To assess if there was a division of labour in nest maintenance, we introduced a nest contaminant (a small rock) into the terra cotta pot and measured behaviours aimed at removing the contaminant from the nest (strikes, bites, and latency). We found a slight but significant effect of sex on the sum of aggressive behaviours (no transformation, χ² ₁ = 4.081, p = 0.043, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.025; supplementary Figure 4D) and latency to address the stimulus (log transformed, χ² ₁ = 5.943, p = 0.015, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.005; supplementary Figure 4E). Females performed 1.1‐fold more aggressive behaviours than males and addressed the contaminant 35% faster than males (see Supplementary Excel File 18 for ANOVA tables, and 19 for means and standard errors).

We also found a significant effect of egg condition on several recorded behaviours. This included the sum of behaviours (no transformation, ${\mathrm{\chi }}_2^2$ = 12.246, p = 0.002, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.180; Figure 4a), strikes (no transformation, ${\mathrm{\chi }}_2^2$ = 10.809, p = 0.004, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.198; supplementary Figure 4F), and bites (square root transformation, ${\mathrm{\chi }}_2^2$ = 9.436, p = 0.009, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.115; supplementary Figure 5A). Pairs were significantly more aggressive toward the contaminant with eggs in the nest (own or surrogate) as opposed to those without eggs. No differences between own and surrogate egg conditions were detected. Pairs responded to the contaminant 63% faster with eggs in the nest compared to without and produced nearly double the number of behaviours aimed at removing the contaminant (see Supplementary Excel File 4 for pair‐wise comparisons, 18 for ANOVA tables, and 19 for means and standard errors).

In addition, we found a significant effect of round on strikes (no transformation, ${\mathrm{\chi }}_2^2$ = 6.883, p = 0.032, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.003; supplementary Figure 5B). Similar to previous assays, as rounds progressed, pairs got more aggressive, though no contrasts were significant in post hoc tests (see Supplementary Excel File 4 for pair‐wise comparisons, 17 for ANOVA tables, and 18 for means and standard errors).

We also found a significant interaction between sex and round on the sum of behaviours (no transformation, ${\mathrm{\chi }}_2^2$ = 8.399, p = 0.015, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.058; supplementary Figure 5C), strikes (no transformation, ${\mathrm{\chi }}_2^2$ = 10.525 p = 0.005, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.072; supplementary Figure 5D), and latency (log transformed, ${\mathrm{\chi }}_2^2$ = 10.761, p = 0.005, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.073; supplementary Figure 5E). Males significantly increased nest maintenance behaviours and decreased latency as rounds progressed, whereas females significantly decreased strikes as rounds progressed (see Supplementary Excel File 4 for pair‐wise comparisons, 18 for ANOVA tables, and 19 for means and standard errors).

Further, we found a significant interaction between sex and eggs for latency (log transformed, ${\mathrm{\chi }}_2^2$ = 10.761, p = 0.004, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.073; supplementary Figure 5F). Males addressed the contaminant significantly faster with eggs in the nest compared to those without, whereas females did not significantly differ based on egg condition (see Supplementary Excel File 4 for pair‐wise comparisons, 17 for ANOVA tables, and 18 for means and standard errors). All recorded nest maintenance behaviours were significantly repeatable (see Table 3 for ICC and p‐values, and Supplementary Excel File 5 for values calculated separately by sex).

3.9. Principal component analysis

In addition to analysing each behavioural assay separately to determine effects of sex, round, and egg condition, we sought to evaluate how a composite of all the recorded assays might be a useful metric for differentiating the sexes. Thus, we performed a PCA analysis on the sum of behaviours from each assay collapsed across all rounds and egg conditions within individuals. We found that the first and second PCs explained 65.7% of the variation. A plot of PC2 on PC1 shows the individuals largely separate based on a parental‐aggression axis (Figure 5), which effectively separates the sexes, except for a few individuals at the margins. These include one male (M2 in Figure 5a), which displays similar levels of aggression and parental care as females, and two females that do more parenting and display less aggression than usual (F4 and F8 in Figure 5a). PC1 from this composite PCA analysis significantly differentiated the sexes (t[16] = 3.512, p = 0.003), whereas PC2 did not (p > 0.05). Additionally, we found that PC2 captured the variation among aggression assays (Figure 5b). Conspecific‐oriented aggression assays loaded negatively, whereas non‐conspecific aggression assays loaded positively (with the exception of immediate reaction to a threat), suggesting conspecific aggression and non‐conspecific aggression constitute two distinct types of aggression.

