Effects of social environment and energy efficiency on preferred swim speed in a marine generalist fish, pile perch (Phanerodon vacca)

The effective use of movement is essential for survival in many animal species. Movement is used to locate food, find mates, explore the environment, defend territories and avoid danger, at times simultaneously (Fortin et al., 2005; Gilroy and Lockwood, 2012; Pyke, 1978; Rosell et al., 1998; Teyke, 1989). Animal locomotion results from a continuous series of choices informed by the summation of biotic, abiotic and endogenous inputs received by an organism. These inputs might include an individual's perception of its proximity to predators, mates or other organisms (Hemmi, 2005; Van Dyck and Baguette, 2005), the local distributions of food and shelter (Tang and Bennett, 2010), and the energetic costs associated with acting on this information (Shepard et al., 2013). Determining the patterns (e.g. speed, trajectories) and underlying causes (e.g. finding a better environment, escaping from threats) of organismal motion requires an understanding of these varied inputs, as well as how they interact with one another.

The energetic cost of movement varies nonlinearly with an animal's speed (Wilson et al., 2015); as fish swim at increasing speeds, their oxygen consumption rises exponentially (Korsmeyer et al., 2002). Experiments have consistently shown that fish have an optimal speed, or U_opt, defined as the speed at which the mass-specific energetic cost of moving a given distance is minimised (Pyke, 1981; Tucker, 1970; Weihs, 1973). U_opt is an individual's most energy-efficient swimming speed, accounting for metabolic rate, body size and shape, and likely other morphological and physiological parameters. Therefore, U_opt varies among species according to habitat and life history strategy (Tucker, 1970; Tudorache et al., 2008; Wakeman and Wohlschlag, 1981; Ware, 1978; Weihs, 1973). In fish species, U_opt may vary from as low as 1 to as high as 5 body lengths (BL) s⁻¹ (Palstra et al., 2020; Tudorache et al., 2008, 2011).

Although the benefits of energy conservation promote moving at or near an individual's U_opt, the speed at which fish choose to swim (U_pref) is often driven by additional factors. For example, habitat and feeding strategy may influence U_pref. Fishes occupying complex habitats may tend to swim more slowly than their U_opt in order to avoid obstacles (Priyadarshana et al., 2001), and fishes with sedentary hunting strategies (i.e. ambush predators) spend the majority of their time completely immobile (Armstrong, 1986; Tolley and Torres, 2002). Further, behavioural tasks more complex than moving along a straight path, such as searching for food, often rely on sensory inputs via visual and olfactory cues. These tasks may not be optimised while swimming constantly, requiring deviation from U_opt (Kramer and McLaughlin, 2001). Fish also display considerable intraspecific behavioural variation, which may cause further variation in U_pref. Empirically derived associations between stable physiological and behavioural traits, termed ‘pace of life syndromes’, can provide an energetic basis for this variation in some cases (Vasilieva, 2023). For example, individuals with greater metabolic rates often display a greater propensity for exploratory behaviour than individuals with slower metabolic rates (Binder et al., 2016; Fu et al., 2021), potentially exacerbating intraspecific variation in U_pref. Despite these influences, maximising energy efficiency by aligning U_pref with U_opt may provide significant fitness benefits to pelagic and generalist species prone to continuous movement (Tudorache et al., 2011). In particular, migratory species undertake prolonged journeys characterised by constant swimming, and without the frequent responses to sensory inputs necessary for other behavioural states such as foraging, searching for mates or navigating a structurally complex environment. Migratory species then stand to gain from conserving energy by aligning U_pref with U_opt. The brook trout (Salvelinus fontinalis) has been experimentally shown to choose energetically optimal swimming speeds when presented with a flow gradient (Tudorache et al., 2011), and the volitional (i.e. preferred) swimming speed of captive sharks (Carcharhinus leucas and C. plumbeus) was shown to match closely their theoretical U_opt (Weihs et al., 1981). However, U_pref and U_opt may not match as closely in non-migratory, non-pelagic species, as their swimming behaviour is likely to be related to factors other than energetics, such as sensory awareness (Kramer and McLaughlin, 2001). In largemouth bass (Micropterus salmoides), U_pref was found to generally be below U_opt, with this study speculating that slower swimming speeds may improve foraging efficacy (Han et al., 2017). Large fish such as basking sharks (Cetorhinus maximus) have also been found to swim below their optimal swimming speed while feeding (Sims, 2000). Analysis of multiple shark species found that both body length and tail size are key predictors for cruising speed, indicating that both energetics and ecology, which relates closely to tail length, inform U_pref (Ryan et al., 2015).

