Impact of intraspecific variation in teleost fishes: aggression, dominance status and stress physiology

A dominance hierarchy describes a social situation in which animals are physically or chemically (e.g. urinary signals) dominant over other animals in their social group (Breed and Moore, 2012; Tibbetts et al., 2022). The prevalence of dominance hierarchies in nature suggests they are an evolutionary solution to overt fighting once the hierarchy is stable. This allows group-living animals to coexist with one another harmoniously if individuals know their place and do not engage higher-ranked individuals in aggressive interactions. Therefore, social animals often develop dominance-based social hierarchies and an individual's social rank greatly affects behaviour, development and physiology (Cubitt et al., 2008). Living within a group presents many advantages as there are more individuals to locate food, identify predators and dilute the predation risk, but social rank can influence the differential distribution of resources dependent upon dominance status (Culbert et al., 2019). Being subject to aggressive interactions which the individual does not win may lead to a chronic stress response where the animal perceives the constant threat from a higher ranked or dominant individual as a danger disrupting homeostasis. Therefore, low social rank may result in chronic stress and lower-ranking animals (i.e. not dominant) typically show submissive behaviour and a general behavioural inhibition. Dominant animals, on the other hand, show the opposite behavioural profile: they are aggressive, active, highly competitive and monopolize limited resources such as food, mates and shelter. The hierarchy is often established by pairwise agonistic interactions and the outcome tends to be based upon the relative competitive ability of the contestants. Competitive ability is determined by an interaction between genetic factors, social environment and previous experience. The behavioural inhibition displayed by subordinate animals could serve as a strategy to avoid aggression from the more dominant, whereby they sneakily obtain resources while waiting for an opportunity to climb up the hierarchy. Thus, when given the opportunity to increase in rank, individuals of low rank will have to shift from a subordinate to dominant behavioural state. This behavioural switch includes modifications in endocrine signalling, transcriptome profiles and changes in the activity of various neurotransmitter systems. In this Review, we will focus on teleost fish since there is a wealth of empirical data we can draw upon, and on the interaction of genetic and environmental effects on competitive ability. Teleost fishes are becoming increasingly more important as vertebrate models in a wide range of scientific research and thus it is important to understand how their social structure could potentially affect physiological studies and introduce high variation within data sets.

Fishes provide highly tractable models for the study of dominance behaviour since they often form easily measured linear dominance hierarchies (Holekamp and Strauss, 2016; Wallace et al., 2022). Typically, the dominant or top-ranked individual displays more aggressive behaviour, is relatively larger or has some other intrinsic property greater/better than that of others within the hierarchy. This provides the dominant individual with exclusive or priority access to resources, which may include mating opportunities (Rodriguez-Santiago et al., 2020). Size can often be used as a predictor of dominance in fishes (e.g. Reddon et al., 2011). Lower-ranked (or middle ranked) individuals are termed subdominants, and they may challenge the dominant for the top position. Beneath subdominants, subordinate individuals occupy the very lowest rank positions. Sticklebacks (Gasterosteus aculeatus) held in groups of four individuals form a linear hierarchy with aggression being highest in the dominant individual, who obtains most food (Sneddon et al., 2006; Fig. 1). The subdominants in rank positions 2 and 3 obtain less food; however, the subordinate in rank 4 is much less aggressive and adopts a sneaky foraging strategy, which allows it to obtain food when the other fish are engaging in agonistic interactions. These rank positions also influence growth rates: the rank 1 dominant sticklebacks grow more than the subdominants, who display negative growth; this may be due to them engaging in unsuccessful fights with the dominant fish, leading them to expend energy to fuel the costs of aggression (Sneddon et al., 2006). In contrast, the sneaky foraging strategy of the subordinate fish pays off and these individuals exhibit positive growth rates. Thus, rank position can have profound effects on biological function and physiology.