FIGURE 5.

Males and females separate largely along a parental care‐aggression axis. (a) A principal component analysis (PCA) was performed on the sum of behaviours (collapsed across rounds and egg conditions) for each of the seven behavioural assays. PC1 is plotted against PC2 showing males and females separately. In addition, the global female and male averages are shown as large red circle and large blue triangle, respectively. (b) Loadings of the corresponding assays are shown on PC1 and PC2. Males and females largely split along a parental care‐aggression axis, where parental care loads to quadrant 1, whereas the various aggression assays load to quadrants 2 and 3.

Additionally, we performed multiple separate PCA analyses for all of the recorded behaviours within each assay (supplementary Figures 6–8). We then calculated the Euclidean distance between the mean male and mean female PC1 and PC2 coordinates. Based on this metric, the assays that best differentiated the sexes were parental care (2.767), female‐oriented aggression (2.645), and large intruder aggression (2.297; Table 4). PC1 extracted from the analyses of parental care, large intruder aggression, female‐oriented aggression, and immediate reaction to a threat significantly differentiated males and females but did not significantly differentiate the sexes in the other assays, whereas PC2 extracted from male‐oriented aggression significantly differentiated the sexes (see Supplementary Excel File 20 for statistical analyses).

TABLE 4.

Euclidian distances between mean male and mean female PC1 and PC2 coordinates.

Assay	Euclidean distance
Parental care	2.767
Small intruder aggression	1.165
Large intruder aggression	2.297
Male‐oriented aggression	1.565
Female‐oriented aggression	2.645
Immediate reaction to a threat	2.147
Nest maintenance	0.577

Note: Parental care, female‐oriented aggression, and large intruder aggression produce the largest Euclidean distance between the average male and average female.

3.10. Pair dynamics

To estimate the influence of one member of the pair's behaviour on the other, we analysed each behaviour separately by sex and entered the alternative member of the pair's value for the behaviour under consideration as a covariate in the linear mixed model analyses. We hypothesized that each sex would have a negative influence on the other reasoning that a deficiency in one partner's behaviour would be compensated by the other. However, we observed the exact opposite. In 22 out of 36 behaviours, the behaviour of one sex was a positive predictor of the behaviour of the other; that is, behaviours were positively correlated between sexes (see Supplementary Excel File 21).

3.11. Behavioural syndromes

Of the 21 correlations calculated for males, 3 were significant. Both female‐oriented aggression and male‐oriented aggression were significantly positively correlated with an immediate reaction to a threat (r [7] = 0.897, Q = 0.021; Figure 6a; r [7] = 0.831, Q = 0.049; Figure 6b, respectively). Female‐oriented aggression was also positively correlated with male‐oriented aggression (r [7] = 0.819, p = 0.049; Figure 6c). No significant correlations between behaviours were observed for females.

FIGURE 6.

Males display a few behavioural syndromes, whereas females display none. All data points represent the sum of behaviours collapsed across round and egg condition within each individual. (a) The sum of aggressive behaviours in immediate reaction to a threat plotted against the sum of aggressive behaviours in female‐oriented aggression assay, males only (Q = 0.021). (b) The sum of aggressive behaviours in immediate reaction to a threat plotted against the sum of aggressive behaviours in male‐oriented aggression, males only (Q = 0.049). (c) The sum of aggressive behaviours in male‐oriented aggression plotted against sum of aggressive behaviours in female‐oriented aggression, males only (Q = 0.049). Blue lines represent simple linear regression lines for males. Ribbons show ±95% CI.

