Energetic cost determines voluntary movement speed only in familiar environments

Locomotor performance is an essential component in the ecology of animals, because it can influence dispersal, foraging and predation, and behavioural interactions (Hillman et al., 2014; Husak et al., 2006; Irschick and Garland, 2001). Maximal locomotor capacity is important for ecology when it limits movement speed to the extent that individuals cannot perform effectively in an ecological context, such as escaping predators or in aggressive interactions (Grigaltchik et al., 2012; Husak and Fox, 2006; Sinclair et al., 2014). Often, however, animals do not reach their maximal locomotor capacity when moving under undisturbed conditions (Irschick and Garland, 2001; Wilson and Husak, 2015). Instead, voluntary movement speed is likely to be lower than maximal speeds during foraging and dispersal, for example (Weihs, 1973; Irschick and Losos, 1998; Husak and Fox, 2006; Humphries et al., 2010; Wilson et al., 2013). It is therefore important to understand the mechanisms that determine voluntary speed, because these will influence movement patterns of individuals across environmental contexts. For example, dispersal rates may differ across environmental gradients (Clobert et al., 2009; Pigot and Tobias, 2015), and differences in dispersal speed may alter population structures (Evans et al., 2012). Ultimately, this could influence population genetics if animals are sorted non-randomly with respect to the voluntary speed at which they move (Hillman et al., 2014).

Locomotor performance is determined by the relationship between muscle power output and the energetic cost to achieve that power output (Curtin and Woledge, 1991; Lichtwark and Wilson, 2005; Woledge et al., 2009). In other words, locomotor performance is determined by energy metabolism and the intrinsic functioning of muscle proteins involved in calcium cycling to mediate contraction and relaxation (Berchtold et al., 2000; Allen et al., 2008; Bruton et al., 2010). Theory predicts that voluntary speed reflects minimisation of the cost of transport (i.e. the energy used for a given distance travelled), which occurs at a locomotor speed that is a fraction of maximal speed (Weihs, 1973; Pettersson and Hedenstrom, 2000; Wickler et al., 2000; Claireaux et al., 2006; Palstra et al., 2010).

Voluntary locomotor speed may therefore be determined by the mechanical efficiency of muscle; that is, by the energy used for a given muscle power output (Barclay et al., 2010). Mechanical power output is defined as force produced multiplied by muscle shortening velocity (Curtin and Woledge, 1996; James et al., 1995). Power output is affected by muscle activation as a result of calcium release from the sarcoplasmic reticulum, and the duration of activation (duty cycle). Increasing the duty cycle increases power output, but it also decreases efficiency (Curtin and Woledge, 1991,, 1996). Maximal efficiency is achieved at a duty cycle and muscle shortening velocity that are a fraction of those producing maximal power output (Curtin and Woledge, 1991,, 1996; Barclay et al., 2010). During maximal locomotor performance, muscle works at a relatively low efficiency in terms of converting chemical energy to mechanical energy (Lichtwark and Wilson, 2005; Woledge et al., 2009), and it may be constrained by maximal ATP supply as a result. In contrast, animals moving at their voluntary speed should optimise mechanical efficiency rather than maximising power output. Nonetheless, there are considerable differences in muscle characteristics between individuals, because mechanical efficiency is optimised at a duty cycle and shortening velocity characteristic of each individual muscle. Individuals with muscle that possesses relatively high maximal shortening velocities should therefore also have greater muscle power output both when performing maximally and at optimal efficiency (Herrel et al., 2007; James et al., 2012; Seebacher et al., 2012; James, 2013). Consequently, it would be expected that voluntary speed is positively associated with maximal speed within individuals.

The processes outlined above are sensitive to changes in temperature, and there can be substantial differences in the thermal sensitivity of the various steps involved in muscle contraction and relaxation, and their relationship to ATP use (James, 2013; Seebacher et al., 2014). Maximum locomotor performance therefore is strongly dependent on temperature, and it is likely that voluntary speed and its relationship to maximal performance change in different thermal environments.

Our aim was to determine the relationship between voluntary movement speed, maximal locomotor capacity and metabolism across a temperature range. We analysed differences between individuals to test the following hypotheses (Fig. 1). Hypothesis 1: maximal sustained locomotor performance (U_crit) is constrained by absolute ATP supply; that is, it increases with increasing metabolic scope (Fig. 1A). Hypothesis 2: voluntary speed occurs at a fraction of maximal sustained speed; that is, voluntary speed is positively related to U_crit, because muscle efficiency is optimised at a given fraction of the maximum for each individual (Fig. 1B). Hypothesis 3.1: voluntary speed decreases with an increase in the energetic cost of locomotion within individuals, because voluntary speed is determined by minimisation of metabolic cost (Fig. 1C). Hypothesis 3.2: if hypothesis 3.1 is rejected, an alternative is that animals regulate voluntary speed per se. This may have evolved in group-living animals to stay with a shoal (Conradt and Roper, 2000), and it may reflect constraints on movement imposed by information processing (Stewart et al., 2012; Baumann and Mattingley, 2014). In this case, the energy used to travel at voluntary speed is positively related to the cost of locomotion, so that individuals with a greater cost of locomotion also invest more when moving voluntarily (Fig. 1D).