Generally, the dominant individual controls the behaviour of the group, and the subdominants and subordinates are constrained by the use of aggression by the dominant. In the cichlid fish, Astatotilapia burtoni, dominant males are aggressive and exert influence over normal movement of the group. In contrast, subordinate males are non-aggressive, socially peripheral and do not direct group movement (Rodriguez-Santiago et al., 2020). This behavioural suppression and lack of access to resources may result in chronic stress in lower-ranked individuals, thus eliciting tertiary or long-term responses to stress, including impaired immune function, growth and reproduction (Rodriguez-Santiago et al., 2020). These biobehavioural differences between animals of differing rank position may increase energetic demands and consequently affect both energy intake and metabolism (Milewski et al., 2022). Further changes are known to take place in a variety of physiological systems, such as the hypothalamic–pituitary–interrenal (HPI) and hypothalamic–pituitary–gonadal (HPG) axes, and include altered gene expression and responsiveness of neural circuits that regulate dominance-related behaviours (Maruska and Fernald, 2011; 2012; Øverli et al., 2004; Almeida et al., 2019). In contrast, low-ranking animals inhibit aggression, and this may also include reproductive behaviours. Indeed, subordinates exhibit physiological changes to allow them to adapt to the stress of their low rank within the hierarchy (Backström and Winberg, 2017; Höglund et al., 2019). Chronic social stress, where low-ranked members of a hierarchy experience a long-term stress response, can lead to the occurrence of health issues since this impairs immune function (Dara et al., 2022). Yet, studies have demonstrated that rapid changes in dominance status occur as fish socially increase or decrease their rank and these changes are associated with dynamic changes in the brain (Sneddon et al., 2011; Simões et al., 2015). Gene expression profiles obtained from rainbow trout (Oncorhynchus mykiss) brains with different rank positions demonstrate that there is a specific gene expression signature for dominants versus subdominants and subordinates (Sneddon et al., 2011). Upon removal of the dominant fish, subdominants display a rapid shift in brain gene expression to that of a dominant profile within 48 h (Sneddon et al., 2011). Therefore, caution should be applied to the measurement of responses from groups of fish, since it appears that social rank influences global gene expression and physiology, through to behaviour, and thus knowledge of each individual's rank position within a group may provide new insights into intraspecific variation. Transcript profiles were obtained from whole brains, and the genes expressed have functions in protein turnover, metabolism, cell structure, transport and stress; these were upregulated in low-ranked trout compared with dominant fish. This differential expression profile may reflect the compromising nature of low social status in addition to the elevated stress response in lower-ranked fish compared with dominant individuals (Sneddon et al., 2011). Behaviourally relevant candidates were identified and included gamma aminobutyric acid-receptor associated protein (GABA-RAP) with roles in stress and aggression. Here GABA-RAP was upregulated in subordinates compared with dominant fish and this correlated with elevated cortisol in subordinates and thus this may be a consequence of increased stress in subdominants (Sneddon et al., 2011). Knowledge of the life history of the specific species being investigated is crucial to meaningfully understand the impact of dominance on physiology since this is likely to vary across species. Dominant A. burtoni, for example, form breeding territories and also divert energy to gonadal growth so dominant males lose weight compared with subordinates, possibly because of the costs of territory defence and reproductive investment (Hofmann et al., 1999).

Winning or losing fights for social dominance has a dramatic effect on future chances of winning fights (Hsu and Wolf, 1999; Oliveira et al., 2011). The winner increases its chance to win future contests, whereas the loser is more likely to continue losing; a phenomenon which has been referred to as the winner/loser effect. Even though body size is usually a strong predictor of the outcome of fights for social dominance (Reddon et al., 2011), Abbott et al. (1985) showed that the effect of previous social experience may outweigh the effect of size in dyadic fights in steelhead trout (Oncorhynchus mykiss). In their experiment, the outcome of initial fights was clearly related to body size, with larger fish having greater fighting ability than smaller conspecifics. However, surplus feeding of subordinates did not reverse the dominance relationship, even though subordinates ended up being 14–50% larger than their dominant contestants (Abbott et al., 1985). Apparently, fighting experience alters a contestant's estimated and/or actual fighting ability (Hsu and Wolf, 1999). Long-term social interaction, as in the experiment by Abbott et al. (1985), may cause injuries in subordinates such as fin damage or loss of weight and physical strength, and this may reduce their actual fighting ability. However, losing fights appears to inhibit aggression and motivation or willingness to initiate agonistic interactions, even with opponents that are naïve and considerably smaller in body size and thus have less resource-holding power (RHP). Using a resident/intruder test, where an isolated animal is challenged by the introduction of a conspecific intruder, Höglund et al. (2001) quantified aggressive behaviour, in isolated juvenile Arctic charr (Salvelinus alpinus) prior to and following 2 days of dyadic interaction with a size-matched opponent. Following dyadic interaction, losers showed a significant decline in the number of aggressive acts performed towards the small intruder, whereas the opposite was observed in winners from the dyadic interaction. Moreover, losers showed a profound increase in the latency to attack the intruder (Höglund et al., 2001).

The inhibition of aggressive behaviour observed in subordinates is likely to be mediated by factors involved in the neuroendocrine stress response (Summers and Winberg, 2006). In juvenile Arctic charr kept in groups of four, the most subordinate fish, social rank 4, displayed a longer-lasting behavioural inhibition compared with the subdominant fish (Winberg et al., 1992; Fig. 2). In a stable dominance hierarchy, the most subordinate animal is subjected to more intense stress than individuals of higher social rank. It has been shown repeatedly that there is an inverse correlation between social rank and plasma levels of cortisol (Winberg and Lepage, 1998; Bessa et al., 2021) and a similar negative relationship has also been reported between social rank and brain serotonergic activity (Winberg et al., 1991). Thus, subdominant animals are less inhibited in their behaviour and appear ready to increase their aggressive behaviour and gain dominance as soon as an opportunity arises. In fact, Øverli et al. (2004) found that in rainbow trout, fish of intermediate social rank increase their aggression towards lower-ranking fish in response to short-term defeat (Fig. 3). However, it is also obvious that there is a negative correlation between the number of aggressive acts received during this short-term defeat and aggression shown towards the subordinate partner (Fig. 3). Thus, being on the receiving end of intense aggression results in behavioural inhibition, whereas being in contact with a dominant, highly aggressive fish, but receiving only low levels of aggression stimulates aggressive behaviour in the smaller subdominant trout (Fig. 3). Interestingly, in the study by Winberg et al. (1992), the subdominant fish (social rank 2) regained aggressive behaviour faster than the subordinates (social rank 3 and 4), but they showed lower levels of aggressive behaviour than these subordinates throughout the experiment (Winberg et al., 1992; Fig. 2).