4. DISCUSSION

Here we significantly broadened the range of quantifiable behaviours in A. ocellaris, enabling a more comprehensive examination of their unique characteristics in a laboratory setting. To this end, we established significant individual differences, effects of sex, reproductive contexts (e.g., whether eggs are present or not), and repeated testing (e.g., round) on multiple measures of territory defence, aggression, and parental care. All assays contained measures that were significantly repeatable across contexts and rounds. The average ICC of measured behaviours was 0.36, consistent with estimates of repeatability of behaviour of A. ocellaris from other laboratory studies (Wong et al., 2013) and animals in general (Bell et al., 2009).

4.1. Consistent sex differences

All of the assays described in this paper except for male‐oriented aggression had at least one behaviour that significantly differed based on sex. For the aggression measures, this was almost always an increase in aggression in females, and an increase in latency in males. Interestingly, in female‐oriented aggression, the sexes did not significantly differ in the sum of aggressive behaviours, but did differ in their behavioural composition, with males displaying indirect aggression (displays) while females displaying more direct aggression (bites) (Figure 3c,d). Thus, future studies that use this assay to evaluate sex differences should evaluate displays and bites separately. The different ways in which males and females direct aggression toward a female intruder are likely due to the significant size difference between males and females and their unique natural history. Males may be less willing to directly confront a female because of possible injury and mating potential with it, whereas a female will fight to the death to maintain/usurp the α position in the hierarchy as searching for a new anemone means almost certain death (Da Silva & Nedosyko, 2016; Iwata & Manbo, 2013).

4.2. Best assays for detecting sex differences

We found that the best assays for detecting sex differences were parental care, female‐oriented aggression, and large intruder aggression. These assays detected large and significant sex effects across conditions and rounds. Because of the large effect size and internal consistency, these assays would be the most useful for evaluating whether a fish in the middle of a sex change behaves more as a male or as a female, that is, for experimentally identifying when a behavioural sex change occurs. However, we found that none of these, whether considered separately or combined into a linear combination, could perfectly distinguish the sexes. This was due to strong and repeatable individual differences in behaviour with some females behaving similarly to males (e.g., F8 in Figure 5a and supplementary Figures 6C, 7A, 7C, and 8A), and some males behaving similarly to females (e.g., M2 in Figure 5a and supplementary Figures 6A, 6C) in specific assays.

4.3. Sex‐specific behavioural syndromes

In addition to uncovering large sex differences for independent, repeatable behaviours, we observed strong evidence for sex‐specific behavioural syndromes in A. ocellaris. In males, we observed positive correlations between male and female‐oriented aggression and boldness. However, in females, none of the correlations reached statistical significance beyond a false discovery rate of 5%. Most of the sex differences we observed including the syndromes are likely caused at some level by the organizational and activational effects of the sex steroid hormones 11‐ketotestosterone (11‐KT) and estradiol (E2), which have been shown to have broad effects on multiple aspects of physiology and behaviour (Gegenhuber et al., 2022; Godwin, 2010; Stennette & Godwin, 2024).

4.4. Positive behavioural feedback

In anemonefish, male–female pairs work together as a reproductive unit and display a strong division of labour (Barbasch et al., 2022; Casas et al., 2022). Because of this we expected behaviours to be negatively correlated between members of the same pair. However, we found the opposite. In 22 out of 36 behaviours, the behaviours were positively correlated with each other. It is notable that the only assay where this was largely not true was parental care, where parental care acts by the male were performed largely independently of female parental care acts. Aggression levels may be positively correlated within pairs due to a positive feedback loop, where one individual's aggression triggers and amplifies the other's, as seen in Betta splendens (Vu et al., 2020). For example, a male may become more aggressive when its female partner does, and vice versa. Additionally, simultaneous aggression from both would likely present a more formidable threat to an intruder.