An individual's social environment may also determine the extent to which U_pref matches U_opt. Group living, in the form of shoaling or schooling behaviour, is present in many fish species at some point in their life history (Pitcher, 1983; Pitcher et al., 1982), and group members may derive significant advantages from swimming as a group. These advantages include improved foraging potential, predation protection and the possibility of finding a mate (Krause and Ruxton, 2010; Pitcher et al., 1982; Ward and Webster, 2016; Wright et al., 2006). To maintain cohesion as a group, the swimming speeds of group members must respond to the movement of those around them (Herbert-Read et al., 2011). Individuals within social groups may therefore choose to compromise their U_pref to align with the rest of the group. Members of some social species have been found to alter their preferences for abiotic parameters such as temperature in order to maintain proximity to conspecifics (Cooper et al., 2018; Nay et al., 2021); however, whether the presence of a conspecific disrupts the relationship between U_opt and U_pref remains unclear. As different individuals are likely to have different U_opt and U_pref values, determining whether a fish will sacrifice social proximity to swim at their preferred speed or will sacrifice their preferred swimming speed in favour of social proximity is a key consideration for social species (Jolles et al., 2017).

The present study aimed to characterise the relationship between U_opt and U_pref in a social, non-migratory species, the pile perch, Phanerodon vacca (formerly Rhacochilus vacca) (Longo et al., 2018). Phanerodon vacca was chosen for its complex coastal habitat, gregarious nature and non-migratory life history (Munsch et al., 2016). We used a Steffensen-type swim tunnel respirometer and a novel circular flow-gradient tank to address whether U_opt and U_pref align in a marine generalist species that does not migrate, as well as how a social species might prioritise energetically optimal swimming compared with social proximity. We predicted that: (1) U_pref would be positively correlated with U_opt; (2) U_pref would be lower than U_opt in non-migratory P. vacca; and (3) mean U_pair, the average swimming speed of fish in the paired trial, would represent a compromise between the U_pref of each fish in a pair, and would therefore be intermediate between U_pref values for each fish.

Pile perch, Phanerodon vacca (Girard 1855), were collected between 18 July and 13 August 2023 from Jackson Beach, San Juan Island, WA, USA (N 48.51995, W 123.01105), and held under IACUC permit number 4238-03 for Friday Harbor Laboratories. Fish were caught by seine netting from the shore. Individuals with masses of at least 30 g were selected for study; individuals less than 30 g were not used owing to the increased noise associated with oxygen uptake measurements in a large respirometer. Fish were transported to the University of Washington Friday Harbor Laboratories, where they were maintained in a flow-through tank (130×60×15 cm, filled with seawater to a depth of 9 cm) with continuously refreshing seawater at ambient temperature (approximately 14°C). Fish were acclimated to aquarium conditions for at least 48 h. Prior to respirometry, morphological measurements (total length, body depth, body width) were collected to account for the solid blocking effect (Bell and Terhune, 1970; Kline et al., 2015) of each fish on swim tunnel flow speed. Data were collected for a total of 16 individuals, comprising 16 respirometry trials as well as 16 individual and seven paired swimming trials as described below. Data were collected from 2 to 16 August 2023. To validate our methods, two pilot fish were also run between 31 July and 2 August 2023.