Winning or losing fights for social dominance sometimes also affects individual visual, acoustic and chemical cues, thereby signalling social experience to conspecifics. These social signals, which may work in tandem, are important in reducing agonistic interactions in established dominance hierarchies. The cichlid Metriaclima zebra uses visual signalling or displays during aggressive interactions and when these are combined with acoustic signals, increased modulation of aggressive behavioural decision making was observed: acoustic signals were used to signal decreased escalation of fights (Bertucci et al., 2010). In salmonids, social subordination results in a darkening of the sclera of the eye, followed by darkening of body coloration (O'Connor et al., 1999; Höglund et al., 2000, 2002). Höglund et al. (2000, 2002) showed that social subordination induces skin darkening in Arctic charr. This skin darkening is slow to develop and seems to represent a morphological colour change induced by chronic stress, especially stress-induced elevation in plasma levels of α-melanocyte-stimulating hormone. The darkening of the sclera that is observed in juvenile Atlantic salmon (Salmo salar) in response to losing fights for social dominance occurs much faster and is most likely under neural control (O'Connor et al., 1999). Similarly, the eye-bar present in many cichlids appears to be under neuronal control since it can be turned on or off very rapidly (Price et al., 2008); however, in this case, the presence of a dark eye-bar is associated with dominance and high levels of aggression. Similarly, in zebrafish (Danio rerio), dominant fish tend to become darker with more obvious stripes, whereas subordinates turn pale (Price et al., 2008). Thus, fishes can quickly signal their social status, which may halt further aggressive interactions.

Chemical communication is common in fish, and chemical signals related to social rank have been demonstrated in cichlids (Keller-Costa et al., 2015). In the Mozambique tilapia (Oreochromis mossambicus), urinary odorants are actively released during agonistic interactions and appear to signal social status (Barata et al., 2007). A similar system has been described in the Tanganyika cichlid Astotilapia burtoni (Maruska and Fernald, 2012). Genomic analysis of the olfactory bulbs and olfactory cortex of the brain in Mozambique tilapia demonstrated specific changes in response to the urine of either dominant or subordinate males (Simões et al., 2015). Olfactory stimuli from individuals with differing dominance status had a major impact on the brain transcriptome, with different urinary social cues prompting specific patterns of transcript expression in the brain. These results illustrate that rapid changes in social status can be signalled to another conspecific, which responds at the level of gene expression in the brain and underpins behavioural plasticity in social status. An interesting point to note is that A. burtoni can undergo dynamic and fast transitions from dominant to subordinate and back again; their behaviour, colouration, transcriptome and stress responses also show a dramatic alteration when social status is altered (Maruska, 2015).

Huntingford (1976a,b) found that there is a positive relationship between boldness and aggressive behaviour in the three-spined stickleback (Gasterosteus aculeatus). Moreover, using zebrafish, Dahlbom et al. (2011) showed that bold males have an advantage in dyadic fights for social dominance. In their experiment, adult zebrafish males were screened for boldness prior to dyadic interaction using a series of behavioural assays. In all cases, boldness, as determined by these behavioural assays, predicted the outcome of fights for dominance. Thus, in addition to size and previous experience, factors such as personality traits appear to be important for the outcome of fights for dominance. Such a relationship between consistent behavioural traits and competitive ability in fights for social dominance is further supported by Schjolden et al. (2005), who showed that in juvenile rainbow trout, there is a relationship between how fast the fish resume feeding after transfer to isolation in a non-familiar environment and aggression. Fish acclimating faster start feeding sooner and display consistently higher attack frequencies and shorter attack latencies when tested in resident/intruder tests. Moreover, rapidly acclimating fish also have lower post-stress cortisol levels and become socially dominant in dyadic fights more often than fish displaying high post-stress cortisol (Schjolden et al., 2005). Fast acclimation to a novel environment, being aggressive and competitive in fights for social dominance along with low post-stress plasma cortisol is typical of a ‘proactive’ stress-coping style. Animals displaying a reactive stress-coping style show the opposite profile: high post-stress cortisol levels, neophobia with slow acclimation to novel environments and low aggression (Koolhaas et al., 2007). These divergent stress-coping styles were originally described in rodents (Koolhaas et al., 2007), but recent studies have shown that teleost fish show similar coping methods (Castanheira et al., 2017; Øverli et al., 2007; Winberg et al., 2007).

In fish, divergent stress-coping styles have been extensively studied in rainbow trout strains selectively bred for high (HR trout) and low (LR trout) post-stress plasma cortisol levels (Pottinger and Carrick, 1999). The LR and HR trout differed in behaviour in a way that could be interpreted as LR trout displaying a proactive coping style, with HR trout being reactive. In line with this, Pottinger and Carrick (2001) showed that LR trout have an advantage over HR trout in staged fights for social dominance in size-matched pairs.