4.5. Increase in aggression over time

In many of our assays, we found at least one behaviour that significantly changed as rounds progressed. This was almost always an increase in aggressive behaviour or a decrease in latency to address the stimulus. These round differences are likely caused by acclimation to the stimulus following repeated exposure leading to increased confidence that aggression does not pose a risk. In male and female‐oriented aggression, the increase in aggression as round progressed may also be caused by winner effects as, in most cases, the focal pair would likely perceive they ‘won’ the encounter (Dijkstra et al., 2012; Dugatkin, 1997; Hsu & Wolf, 1999).

4.6. Effect of reproductive context on behaviour

We also found in many assays that the fish's behaviour depended on the reproductive context, whether pairs had their own, surrogate, or no eggs present in the nest when the behaviours were recorded. In nest maintenance, when the pairs had no eggs in the nest, they displayed fewer behaviours aimed at clearing the contaminant than when they had their own or surrogate eggs. This was expected, as a contaminant in the nest serves as an obstacle to the care of the eggs meaning motivation to clear the contaminant would be higher with eggs in the nest compared to that without eggs. Additionally, when eggs are in the nest, the male spends more time in the nest than when it is empty (DeAngelis & Rhodes, 2016), meaning the male is more likely to notice the contaminant and will more rapidly address it (supplementary Figure 5F). However, we did not expect to find the consistent differences between own and surrogate conditions we observed. There were significant differences in behaviour between the own and surrogate conditions in male‐oriented aggression (Figure 3a), female‐oriented aggression (Figure 3f, supplementary Figures 2E, 2F, 3A), and immediate reaction to a threat (supplementary Figure 4B) that were consistently characterized by increased latency and decreased aggression in the own condition relative to none or surrogate. For the most part, we were not able to explain these significant differences with the count or age of the eggs suggesting two possibilities. The first possibility is that pairs can differentiate their own eggs from surrogate eggs, though this seems unlikely as own and surrogate eggs are treated the same in all other assays. The second possibility is that the replacement of the terra cotta pot serving as the pair's nest leads to subtle changes in behaviour in these assays even after the 24 h acclimation period. Future studies should examine under what contexts pairs can detect differences between own and surrogate eggs to better understand the scope of this difference.

4.7. Limitations

One limitation of this study is the small sample size, as we studied only nine pairs of A. ocellaris. Future studies with a larger sample size are needed for generalization. With a larger sample size it may also be possible to estimate the effect that the specific individual intruders have on the focal fish (e.g., specific damselfish individuals or specific crabs) that could account for some of the individual variation in the present study. Future studies should include a control group that has their terra cotta pot with eggs removed and then replaced to control for the disturbance of replacing the pot, in addition to the 24‐h acclimation period.

5. CONCLUSION

In conclusion, we found 33 out of 35 behaviours were significantly repeatable across three different reproductive contexts. This result is consistent with what has been observed for A. ocellaris and other species (Bell, 2007; Wong et al., 2013). We detected a few behavioural syndromes in males but none in females. Behavioural syndromes were expected based on previous work (Barbasch & Buston, 2018; Schmiege et al., 2016; Wong et al., 2013), but a stronger signal in males than females was not expected and requires additional theoretical and empirical work to understand. It may be that because females are at the end of the plasticity window, in the terminal phase, there is less individual variation in behaviour to detect significant correlations. This was supported by overall weaker ICC values in females than males (See Supplementary Excel File 5). The observation that the behaviours of males and females within a pair were positively correlated was also unexpected. Overall, with the strong individual repeatability of behaviours, this result suggests that even though there is extraordinary plasticity in the behaviours displayed by anemonefish transforming from male to female, idiosyncrasies of individual members remain largely independent. Individual behaviours are only correlated between members of the reproductive unit to the extent that aggressive defensive behaviours positively feed off each other. Additionally, we found that levels of territorial aggression tended to go up when the pair had no eggs in the nest, consistent with the hypothesis that without eggs, the fish have more time and energy to devote to defence. However, the observation that the fish are more aggressive defending the territory when they have surrogate eggs compared to their own was surprising. This intriguing result suggests that the fish may in fact be able to discriminate their own eggs from surrogate, though additional studies are necessary to eliminate all possible confounding variables. Our study significantly broadens the number of laboratory behavioural assays available for studying female behavioural dominance and territoriality, protandrous behavioural sex change, and behavioural plasticity in the iconic anemonefish, A. ocellaris. In particular, we are poised to use the female‐oriented aggression and large intruder assay, along with the standard parental care assays to quantify territorial aggression and parental care in A. ocellaris males in the midst of sex change to evaluate the behavioural transformation during the process.