We tested these hypotheses at three different temperatures in zebrafish (Danio rerio). Additionally, the novelty of an environment decreases the longer animals are exposed to it, and behaviour changes concomitantly (Wong et al., 2010; Hills et al., 2015). Hence, we analysed movement data at two different time points (1st and 10th minute) after fish were released into an open-field arena.

All experiments were performed with the approval of the University of Sydney Animal Ethics Committee (approval number 587). We chose zebrafish, Danio rerio (Hamilton 1822), because the species is an established model for studies on spatial behaviour (Stewart et al., 2013). Zebrafish were obtained from a commercial supplier [Livefish, Bundaberg, Australia; mean (±s.e.m.) mass=0.51±0.02 g; mean standard length=31.27±0.07 mm]. Fish were kept for at least 1 week in holding tanks (0.60×0.46×0.25 m at a density of approximately one fish per litre) before experiments started. Water temperature was 24°C except for experiments at different acute test temperatures, and the light cycle was 12 h light:12 h dark. Fish were fed twice a day with commercial flake food (Wardley's Tropical Fish Flakes, The Hartz Mountain Corporation, Secaucus, NJ, USA), but were not fed for 24 h before experiments.

In each individual fish we measured movement in an open-field arena, followed by measurements of sustained swimming speed. We then rested the fish for 20 h before measuring oxygen consumption at rest and while swimming at different speeds. We measured these responses at 18, 24 and 30°C using different fish for each test temperature (N=15 fish per test temperature). To identify individual fish in between experimental measurements, we kept individuals in cylindrical plastic containers (1000 ml) with longitudinal slits in their sides, which prevented fish from leaving the container but allowed water flow through the container and visual and olfactory contact between fish. In each holding tank (0.60×0.46×0.25 m), 5–6 fish were kept in their individual containers during experiments.

Open-field arenas were constructed from white plastic (0.6×0.4 m, with a water depth of 0.03 m). Arenas were located in a constant temperature room, which was adjusted to achieve the desired water test temperatures; for the 30°C test temperature, we additionally used aquarium heaters, which were removed before fish were introduced into the arenas. Individual fish were tested in each trial, and at the start of each trial fish were introduced into an opaque plastic cylinder standing on its end in one corner of the arena. Fish were kept in the cylinders for 30 min, after which cylinders were raised remotely without disturbing the fish. The movement of each fish was then filmed for 10 min (at 30 frames s⁻¹; HD910, Logitech, China). We analysed the 1st and the 10th minutes of each trial to capture movement in novel and familiar environments, respectively. Videos were analysed in Tracker software (Open Source Physics, www.opensourcephysics.org) to determine the average voluntary swimming speed of each fish.

We tested the repeatability of voluntary swimming speed in a separate, preliminary experiment. We conducted open-field tests as described above (in N=26 fish), and repeated the trials 7 days later, keeping track of individual fish identity between trials as described above. We ranked fish for each trial, and compared ranks between trials by Spearman correlations to determine consistency in voluntary speed over the 7 day period.

Sustained swimming performance was measured as critical sustained swimming speed (U_crit; Hammer, 1995) in a Blazka-type swimming flume according to published protocols (Hammill et al., 2004; Seebacher et al., 2015a). U_crit was measured in a cylindrical clear Perspex flume (150 mm length and 34 mm diameter) that was fitted tightly over the intake end of a submersible pump (12 V DC, iL500, Rule, Herts, UK), which drew water through the flume. A plastic grid separated the flume from the pump, and bundles of hollow straws at the inlet end of the flume helped maintain laminar flow. The flume and pump were submerged in a plastic tank (0.65×0.42×0.28 m). Flow speed was adjusted by changing the voltage input to the pump with a variable DC power source (NP9615; Manson Engineering Industrial, Hong Kong); flow was measured in real time individually for each flume by flow meters (DigiFlow 6710M, Savant Electronics, Taichung, Taiwan) connected to the outlet of each pump. Fish were allowed to swim at an initial flow rate of 0.06 m s⁻¹ [approximately 2 body lengths (BL) s⁻¹] for 10 min and the flow speed was then increased by 0.06 m s⁻¹ every 10 min until fish were exhausted and stopped swimming. The first time a fish stopped swimming and fell back onto the grid, we stopped the flow for 5–10 s, after which it was increased again to the previous setting. The second time a fish stopped swimming, the trial was terminated and the time until exhaustion was recorded. U_crit was determined as U_crit=U_f+T_f/T_i×U_i, where U_f is the highest speed maintained for an entire interval (T_i=10 min), T_f is the time until exhaustion at the final speed interval and U_i is the speed increment. U_crit is reported as BL s⁻¹.