Fish were placed in a Steffensen-type 8.5 l swim tunnel respirometer (Steffensen et al., 1984) with a cross-sectional area of 99 cm² (Fig. 1A). Swim tunnel flow rate was calibrated using a Höntzsch flowmeter and SEW Eurodrive voltmeter. Temperature was maintained at a constant 14°C using a thermostat and Thermo Fisher Scientific cooler, setpoint 14°C with a hysteresis of 0.1°C. Oxygen saturation was measured using a Fibox 3 oxygen meter. Intermittent flow respirometry was carried out in cycles of 10 min, comprising a 270 s flush period, 30 s wait period and a 300 s measurement period. Fish were acclimated in the swim tunnel for 6–12 h at a constant swim speed of 0.5 BL s⁻¹. Respirometry data were collected using the software Autoresp (Loligo Systems, Denmark). Following acclimation, swim tunnel speed was increased by increments of 0.5 BL s⁻¹ every 30 min. For each speed increment, the fish was recorded with a Logitech camera for a minimum of 30 s to capture pectoral fin beat frequency as pectoral fin beats are consistent during navigational swimming and primarily responsible for locomotion in this species (Mussi et al., 2002). The camera was mounted approximately 40 cm above the tank and a mirror was placed at a 45 deg angle at the side of the tank to simultaneously record dorsal and lateral views of the fish (Fig. 1A). Speed increase continued until the fish ceased swimming and rested against the back grill of the swim tunnel for at least five consecutive seconds (critical swimming speed; U_crit). At this point, the flow rate was decreased back to 0.5 BL s⁻¹ and oxygen consumption was recorded until it returned to acclimation levels, to determine the oxygen debt of the fish. After fish returned to pre-trial oxygen consumption rates, approximately 40 min post-exhaustion in most cases, each fish was moved to the free swim trial arena. To determine background respiration, oxygen measurements were taken before and after each respirometry trial using a 600 s measurement period and otherwise the same settings. Background respiration remained stable during each measurement so an assumed linear growth between pre- and post-trial background respiration was used to determine the real oxygen consumption of each fish.

Note that there is a time conversion as Ṁ_O2 was per hour and swim speed is per second.

Optimum swimming speed (U_opt) was defined as the speed at which COT is minimised (Fig. 2) and was obtained by setting the first derivative to zero.

Fish were sorted into pairs at random. Data were collected for each pair of fish over 3 days. The A individual within a pair underwent respirometry on day 1 and a preferred swimming trial on day 2, then was placed alone into a holding tank to distinguish it from other study fish until day 3. The B individual within a pair underwent respirometry on day 2 and a preferred swimming trial on day 3. Immediately after completing the individual preferred swimming trial on day 3, the A fish was retrieved from the holding tank and placed back into the free-swimming tank with the B fish (Fig. 4). In the case of pairs 3 and 4, paired trials were delayed because of mortality in the A, and in the case of pair 4, B individuals prior to paired trials being carried out. New individuals therefore had to be included in pairs and were given IDs of C and D to account for any effects on fish caused by the delay. The two individuals were then allowed 4 h to acclimate to the tank and to one another, at which point the small pump was reconnected. Following an additional 30 min of acclimation, a 1-h paired preferred swimming speed trial was recorded.

Free-swimming trials were recorded at a frame rate of 30 frames s⁻¹, saved as mp4 files, and subsequently converted to binary data ‘pv’ files for analysis using the animal tracking software Trex (Walter and Couzin, 2021). Trex was used to determine the position of each fish in each video, with coordinate data exported in pixels. Data were then converted to cm coordinates as defined by the flow speed heatmap. Coordinates from one frame per second were utilised in analyses. For paired trials, neighbour distances were also exported from Trex and converted to centimetres by multiplying by the conversion factor, found in Trex for each video. To find the flow speed experienced at each point in the trial according to their coordinates, heatmap coordinate data and trial coordinate data were each converted to separate spatial points data frames using the sp package in R (Bivand et al., 2013). Trial coordinates were then interpolated to heatmap coordinates using the dplyr package (https://CRAN.R-project.org/package=dplyr), allowing speed data for the trial coordinates to be extracted based on known speed values in the heatmap. These data were converted to Excel using the clipr package (https://CRAN.R-project.org/package=clipr).