The results from studies on the HR and LR trout clearly show that stress-coping styles are at least in part a heritable characteristic. However, environmental factors are likely to affect the expression of divergent stress-coping styles. For instance, transporting the selected HR and LR trout strains from Windermere to Oslo strongly affected their behavioural profile (Ruiz-Gomez et al., 2008). When tested prior to transport, LR fish showed a more rapid acclimation to a novel environment with low post-stress plasma cortisol and were more likely to become dominant in dyadic fights with size-matched HR trout, whereas HR trout took longer to acclimate, had higher post-stress cortisol and were more likely to end up subordinate in fights with LR trout. Following transport, the behaviour was reversed: HR trout acclimated faster than LR trout and also became dominant in dyadic fights more often than LR trout. However, HPI axis reactivity appeared to be stable since even after transport, LR trout showed lower post-stress cortisol than HR trout (Ruiz-Gomez et al., 2008).

Early studies suggested that boldness/personality and aggressiveness are fixed traits, meaning that individuals would have a permanent phenotype (review in Dingemanse and Réale, 2005). Genetic studies of boldness have shown it to be moderately heritable, although heritability can decline over the lifetime of offspring as a result of exposure to abiotic and biotic factors (e.g. Chervet et al., 2011; Ferrari et al., 2016). However, studies in rainbow trout investigating plasticity in boldness have provided insights into the dynamic nature of bold and shy traits. Shy trout given the experience of winning dyadic contests increase their boldness in subsequent risk-taking tests; thus, success has an emboldening effect (Frost et al., 2007). In contrast, losing fights results in bold trout becoming much shyer. Therefore, the social environment can also influence boldness. Trout which were classified as bold became shyer when placed in groups containing either 100% bold or 100% shy individuals, whereas shy trout remained shy (Thomson et al., 2016). When subject to a stressor, plasma cortisol concentrations reflect the original personality of these trout. Therefore, initially shy individuals had higher plasma cortisol than initially bold trout. These results are indicative of behavioural and physiological parameters of coping styles becoming uncoupled, whereby behavioural changes in boldness are not correlated with stress responses (see Sadoul et al., 2021). In a further study, the impact of predation threat was investigated when rainbow trout were under no, low or high predation risk conditions (Thomson et al., 2012). Under no risk, bold and shy fish retained their original behavioural responses to a novel challenge during the experimental period. However, under low and high risk, bold individuals became shyer, demonstrating their personality was labile and able to respond to predation. Shy individuals remain shy despite the risk; of course, being shy and cautious in nature in the face of the risk of predation is adaptive. Although bold fish become shyer under low risk, they still had a much lower cortisol response compared with shy fish, demonstrating again a decoupling of stress reactivity and risk-taking behaviour (Thomson et al., 2012). The dynamic nature of risk taking and behavioural flexibility in bold rainbow trout contrasts with studies conducted on the same species in other geographical locations. Several of these studies have shown that bold phenotypes are inflexible and that it is the shy personality that exhibits flexibility in behaviour (Moreira et al., 2004; Ruiz-Gomez et al., 2011; Johansen et al., 2012; Höglund et al., 2017). These disparate findings between laboratories in different countries may be due to genetic differences or it is possible that the previous experience of the fish also differed substantially. To dissect this, it would be useful to use the same animals from one location in studies conducted in geographically distant laboratories.

A social decision-making network (SDMN) has been identified in the vertebrate brain and in teleosts. The SDMN includes brain areas such as ventral pallium (Vs, the homologue of the mammalian extended medial amygdala), ventral nuclei of ventral telencephalic area (Vv, homologue of mammalian lateral septum), pre-optic area (POA, homologue of mammalian POA/paraventricular nucleus), ventral tuberal nucleus (vTn, homologue of mammalian anterior hypothalamus), anterior tuberal nucleus (aTn, homologue of mammalian ventro medial hypothalamus) and griseum centrale (GC, homologue of mammalian periaqueductal grey). These brain areas are interconnected and to some degree overlap with the mesolimbic reward system, for example, the Vv and Vs in teleosts (O'Connel and Hoffman, 2011a, b). The SDMN expresses sex steroid hormone receptors and controls multiple social behaviours, such as aggression, reproduction and parental care. However, it appears that an animal's behavioural profile is determined by the overall activity of this network and not by specific individual brain areas (Teles et al., 2016a, b). The activity of the network is affected by developmental factors, hormone levels, reproductive status, but also by the social information received, i.e. by social experience (O'Connel and Hoffman, 2011a, b). For instance, in zebrafish, winning or losing fights for social dominance is reflected in divergent expression patterns of immediate early genes in the SDMN, which in turn are paralleled by changes in plasma sex steroids and brain levels of arginine vasotocin (AVT) and isotocin, the teleost homologues of arginine vasopressin and oxytocin, respectively (Teles et al., 2016b; Almeida, et al., 2019).