AUTHOR CONTRIBUTIONS

G.J.G. conceived the experimental design, conducted data collection and analysis, created all graphs, and contributed to writing the manuscript. I.M.W. and E.K.P. contributed to the experimental design and data collection. J.S.R. assisted with the experimental design and manuscript writing.

FUNDING INFORMATION

Funding was from gifts and indirect costs recovered from federal grants to J.S.R.

Supporting information

Data S1. Supporting information.

Data S2. Supporting information.

Data S3. Supporting information.

Data S4. Supporting information.

Data S5. Supporting information.

Data S6. Supporting information.

Data S7. Supporting information.

Data S8. Supporting information.

Data S9. Supporting information.

Data S10. Supporting information.

Data S11. Supporting information.

Data S12. Supporting information.

Data S13. Supporting information.

Data S14. Supporting information.

Data S15. Supporting information.

Data S16. Supporting information.

Data S17. Supporting information.

Data S18. Supporting information.

Data S19. Supporting information.

Data S20. Supporting information.

Data S21. Supporting Information.

Figure S1. Males perform more parental care behaviours, and pairs get more aggressive as rounds progress. All data points, including repeated measures, are shown. (a) The square root of nips in parental care is shown by sex. Males performed 4.6‐fold more nips than females. (b) The square root pectoral fans in parental care are shown by sex. Males performed 10.2‐fold more pectoral fans than females. (c) The log of chases/charges toward the stimulus animal in large intruder aggression is shown by round and sex. Pairs performed 2.5‐fold more chases/charges as rounds progressed. (d) The square root displays in large intruder aggression are shown by sex and round. Pairs produced 1.9‐fold more displays as rounds progressed. (e) The log of latency to address the stimulus male in male‐oriented aggression is shown by round and sex. Pairs addressed the stimulus male 76% faster as rounds progressed. (f) Displays in male‐oriented aggression are shown by round and sex. Pairs produced 1.5‐fold more displays as rounds progressed. Error bars show the standard error estimate from the linear mixed model that represents individual variance centred around the mean estimate from the model, ***p < 0.001, **p < 0.01, *p < 0.05, and ‘NS’ indicates not significant.

FIGURE S2. Females perform significantly more direct aggression than males, pairs get less aggressive as the focal female increases in size relative to the stimulus female, pairs are the least aggressive in the own eggs condition, and egg age and count are negative predictors of aggression. All data points, including repeated measures, are shown. (a) Strikes toward the stimulus female in female‐oriented aggression are shown by sex. Females produce 3.2‐fold more strikes than males. (b) The square root of chases/charges in female‐oriented aggression is shown by sex. Females chased the stimulus female 4.1‐fold more than males. (c) Strikes toward the stimulus female in female‐oriented aggression are shown by the difference in length between the focal female and the stimulus female and sex. Pairs got significantly less aggressive as stimulus female got larger, particularly females. (d) The square root of chases/charges toward the stimulus female in female‐oriented aggression is shown by sex and egg condition. Pairs produced 2.7‐fold more aggressive behaviours in the none condition relative to the own condition. (e) The square root of displays toward the stimulus female in female‐oriented aggression is shown by sex and eggs. (f) Strikes toward the stimulus female in female‐oriented aggression are shown by sex and eggs. Pairs were 32% less aggressive with their own eggs relative to none or surrogate. (g) The square root of the sum of aggressive behaviours in female‐oriented aggression is shown by egg age and sex. For each day the age of the eggs increased, and the sum of aggressive behaviours decreased by 6. (h) The square root of the sum of displays in female‐oriented aggression is shown by egg age and sex. For each day the age of the eggs increased, and the number of displays decreased by 2. (i) The square root of the sum of behaviours in female‐oriented aggression is shown by egg count and sex. For each 100 egg increase, the sum of aggressive behaviours decreased by 0.6. In C and G‐I, lines represent linear regression estimates, and ribbons represent the ±95% CI from the mixed model. Error bars show the standard error estimate from the linear mixed model that represents individual variance centred around the mean estimate from the model, ***p < 0.001, *p < 0.05, and ‘NS’ indicates not significant.