Videos were also manually reviewed to assess the percentage of time each fish spent steady swimming as opposed to moving around the tank. Time spent steady swimming was defined as when the fish stayed in approximately the same position within the flow. Non-steady swimming was defined as when: (a) the fish moved more than 90 deg (one-quarter of the circumference of the tank) forward or backward in the tank in a continuous movement in under 10 s, while still facing into the flow; (b) the fish turned around and swam with the flow, regardless of how far it swam with the flow; or (c) in paired trials, fish overtook one another. Non-steady swimming times were then excluded before the modal swim speed for each trial was calculated.

In paired trials, each individual's position relative to its partner was determined based on the position of both fish. When fish were more than two body lengths apart, position was characterised as ‘apart’. If fish bodies overlapped laterally by at least one-third of the body length of the smaller fish, their position was coded as ‘alongside’. When fish swam in line with each other or with less than one-third of body length overlapping laterally, their position was coded as ‘ahead’ or ‘behind’. Modal swim speed for each position adopted by an individual throughout the trial was then calculated. When an individual adopted a given position for less than 5% of the trial (3 min, ∼150 data points) this position was excluded from the dataset owing to inability to determine a singular modal speed in a number of cases.

Fin beat frequency for each fish in both individual and paired trials was determined by counting fin beats for ten 1-min periods at regular intervals (every 6 min, beginning at the start of the trial) during both individual and paired swimming trials. Where fish did not swim steadily for 1 min at these intervals, fin beats were counted for at least 30 s at the nearest point where 30 s of continuous steady swimming could be found. Mean fin beat frequency (beats min⁻¹) for each individual in both paired and individual trials was then calculated. It was not possible to calculate amplitude of fin beats, either in the swim tunnel or in free-swimming trails, as the video footage was not of sufficient quality to allow fin beat amplitude to be accurately measured.

All models were constructed in R (version 4.2.2). Data were visually checked for normality and homoscedasticity of variance using the qqnorm() and hist() commands in R prior to modelling. Individual U_pref was cube root transformed to improve residual distribution. To determine whether U_opt had a significant effect on U_pref, a linear model was constructed using the lme4 and lmertest packages (Bates et al., 2015; Kuznetsova et al., 2017), with U_pref from individual trials as the resultant variable. The explanatory variables were U_opt, fish total length and fin beat frequency from the free-swimming trial. The interaction U_opt×fish total length was also included. Models with and without interaction terms were compared using Akaike's information criterion (AIC) to find the best fit model (https://cran.r-project.org/package=AICcmodavg). The interaction effect was not retained in the best fit linear model. A paired sample t-test was carried out on cube root transformed data to determine whether U_opt differed significantly from U_pref.

To determine whether the presence of a conspecific would affect swimming speed, U_pref of each individual in a pair was compared with mean U_pair between both fish. This was calculated by dividing the sum of the modal speed for each fish by 2, meaning that if fish A had a modal speed of 6 cm s⁻¹ and fish B had a modal speed of 8 cm s⁻¹ in the paired trial, mean U_pair would be 7 cm s⁻¹. Mean U_pair in cm s⁻¹ was converted to a percentage relative to the U_pref (also in cm s⁻¹) of each fish in the pair, where the value of U_pref for the faster fish was set as 100% and the value of U_pref for the slower fish was set as 0%. Mean U_pair was then calculated as a percentage change based on this scale. For example, if the U_pref of the slower fish in a pair were 8 cm s⁻¹, the U_pref of the faster fish were 12 cm s⁻¹ and the mean U_pair were 10 cm s⁻¹, the percentage value for U_pair would be 50%. If mean U_pair were 7, the percentage value would be −25%. Two one-way t-tests were then carried out on mean U_pair percentage data, to determine whether these values differed significantly from the set values of 0 and 100, representing the U_pref of each fish in the pair. A linear mixed model was also constructed in order to analyse the effects of fish position relative to conspecific, and fin beat frequency in paired trials on U_pair. The response variable for this model was U_pair, which was log transformed to improve residual distribution. The explanatory variables were mean nearest neighbour distance (NND), position relative to conspecific and mean fin beat frequency. The interactions between mean fin beat frequency×position, mean fin beat frequency×NND and position×NND were also included. Pair ID and fish ID were included as random effects. Models with and without interaction terms were compared using AIC to find the best fit model (https://cran.r-project.org/package=AICcmodavg). The best fit linear model did not include any interaction effects. An additional model was constructed to compare the correlation between fin beat frequency and speed in the swim tunnel compared with in the free-swimming tank, in case differing hydrodynamics in the two arenas meant that fish had to expend more effort to maintain the same swim speed in one or the other. The response variable in this model was swimming speed, and the explanatory variables were mean fin beat frequency (beats min⁻¹) and test type (swim tunnel or free swimming). The interaction between mean fin beat frequency and test type was also included. Pair ID and individual ID were included as random effects. Graphs were plotted in the ggplot2 package (Wickham, 2016). A total of 16 fish were analysed for U_opt and individual U_pref. Of these, 14 were included in paired trials to assess U_pair.