The brain monoamines, dopamine (DA), noradrenaline (norepinephrine) and serotonin (5-hydroxytryptamine, 5-HT), appear to have modulatory effects on the SDMN (Soares et al., 2018). Agonistic interactions result in a rapid activation of these neurotransmitter systems (Backström and Winberg, 2017). A brain area that interacts with the brainstem monoaminergic systems and highly important in controlling agonistic behaviour is the habenula (Hb). This diencephalic brain area, which is conserved across vertebrates, seems to be a satellite of the SDMN (Ogawa et al., 2021). In zebrafish, the medial subregion of the dorsal habenula (dHbM) projects to the median raphe (MR) via the ventral interpendicular nuclei (IPN), whereas the lateral subregion of the dorsal habenula (dHbL) projects to the GC via the dorsal IPN (Okamoto et al., 2020; Nakajo et al., 2020). These two pathways of the habenula act like a switch. Socially dominant zebrafish display a more intense activation of the dHbL–GC projection, whereas fish losing fights for social dominance show elevated activity of the dHbM–MR pathway (Okamoto et al., 2020). MR is one of the major serotonergic brain areas and 5-HT has been found to play an important role in coordinating behavioural, autonomic and neuroendocrine stress responses, as well as mediating the behavioural inhibition characterizing socially subordinate animals (Backström and Winberg, 2017). The GC, on the other hand, is rich in dopaminergic neurones. To some degree, dopamine DA and 5-HT have opposite effects on agonistic behaviour.

When establishing a dominance hierarchy, both winners (i.e. animals becoming dominant) and losers (i.e. animals becoming subordinate) show a rapid activation of the brain 5-HT system (Øverli et al., 1999). In this initial phase of agonistic interaction, both winners and losers are actively performing agonistic behaviour, showing no signs of behavioural inhibition. However, brain serotonergic activity gradually returns to control levels in animals becoming dominant, whereas it remains elevated in subordinate animals even after long-term social interaction (Winberg and Nilsson, 1993; Øverli et al., 1999). Subordinates also show elevated plasma cortisol levels (Winberg and Lepage, 1998; Øverli et al., 1999) and a pronounced behavioural inhibition (Winberg et al., 1992). The chronic elevation of brain serotonergic activity is likely to be part of the mechanism mediating this behavioural inhibition (Winberg and Thörnqvist, 2016; Backström and Winberg, 2017). Interestingly, only long-term activation of the serotonergic system results in behavioural inhibition. This is also obvious when brain serotonergic activity is stimulated by elevated dietary intake of L-tryptophan (TRP), the amino acid precursor of 5-HT (Winberg et al., 2001; Lepage et al., 2003; Höglund et al., 2019). In rainbow trout, fish fed elevated TRP show an inhibition of aggressive behaviour but only after being fed this TRP-supplemented feed for 7 days, even though elevated dietary TRP has rapid effects on brain serotonergic activity. Thus, chronic elevation of brain serotonergic activity most likely results in changes in brain 5-HT functions related to altered expression of 5-HT receptors, re-uptake transporters and the balance between pre-synaptic auto-receptors and post-synaptic receptors. Such plastic changes may result in elevated 5-HT tone in specific brain regions (Winberg and Thörnqvist, 2016; Backström and Winberg, 2017; Höglund et al., 2019). Moreover, behavioural effects of 5-HT may also be mediated in concert with other neuromodulators, e.g. neuropeptides.

AVT and corticotropin-releasing factor (CRF) are involved in the control of the HPI axis and have been suggested to modulate agonistic behaviour in teleosts (reviewed by Backström and Winberg, 2017). Larson et al. (2006) observed that dominant zebrafish have higher numbers of AVT immunoreactive (AVTir) cells in the magnocellular POA (mPOA) than subordinates, and that magnocellular AVTir cells were also larger in dominant than in subordinate zebrafish. However, in the parvocellular POA (pPOA), the reverse relationship was observed, with subordinate zebrafish having larger and higher numbers of AVTir cells than dominant fish (Larson et al., 2006). By contrast, in Mozambique tilapia, subordinate (non-territorial) males showed larger AVTir cells in the mPOA than territorial males, and even though the number of pPOA or mPOA AVTir cells did not differ between non-territorial and territorial males, non-territorials displayed a higher number of gigantocellular AVTir cells (Almeida and Oliveira, 2015). The significance of this observation is unclear, even though Almeida and Oliveira (2015) suggested that it could be related to an association between the HPI axis and the gigantocellular AVT nucleus. Interestingly, in Astotilapia burtoni, another closely related African cichlid species with a mating system similar to that of the Mozambique tilapia, territorial males showed higher expression of AVT mRNA in the gigantocellular nucleus than non-territorial males, and AVT mRNA levels correlated with aggressive behaviour (Greenwood et al., 2008). Thus, even though AVT appears related to agonistic behaviour, the role of this peptide in controlling behaviour is likely to vary between species and contexts (Teles et al., 2016a,b). For example, dominant Neolamprologus pulcher have higher brain AVT gene expression than do subordinates (Aubin-Horth et al., 2007), but subordinates have higher levels of free AVT peptide, the biologically active form (Reddon et al., 2015). An interaction between 5-HT and AVT has been suggested, and Semsar et al. (2004) showed that fluoxetine, which reduces aggression in male bluehead wrasse (Thalassoma bifasciatum) (Perreault et al., 2003), also reduces the expression of AVT mRNA in this species. However, in rainbow trout, intracerebroventricular (ICV) injections of AVT had no effect on brainstem 5-HT (Backström and Winberg, 2009), even though Gesto et al. (2014) reported that ICV injections of AVT increased serotonergic activity in the telencephalon and hypothalamus of rainbow trout 8 h following injection. In the study by Backström and Winberg (2009), the fish were sampled following 60 min of dyadic agonistic interactions and since aggressive encounters result in an immediate activation of the brain 5-HT system, this socially induced activation of the 5-HT system may have concealed potential effects of AVT.