Resting oxygen consumption rate (Ṁ_O2) was measured in cylindrical, clear plastic respirometers (internal volume 27 ml) submerged in plastic tanks (0.3×0.25×0.25 m). Respirometers were sealed with plastic stoppers, and were connected to a peristaltic pump (iPump i150, Loligo Systems, Tjele, Denmark) that provided sufficient flow to replace the oxygen consumed by the fish when switched on; the pump could be turned on and off remotely without disturbing the fish. We affixed oxygen sensor spots (Pst3, Loligo Systems) to the inside of respirometers at their midpoint. A fibre optic cable attached to the outside of the respirometers (with a Velcro strap, Loligo Systems) and connected to an oxygen meter (Witrox 4, Loligo Systems) monitored oxygen concentration inside the respirometers. At each sampling time, the pump was turned off and oxygen concentration was measured until there was a steady decline in concentration (within ∼5–8 min). Oxygen consumption was calculated from the slope of the linear decrease in oxygen concentration (Sinclair et al., 2006). We also conducted measurements of respirometers without fish to control for oxygen consumption or production by microorganisms; in the event, the respirometer set-up was kept dry when not in use and was regularly cleaned with a weak bleach solution so that there was no confounding effect of microorganisms.

Before measuring resting Ṁ_O2, we conducted a preliminary study (with 12 fish not used elsewhere in the study) to determine how long fish needed to be rested after handling before oxygen consumption stabilised. For the pilot experiment, fish were introduced into the respirometers, and the respirometers were sealed without turning on the pump to measure Ṁ_O2 immediately following handling (time=0). We then turned on the peristaltic pump between measurements without disturbing the fish, and measured Ṁ_O2 every 30 min for 120 min. Ṁ_O2 stabilised after 90–120 min (Fig. 2), and we therefore rested fish for 120 min before measuring resting Ṁ_O2 in all subsequent trials.

Ṁ_O2 while fish were swimming was measured in a zebrafish swimming flume respirometer (170 ml, Loligo Systems) according to the manufacturer's instructions. Similar to resting Ṁ_O2, these values can be confounded by handling stress at low speed, and we conducted a preliminary experiment as above to determine that fish needed a minimum of 30 min to relax while swimming at the lowest speed. We initially measured Ṁ_O2 of individual fish swimming at 0.05 m s⁻¹ and then in 0.1 m s⁻¹ increments until fish could not hold their position in the flume. At each measuring point, the swimming flume was sealed and oxygen decrease was measured until there was a steady decline in oxygen concentration (after ∼10 min). The flume was then flushed with fully oxygenated water, and the procedure was repeated at the next flow speed. As above, we conducted trials without fish, and the flume respirometer was drained and cleaned every day to avoid confounding effects from microorganisms.

We used non-linear regression (in GraphPad Prism 5.0 Software) to fit logarithmic curves [y=a+b ln(x)] to Ṁ_O2 at different swimming speeds of individual fish. Note that we used resting Ṁ_O2, which we determined independently, as the value for zero flow. The curves allowed us to determine Ṁ_O2 at the voluntary swimming speed of each fish. We also used Ṁ_O2 data to calculate the metabolic cost of transport (CoT, in µmol kg⁻¹ m⁻¹) at different swimming speeds by dividing Ṁ_O2 (in µmol kg⁻¹ s⁻¹) by swimming speed (in m s⁻¹; Claireaux et al., 2006). As above, we used non-linear regression to fit power curves (y=ax^b) to the CoT data of each fish. We integrated the power curves to estimate the aerobic metabolic cost of movement across all speeds for each fish (‘integrated cost’ in W kg⁻¹). This metric is useful in determining differences between individuals in the investment into locomotion, and their potential consequences for voluntary movement.

We analysed the effect of metabolic scope on U_crit (hypothesis 1) by permutational analysis of co-variance (using the package lmPerm in R v1.1-2) with test temperature (18, 24 and 30°C) as independent factor and metabolic scope as covariate. Similarly, we analysed voluntary speed and Ṁ_O2 at voluntary speed with permutational analyses of co-variance with time (1st or 10th minute in the arena) as random factor, test temperature as fixed factor, and different predictor variables [U_crit (hypothesis 2) and integrated aerobic cost of locomotion (hypothesis 3)] as covariates in separate analyses. Additionally, we used standard length as covariate for all measures of swimming speed, and body mass as covariate for measures of metabolic rate. When ‘time’ was significant, we repeated analyses for each time point separately. We followed 2-way interactions between fixed factors and covariates by conducting linear regression as post hoc analyses. All analyses were performed in R.