U_opt values ranged from 1.43 to 3.59 BL s⁻¹, with a mean value of 2.25±0.46 BL s⁻¹ (Fig. 5). Individual U_pref ranged from 0.25 to 2.48 BL s⁻¹, with a mean value of 1.25±0.68 BL s⁻¹ (Fig. 5). U_pair ranged from 0.20 to 3.64 BL s⁻¹, with a mean value of 1.14±0.85 BL s⁻¹. U_crit ranged from 3.05 to 4.08 BL s⁻¹, with a mean value of 3.72±0.34 BL s⁻¹. U_opt occurred at a mean value of 61% of U_crit, whereas U_pref occurred at 34% of U_crit on average.

There was a significant difference between U_opt and U_pref (mean difference=0.997 BL s⁻¹, t=4.282, P<0.001), with U_pref being significantly lower, meaning that fish chose to swim spontaneously at speeds significantly slower than their energetic optimum (Fig. 5). U_opt was not significantly correlated with U_pref, indicating that individuals with a higher U_opt did not necessarily also have a relatively high U_pref. However, fin beat frequency was positively correlated with individual U_pref (t=4.867, P<0.001, r²=0.31) (Table 1, Fig. 6A). In the model constructed to compare the relationships between fin beat frequency and swimming speed in the swim tunnel and free-swimming tank, fin beat frequency was found to be significantly positively correlated with speed (t=17.65, d.f.=226.6, P<0.001), but this relationship did not differ between the swim tunnel and the free-swimming tank (t=1.223, d.f.=233.2, P=0.223) (Table 2, Fig. 6B).

The one-way t-test comparison between mean U_pair and U_pref of faster fish revealed U_pair to be significantly lower than the U_pref of the faster individual in the pair (mean % difference=−117.7, t=−3.468, P=0.013) (Fig. 7). In contrast, U_pref of the slowest fish in the pair was not significantly different from mean U_pair (mean % difference=−17.7, t=−0.521, P=0.621). There were no significant correlations between U_pair and mean fin beat frequency, fish position relative to conspecific or mean NND (Table 3).

Executing a movement strategy is a continuous, central component of behaviour for all mobile animals. Although energy requirements significantly influence an individual's movement strategy, other factors such as their abiotic and social environment, stable behavioural traits or ‘personality’, and species-level feeding strategies may also impact how an animal chooses to move within its environment. This study provides evidence that for the pile perch, P. vacca, optimising energy usage per distance travelled may not be the most important factor when choosing a swimming speed, whether alone or in a social context. We predicted that U_opt would be correlated with U_pref. However, U_pref in individual trials was significantly lower than U_opt and the two speeds were not correlated, meaning that fish with higher U_opt values did not necessarily have high U_pref values. Fish did appear to exert increased effort when swimming at higher speeds in both free-swimming and swim tunnel trials, as fin beat frequency was positively correlated with swim speed, a relationship that did not vary significantly depending on the swimming environment. We also found that an individual's social environment may influence preferred swim speed; the slower individual within a pair appears to inform U_pair, with faster fish slowing significantly to school with a partner. Rather than swimming at intermediate speeds for both members of the pair, we found no significant difference between mean U_pair and U_pref for the slower fish, signifying that faster individuals appear to sacrifice their own preferred swimming speed to remain in a social setting.