Interactions between 5-HT and CRF have been studied in more detail and it is clear that 5-HT interacts with CRF in the control of several behaviours (Backström and Winberg, 2017). Both stimulating and inhibitory effects of CRF on aggressive behaviour have been reported in mammals (see reviews by Backström and Winberg, 2013, 2017; Hostetler and Ryabinin, 2013). In rainbow trout, ICV injections of CRF have been shown to either increase or decrease dominance. For instance, Carpenter et al. (2009) found that ICV injections of CRF suppressed attack behaviour but increased the ratio of attacks to retreats, which led to dominance (Carpenter et al., 2009). Backström et al. (2011), on the other hand, reported that rainbow trout receiving ICV injections of CRF lost dyadic fights. This discrepancy could be due to differences in interaction times and experimental design. Carpenter et al. (2009) used higher doses of CRF and allowed the fish to interact for 15 min, whereas in the study by Backström et al. (2011) they interacted in pairs for 60 min. Both studies quantified effects on the brain monoaminergic systems. As in the studies of AVT discussed above, the fish were interacting in pairs, which most likely made a stimulatory effect of CRF on 5-HT activity difficult to detect. In agreement with this suggestion, neither Carpenter et al. (2009) nor Backström et al. (2011) detected any effects of CRF on brain 5-HT activity. However, ICV injections of the CRF antagonist α-helical-CRH1-41 decreased 5-HIAA and 5-HT in the brainstem (Backström et al., 2011).

Spexin (SPX), a novel neuropeptide also known as neuropeptide Q, which is highly conserved across the vertebrate subphylum, is another neuromodulator that appears to interact with the 5-HT system in the control of stress and agonistic behaviour (Lim et al., 2019). In teleost fishes, SPX occurs as two orthologues, SPX1 and SPX2, with divergent brain expression patterns. In zebrafish, SPX1 is expressed in ventral pallium, dHB and hindbrain, whereas SPX2 is expressed in the POA. SPX acts on the galanin receptors and, in zebrafish, SPX1 and SPX2 activate galanin receptor 2a and 2b. Chronic social stress results in an upregulation in the expression of SPX1a and SPX1b in the brain of Nile tilapia (Oreochromis niloticus) (Lim et al., 2020). As discussed above, the dHb appears to be important in the control of agonistic behaviour via its projections to MR via the IPN (Okamoto et al., 2020). Transgenic zebrafish overexpressing SPX1 in the dHB show anxiolytic behaviour compared with wild-type siblings (Jeong et al., 2019). Moreover, overexpression of SPX1 also results in an upregulation of the expression of 5HT-related genes in the MR (Jeong et al., 2019).

Aggressive behaviour can differ over a fish's lifetime. For example, marine yellowtail, Seriola quinqueradiata (Carangidae), do not engage in agonistic interactions during the larval period [until 10 mm in total length (TL)] ∼20 days after hatching (Sakakura and Tsukamoto, 1998). Aggressive behaviour is first observed just after metamorphosis to the juvenile period, which is accompanied by an elevation of tissue cortisol levels. The onset of schooling behaviour is observed at 12 mm TL. Within the juvenile schools, three categories of social rank were characterised, including dominants (10–20%), subdominants (10–20%) and subordinates (60–80%) (Sakakura and Tsukamoto, 1998). Similarly, in Japanese flounder (Paralichthys olivaceus), aggression was not recorded during the larval stage [13.2 mm in standard length (SL)] (Sakakura and Tsukamoto, 2002). Aggressive behaviour was first observed on day 39 after hatching, coinciding with completion of metamorphosis from the larval to the juvenile stage, and increased until day 46. Therefore, there are clear age-related changes in aggression that need to be explored.