FIGURE S3. Females significantly differ in aggression based on egg condition, and pairs get more aggressive as rounds progress. All data points, including repeated measures, are shown. (a) The square root of the sum of aggressive behaviours in female‐oriented aggression is shown as eggs by sex interaction. ‘N’ indicates no eggs, ‘O' indicates own eggs, and ‘S' indicates surrogate eggs. Females significantly reduced aggression in the own eggs condition compared to the none or surrogate conditions, whereas males did not. (b) The square root of the sum of aggressive behaviours in female‐oriented aggression is shown by sex and round. Pairs produced 1.4‐fold more aggressive behaviours between rounds 1 and 3. (c) The square root of displays in female‐oriented aggression is shown by sex and round. Pairs produced 1.7‐fold more aggressive behaviours between rounds 1 and 3. (d) Strikes in female‐oriented aggression are shown by sex and round. Although this term was significant in the ANOVA, no pair‐wise comparisons were significant. (e) The log of latency to address the stimulus female in female‐oriented aggression is shown by sex and round. Pairs addressed the stimulus female 59% faster between rounds 1 and 3. Pairs produced 1.2‐fold more displays in the surrogate eggs condition compared to the own eggs condition. (f) Log10 aggression ratio is shown by sex. Females are 2.7‐fold more biased toward female‐oriented aggression than males. Error bars show the standard error estimate from the linear mixed model that represents individual variance centred around the mean estimate from the model, ***p < 0.001, *p < 0.05, and ‘NS’ indicates not significant.

FIGURE S4. Males are less bold than females and significantly differ in boldness based on egg condition, and females perform more nest maintenance behaviours than males. Nest maintenance behaviours are dependent on the presence of eggs in the nest. All data points, including repeated measures, are shown. (a) The log of chases/charges in immediate reaction to a threat is shown by sex. Only females were observed performing chases/charges. (b) Displays toward the stimulus in immediate reaction to a threat are shown as a sex by eggs interaction. Although this term was significant in the ANOVA, no pair‐wise comparisons were significant. (c) Latency to address the stimulus in immediate reaction to a threat is shown as a sex by eggs interaction. Although this term was significant in the ANOVA, no pair‐wise comparisons were significant. (d) The sum of behaviours in nest maintenance is shown by sex. Females performed 1.1‐fold more behaviours than males. (e) Latency in nest maintenance is shown by sex. Females addressed the contaminant 33% faster than females. (f) Strikes in nest maintenance are shown by sex and eggs. Pairs produced 1.9‐fold more strikes with eggs in the nest compared to those without eggs. Error bars show the standard error estimate from the linear mixed model that represents individual variance centred around the mean estimate from the model, **p < 0.01, *p < 0.05, and ‘NS’ indicates not significant.