Although swimming at U_opt maximises energy efficiency, P. vacca appear to choose swimming speeds below their U_opt. U_pref values were significantly lower than U_opt, and U_pref was not correlated with U_opt at the individual level. Swimming speed was positively correlated with fin beat frequency in both individual free-swimming trials and in the swim tunnel, indicating that fish were increasing effort when swimming at greater speeds. The lack of a relationship between U_pref and U_opt indicates that other factors within the environment, in addition to energetic efficiency, likely exert some influence over U_pref in this species. The appropriately named P. vacca frequents complex habitats around coastlines and artificial structures such as piers (Alevizon, 1975; Allen and Pondella, 2006). Behavioural tasks associated with spatially complex and densely populated habitats, such as foraging and avoiding predators, require continuous monitoring of the environment for potential dangers and resources. These behaviours may then be prioritised over swimming at a speed that minimises cost of transport, particularly if that speed is impractically high for an individual's environment (Han et al., 2017; Killen et al., 2007). In migratory and pelagic species, U_pref appears to track U_opt closely (Tudorache et al., 2011; Weihs et al., 1981). This represents an advantageous strategy when selecting a sustained swim speed for covering long distances, and so is an intuitive finding for migratory or pelagic species. In generalist non-migratory species such as P. vacca, however, it appears be of net benefit to swim at a less energetically efficient speed, possibly so as to optimise sensory perception of the environment (Han et al., 2017; Kramer and McLaughlin, 2001; O'Brien et al., 1989). Fitness might then be promoted via increased chances of finding food or a mate, enhanced capacity to manoeuvre around obstacles, and improved ability to detect dangers, by selecting a U_pref lower than U_opt.

Because fish had ample time to explore the flow gradient tank during acclimation, it can be assumed they were aware of the extent of their environment, and thus that steady swimming into the current accomplished no overall navigational progress. Individuals might then have chosen slower U_pref values in this assay than they would have in a natural setting where navigational progress is achievable. However, the same limitation applies to previous U_pref measurements using S. fontinalis with which we principally compare our findings (Tudorache et al., 2011). Despite this potential source of bias, the difference in the U_opt–U_pref relationship between migratory and non-migratory species remains clear. In addition, our protocol for determining U_opt did not account for the accumulation of anaerobic metabolites at sub-critical speeds (Di Santo et al., 2017), though the cost of swimming at low speeds has been found to be negligible in labriform fishes (Svedsen et al., 2010).

Alternative energetic explanations may also be helpful for understanding how preferred swim speed is selected by non-migratory fishes. Although we did not quantify aerobic scope in P. vacca, our results indicate that U_opt represents a much greater proportion of U_crit than U_pref does. The scope for an increase in activity when moving at U_opt is therefore significantly reduced compared with when moving at U_pref, which would represent a survival cost when attempting to escape a predator (Wood, 1991). The benefits of reserving a large portion of the aerobic scope may therefore impose limitations on swimming strategies in some generalist species. The lumpsucker, Cyclopterus lumpus, a benthic species occupying similar habitats to P. vacca, does not swim while foraging and prefers to remain motionless if given the opportunity and provided a minimum density of prey is available (Killen et al., 2007). Killen et al. (2007) speculate that maintaining low swimming speeds may allow fish to maximise the proportion of their aerobic scope that is available at a given time, particularly in a species whose aerobic scope is limited.

Organismal energetics and abiotic environmental components are likely not the only drivers of U_pref in P. vacca, as this species is highly social (Munsch et al., 2016). Fitness benefits derived from social behaviour, including improved chances of avoiding predators, finding food and finding mates (Krause and Ruxton, 2010; Pitcher et al., 1982; Ward and Webster, 2016; Wright et al., 2006), may also influence an individual's movement decisions. The preferred swim speed of faster fish when swimming with a conspecific partner was significantly reduced, resulting in a mean U_pair similar to the U_pref of the slower fish. Slower fish, however, did not appear to significantly adjust their swimming speed. This mismatch in swim speed adjustment is consistent with the hypothesis that U_pref is driven by the need to reserve a portion of an individual's aerobic scope. If swimming at speeds closer to U_opt, i.e. speeds faster than an individual's U_pref, is avoided because of the associated reduction in available aerobic scope, as suggested by Killen et al. (2007), the fitness cost of adjusting swim speed will disproportionately fall on the slower fish as it accelerates past its U_pref. In contrast, reducing swim speed may result in negligible costs to faster individuals whose aerobic scope would be increased under these conditions, while both fish would also benefit from group swimming. Future studies should test whether this mechanism is responsible for the observed mismatch in swim speed adjustments during schooling using direct measurements of aerobic scope.