Early rearing conditions can also influence adult aggressive behaviour in many animals (Holekamp and Strauss, 2016). Adverse conditions during development can mean individuals become more aggressive, and this persists into adult life. Studies in mammals show that early life adversity, including repeated maternal separation and neglect, conflict between parents, post-weaning social isolation and neglect, can influence later life aggression. In Siamese fighting fish (Betta splendens), individuals were subjected to social isolation as juveniles and sub-adults, with fish reared in social groups becoming much less aggressive (Iwata et al., 2021). Threatening behaviour was greater in adult fish isolated as sub-adults, whereas fighting behaviour was significantly higher in adults isolated as juveniles. The influence of rearing conditions on behaviour was greater in females than in males, with female aggression more affected by a lack of enrichment (i.e. held in barren tanks) in the environment. Adults isolated as sub-adults had higher cortisol levels after aggression tests and, additionally, males had higher plasma concentrations of the androgen 11-ketotestosterone than those males reared in a group or isolated at the juvenile stage (Iwata et al., 2021). Social isolation in males also elicits intense fighting and elevated plasma cortisol concentrations after contests in the territorial convict cichlid Archocentrus nigrofasciatus (Earley et al., 2006). Thus, rearing environment can influence aggression, and affects the HPI and HPG axes underlying stress and reproduction. Many other factors can influence aggression in later life. Increased population density and the availability of shelter increase aggression in gopher rockfish, Sebastes carnatus (Hoelzer, 1987). Predation risk influences the aggressiveness of sticklebacks, with those reared in a risky environment being less aggressive (Herczeg et al., 2016).

Maternal behaviour and maternal rank can also significantly influence offspring aggressiveness in fish (D'Amore et al., 2015). Sons of well-fed swordtail fish, Xiphophorus multilineatus, form a positive correlation between aggression and risk taking or boldness, but levels of aggression and boldness are decoupled in offspring from poorly fed females (D'Amore et al., 2015). When offspring from Atlantic salmon mothers treated with cortisol implants were compared with untreated offspring for dominance behaviour, there were differences in both aggressiveness and stress (Eriksen et al., 2011). Offspring from mothers treated with cortisol gained a high dominance rank and were more aggressive after dominance–subordinate relationships were established when compared with offspring from mothers not treated with cortisol (Eriksen et al., 2011). Female guppies, Poecilia reticulata, exposed to mild stress gave birth to offspring that are more aggressive than those from unstressed mothers (Eaton et al., 2015). Thus, the state of the mother has a profound influence on offspring behaviour and physiology, with stressed mothers producing more aggressive and possibly competitive offspring, which would be adaptive in a harsh or challenging environment.

Aggressive and agonistic interactions and the formation of dominance hierarchies occur naturally in wild populations. However, individuals can choose whether to engage in these potentially stressful and energetically costly behaviours and decide when to retreat from an opponent. This is not entirely possible in captive situations where fish are held in a relatively small volume of water and cannot necessarily retreat or escape from an aggressor. Thus, dominance behaviour and aggressive interactions can present a significant problem for low-ranking individuals owing to their low social status, potentially resulting in chronic stress. Many examples of dominance hierarchies described above are from species held in aquaculture (e.g. salmonids), used in the ornamental fish trade and held in home tanks, ponds or public aquaria, or used as laboratory models. Within these contexts, confined environments do not allow escape and researchers have speculated that the high aggression seen in captivity is an artifact (Sloman and Armstrong, 2002). Consequently, although aggression and dominance hierarchies are seen in nature, these may not result in the profound stress responses and elevated cortisol or in the neurobiological differences measured in laboratory-based studies (Sloman and Armstrong, 2002). However, in the natural environment, escape may be difficult in relatively small environments such as a pond or where predation risk is high, making it impossible for individuals to leave the group. Thus, housing aggressive species in close confinement or at high stock densities where lower-ranking individuals cannot seek shelter for respite or escape from the attentions of dominants is likely to be a significant challenge to animal welfare.

Aggression is a significant problem in the aquaculture of fishes. For example, juvenile salmonids aggressively defend feeding territories in nature, which becomes problematic in the close confines of the aquaculture environment, particularly during periods of feed restriction (Brännäs and Alanärä, 1994). Farmed salmonids and other species have observable dominance hierarchies (e.g. Atlantic salmon, rainbow trout), in which subordinates experience chronic stress measured in HPI and serotonergic activity, but HPG activity is reduced (Cubitt et al., 2008). When food is provided in a predictable manner, these relationships can lead to aggressive behaviour, injury and differential feed acquisition, with dominant animals monopolising food, resulting in growth disparity between fish (e.g. Cubitt et al., 2008, Montero et al., 2009). High levels of aggression result in more severe fin damage (e.g. Cañon Jones et al., 2010), with the dorsal and caudal fins mainly affected (Turnbull et al., 2008; Persson and Alanärä, 2014). To overcome differential growth, size grading is often used in fish farms to remove larger individuals; however, this results in subdominants taking the place of dominants so that size grading must be repeated. Handling and size grading have been shown to be a stressor in numerous species, for example, in the red sea bream Pagrus major (Biswas et al., 2006), Atlantic salmon (McCormick et al., 1998) and common carp (Saeji et al., 2003). An additional issue to consider is the variable reactions of individuals with differing stress-coping styles such that proactive, bold individuals with a low cortisol response could potentially tolerate or cope with stress in the aquaculture environment much better than reactive, shy fish.