FIGURE S5. Pairs perform fewer nest maintenance behaviours without eggs in the nest, and differences in behaviour based on round or egg condition are largely driven by males. All data points, including repeated measures, are shown. (a) Bites toward the contaminant in nest maintenance are shown by sex and eggs. Pairs produced 1.9‐fold more bites with eggs in the nest compared to those without eggs. (b) Strikes toward the contaminant in nest maintenance are shown by sex and round. Although this term was significant in the ANOVA, no pair‐wise comparisons were significant. (c) The sum of behaviours in nest maintenance is shown as a sex by round interaction. Females did not significantly differ as rounds progressed, whereas males performed 1.7‐fold more aggressive behaviours between rounds 1 and 3. (d) Strikes in nest maintenance are shown as a sex by round interaction. Males did not significantly change as rounds progressed, whereas females produced 20% fewer strikes between rounds 1 and 2. (e) The log of latency to address the contaminant in nest maintenance is shown as a sex by round interaction. Females did not significantly change in latency to address the contaminant as rounds progressed, whereas males addressed the contaminant 72% faster in round 3 compared to round 1. (f) The log of latency in nest maintenance is shown as a sex by eggs interaction. Females did not significantly differ in latency to address the contaminant based on egg condition, whereas males addressed the contaminant 75% faster with eggs in the nest compared to those without. Error bars show the standard error estimate from the linear mixed model that represents individual variance centred around the mean estimate from the model, ***p < 0.001, **p < 0.01, *p < 0.05, and ‘NS’ indicates not significant.

FIGURE S6. Principal component analysis (PCA) for behavioural composition of parental care, small intruder aggression, and large intruder aggression. For each figure, the left plot displays the loading of individuals based on PCs 1 and 2, and the right plot displays the variable loadings within each assay. Each PCA contains the mean of each behaviour recorded in each assay collapsed across round and egg condition. In all plots, the larger blue triangle indicates the mean of males, and the larger red circle indicates the mean of females. (a) Parental care PC1 versus PC2. Males separate from females based on any parental behaviour. (b) Small intruder aggression PC1 versus PC2. Males and females separate based almost exclusively on latency, with males displaying higher latency. (c) Large intruder aggression PC1 versus PC2. Males and females separate largely based on latency and displays, with males producing high latency and low displays, and females producing low latency and high displays.

FIGURE S7. Principal component analysis (PCA) for behavioural composition of male and female‐oriented aggression. For each figure, the left plot displays the loading of individuals based on PCs 1 and 2, and the right plot displays the variable loadings within each assay. Each PCA contains the mean of each behaviour recorded in the assay collapsed across round and egg condition. In all plots, the larger blue triangle indicates the mean of males, and the larger red circle indicates the mean of females. (a) Male‐oriented aggression PC1 versus PC2. Males and females separate based on displays and latency; however, unlike in other assays, males are characterized by high displays and low latency, with females displaying the opposite pattern. (b) Female‐oriented aggression PC1 versus PC2. Similar to in our statistical analysis, males and females separate based on direct aggression (bites, chases, strikes), and indirect aggression (displays) with males displaying high indirect aggression and low direct aggression, and females displaying the opposite pattern of aggression.

FIGURE S8. Principal component analysis (PCA) for behavioural composition of immediate reaction to a threat and nest maintenance. For each figure, the left plot displays the loading of individuals based on PCs 1 and 2, and the right plot displays the variable loadings within each assay. Each PCA contains the mean of each behaviour recorded in an assay collapsed across round and egg condition. In all plots, the larger blue triangle indicates the mean of males, and the larger red circle indicates the mean of females. (a) Immediate reaction to a threat. Males and females separate based on latency and aggression, with males being less aggressive and producing higher latency than females. (b) Nest maintenance. Males and females do not separate particularly well; however, males produce slightly higher latency, whereas females produce slightly more bites and strikes.

FIGURE S9. Male and female levels of aggression toward a female intruder are correlated. The mean of the sum of aggressive behaviours recorded for females against the sum of aggressive behaviours recorded for males in female‐oriented aggression collapsed across egg condition and round. Females' behaviour significantly predicts males' behaviour. The blue line represents simple linear regression lines. Ribbons show 95% CI.

Graham, G. J. , Wilton, I. M. , Panczyk, E. K. , & Rhodes, J. S. (2025). Novel behavioural assays reveal sex‐specific behavioural syndromes in anemonefish. Journal of Fish Biology, 106(6), 1845–1862. 10.1111/jfb.16069

DATA AVAILABILITY STATEMENT

Raw data and code used in this manuscript can be found at https://github.com/ggraham4/Graham_etal_Novel_Behavioral_Assays/.