Swimming in a group is thought to improve energy efficiency compared with swimming alone (Marras et al., 2015), though we did not find any evidence of fish choosing their swim speed based on the relative position of their partner, as might be expected if a particular configuration confers such an energetic advantage. Studies of performance in larger groups as well as in arenas of different shapes are necessary to clarify this dynamic with greater certainty, as vortices between individuals and effects of arena shape also play a role in flow dynamics and, therefore, energy use (Johansen et al., 2010; Marras et al., 2015; Zheng et al., 2022). The lack of correlation between fin beat frequency and U_pair indicates that swimming effort is altered when in groups, thus altering the correlation between individual effort and resultant speed. It is possible that instead of moving in the tank and therefore altering swimming speed according to relative position, fish may adjust fin beat frequency and therefore swimming effort when swimming in a position that confers greater or lesser energy efficiency. Studies of larger groups may reveal clearer patterns in swim speed relative to group positioning and required swimming effort, as more individuals will create more significant vorticity effects. Further, studies incorporating a larger sample size and number of species would also be helpful in generalising these results to the population level for P. vacca and to other marine species.

Quantifying the relationships between fish physiology and the physical and social environment stands to benefit both behavioural research endeavours, where fish are common model organisms, and burgeoning aquaculture enterprises, upon which a significant portion of humans may soon depend (Tudorache et al., 2011). Optimising the conditions, including environmental complexity, conspecific social landscape, and water flow speed and therefore swim speed, of fish kept in captivity is of key interest to both scientists and aquaculturists, as this maximises the output efficiency of such enterprises while also improving welfare standards, allowing for more efficient and humane food production (Tudorache et al., 2011). For migratory pelagic species such as S. fontinalis, energy conservation during migrations with limited feeding opportunities is likely a primary driver of preferred swim speed. This pattern may be generalised to other migratory species, but our study adds to the evidence that it does not hold for non-migratory, social species such as P. vacca (Han et al., 2017). As species held for both research and aquaculture purposes display varied life history strategies and social behaviours, our findings emphasise the need for further study of diverse species when determining optimal living conditions. Additionally, studies evaluating relationships between physiological and environmental variables are also becoming increasingly relevant in the context of environmental change and conservation, as many habitats are restructured as a result of human influence. A thorough understanding of these dynamics is required to predict how a range of fish species might be affected as anthropogenic processes continue to alter aquatic ecosystems.

Optimising energy usage while swimming appears to be a central concern for some migratory species; however, other ecological and physiological factors may influence the movement strategy of non-migratory species living in more complex environments. Our study demonstrates that a coastal marine generalist species prioritises slower movement over optimal energy use. This strategy likely confers an advantage when navigating a complex habitat and may enable foraging and mate-searching as these behaviours are likely more efficient when moving at slower speeds. Slower movement may also be important for maximising the aerobic scope available, allowing fish to respond rapidly to threats such as ambush predators. In addition, we found that movement strategy is altered by the presence of a conspecific, implying that swimming in groups may confer significant advantages over swimming at a preferred speed. This dynamic also appears to have an energetic component, with faster fish slowing down to remain with slower conspecifics when in pairs. This may allow both fish to reserve a portion of aerobic scope, while still gaining the benefits of group swimming. Our study offers new insight into the interactions between physiology and the physical and social environment, and provides a basis for future study of swimming dynamics and social behaviour across life histories in fishes.

We are grateful to the University of Washington for the use of the Friday Harbor Laboratories site and resources. We thank the editor and two anonymous reviewers for their helpful comments, which have significantly improved our manuscript.