Increased competition over food, increased boldness or even an individual being unable to withdraw from a contest through lack of space may elevate levels of aggression (Huntingford, 2004). Solutions to this problem may be challenging, particularly in fish farms focusing on high productivity. Reducing density and increasing space may limit aggression, although proposed densities to minimise aggressive behaviour varies between studies (Gronquist and Berges, 2013), suggesting that other issues may also be important. Provision of shelter or refuge for fish may also minimise the impact of aggressive encounters, as well as ensuring that space within tanks or sea cages is fully utilised (Huntingford and Kadri, 2014). For example, sea bream have impaired learning ability, brain development and spatial orientation in barren tanks, which are all improved by the provision of environmental enrichment (increasing the heterogeneity or complexity of the living space; Arechavala-Lopez et al., 2019, 2020). Provision of ropes also reduces aggression in sea bream held in sea cages (Arechavala-Lopez et al., 2019, 2020). Future studies should focus on the use of enrichment tools to allow lower-ranking individuals to escape or hide from more aggressive, dominant individuals. This has the potential to reduce stress and improve the welfare of fish with low social status.

Ornamental fish are kept or reared for their attractiveness as companion animals or as exhibits in public aquaria for educational or conservation purposes. It is estimated that 1 in 10 households possess pet fish in the UK (https://www.pfma.org.uk/statistics) and in the USA (Davenport, 1996), with an estimated 100 million housed in tanks and outdoor ponds in the UK alone (https://ornamentalfish.org/aquatic-shops-face-animal-welfare-nightmare-as-energy-prices-rocket/). Approximately 2000 freshwater and marine fish species are kept as companion animals or held in public exhibits such as aquaria or zoos (https://www.liveaquaria.com/article/159/?aid=159). The global trade in ornamental fish is estimated to amount to approximately $5.4 billion (https://www.grandviewresearch.com/industry-analysis/ornamental-fish-market) and is therefore a lucrative industry. Fishes can be readily purchased from numerous retailers and so the likelihood of purchasing aggressive or territorial species to keep in a home tank or pond is high. Aggressive or territorial fishes can harass other fish within a tank community, and this could represent chronic stress and may result in ill health or mortality. Sneddon and Wolfenden (2019) make a variety of suggestions with regards to the housing of aggressive and territorial species. The recommendations include keeping territorial species individually to avoid territories overlapping thereby reducing aggressive behaviour and considering a variety of species that can live together as a ‘community’: for example, the presence of compatible species such as freshwater angelfish, neon tetras, white cloud mountain minnows and tiger barbs leads to no or low aggression between and within these species (Sloman et al., 2011). Social isolation can be stressful for some species, so housing territorial fish completely alone might not be good for welfare and including non-aggressive smaller fishes may be beneficial and provide non-threatening companionship. Knowledge of the social or aggressive behaviour of potential ‘pet’ fish is essential to ensure good welfare.

In the laboratory environment, similar problems to those outlined above exist for aquaculture and ornamental fish-keeping (Sloman et al., 2019). Limited space, inappropriate stocking densities, mixing of unfamiliar individuals and lack of environmental enrichment may all lead to problems with aggression. Again, environmental enrichment has been shown to reduce aggression in some cases. For example, high-level enrichment (8 structures within a tank) in both rockfish (Sebastes schlegelii) and greenling (Hexagrammos otakii) reduced aggression and stabilized the social hierarchy after 3 days (Zhang et al., 2021). Batzina and Karakatsouli (2012) found that gilthead seabream held in tanks with blue or red/brown glass gravel compared with a control (no gravel) or green gravel grew better and were less aggressive. In zebrafish, contradictory results have been obtained on whether enrichment reduces aggression. Some studies have shown a clear reduction in aggressive behaviour, whereas others have shown no change or an increase (reviewed by Stevens et al., 2021). Provision of a refuge during individual testing reduced aggression in zebrafish (Weber and Ghorai, 2013). Moreover, anchored and floating artificial plants and gravel reduced aggression in group-held zebrafish (Carfagnini et al., 2009), as did 12 strips of plastic arranged 3×4 to simulate vegetation (Basquill and Grant, 1998). Zebrafish form highly aggressive dominance relationships in pairs and provision of a floating plastic plant reduced aggressiveness and the occurrence of wounds (Keck et al., 2015). Caution should be applied to using enrichment in that no one size will fit all and enriching the environment may encourage territoriality and aggression in some species of fish. Each species may have different preferences and this may be further affected by age and sex.

Social rank greatly affects behaviour, physiology and performance of fish. High-ranking individuals are active, aggressive and more likely to win future fights for social dominance. Heritable factors, in part, control the stress-coping style of individual fish and thereby also affect competitive ability and the chance to obtain dominance. However, coping styles appear more plastic than previously realised, and environmental factors, especially social experience, largely affect the coping style. Thus, intraspecific behavioural and physiological traits appear to be controlled by a complex interaction between genetic and environmental factors (Fig. 4). In nature, the development of dominance hierarchies may act to reduce overt aggression, but in fish kept in captivity, it may result in severe chronic stress, compromising fish welfare. The impact of dominance and stress-coping style needs further investigation to truly understand how the captive environment affects the welfare of individuals within groups and to inform better husbandry and housing of these animals.

We would like to thank three anonymous referees for valuable comments on the manuscript.