Voluntary swimming speed was a repeatable trait within individuals, and there was a significant correlation in ranks of voluntary swimming speed of fish between the 1st and 2nd weeks (r=0.48, P<0.015; Fig. 2A).

Ṁ_O2 decreased rapidly within the first 30 min of handling, and then plateaued at a low value after 60 min (Fig. 2B).

Metabolic rate increased logarithmically with increasing swimming speed [18°C: y=0.31+0.022ln(x), R²=0.84; 24°C: y=0.53+0.041ln(x), R²=0.90; 30°C: y=0.79+0.062ln(x), R²=0.94; Fig. 3A]. Consequently, CoT decreased with increasing swimming speed (18°C: y=0.0049x^−0.97, R²=0.99; 24°C: y=0.0086x^−0.91, R²=0.99; 30°C: y=0.014x^−0.87, R²=0.90; Fig. 3B).

We reject the hypothesis that U_crit is constrained by maximal ATP supply. U_crit increased significantly with test temperature (P<0.03; means±s.e.m., 18°C: 0.17±0.014 m s⁻¹; 24°C: 0.25±0.016 m s⁻¹; 30°C: 0.28±0.025 m s⁻¹), but there was no effect of metabolic scope on U_crit within test temperatures (both main effect and metabolic scope×test temperature interaction P>0.25; Fig. 4).

The relationship between voluntary speed and U_crit varied depending on temperature and time in the arena. Despite considerable variation between individuals, voluntary speed increased significantly with U_crit (P<0.05; Fig. 5). Additionally, there was an interaction between test temperature and the time at which behaviour was recorded in the arena that influenced voluntary speed (P<0.05). No other interactions were significant (P>0.15). We resolved the interaction between time and test temperature by repeating the analysis for each time point separately.

During the 1st minute after release into the arena, a significant interaction (P<0.03) between test temperature and U_crit influenced voluntary swimming speed, and the relationship between these two variables was not significant at 18 and 24°C (linear regression both F_1,13<1.2, P>0.25), but voluntary speed increased significantly with U_crit at 30°C (F_1,13=10.24, P<0.01; Fig. 5A). However, there was no significant relationship between voluntary speed and U_crit during the 10th minute when neither main effects of the factors nor their interactions were significant (all P>0.6; Fig. 5B).

Hypothesis 3 predicted that voluntary speed decreases with an increase in integrated cost of locomotion. We reject this hypothesis except for fish swimming at 24°C in the 10th minute. There was an interaction between test temperature and the integrated cost of locomotion in determining voluntary swimming speed (P<0.03; Fig. 6), as well as an interaction between time and test temperature (P<0.05). We resolved the interaction between time and test temperature as above by repeating analyses within time periods. During the 1st minute, there was a significant effect of temperature (P<0.01) on voluntary swimming speed, but integrated cost of locomotion had no effect (both main effect and temperature×integrated cost interaction P>0.18; Fig. 6A). However, during the 10th minute, there was a significant interaction between test temperature and integrated cost of locomotion (P<0.04); voluntary speed decreased with increasing cost at 24°C (linear regression F_1,13=5.11, P<0.05), but not at the other temperatures (both F_1,13<3.8, P>0.07; Fig. 6B).

Hypothesis 3.2 predicted that rather than minimising the metabolic cost of locomotion by altering voluntary speed, fish regulate speed regardless of the metabolic cost. Hence, we predicted that metabolic rate at the voluntary speed of each fish is positively related to the integrated cost of locomotion of each fish. We accept this hypothesis for most cases, but there were significant interactions between integrated cost of locomotion and test temperature (P<0.005) and time (P<0.02; Fig. 6C,D). Resolving the analysis for each time period showed that there were significant interactions between test temperature and integrated cost during the 1st (P<0.002; Fig. 6C) and the 10th (P<0.02; Fig. 6D) minutes. In all cases except for 24°C during the 10th minute (linear regression F_1,13=0.13, P=0.72), metabolic rate at voluntary speed increased with increasing integrated cost (1st minute: 18°C, y=−0.0073+0.059x, F_1,13=42.10, P<0.0001, R²=0.78; 24°C, y=−0.15+0.0074x, F_1,13=7.08, P<0.02, R²=0.35; 30°C, y=−0.32+0.056x, F_1,13=58.85, P<0.0001, R²=0.82; 10th minute: 18°C, y=−0.15+0.032x, F_1,13=20.93, P<0.001, R²=0.63; 30°C, y=−0.27+0.0076x, F_1,13=6.81, P<0.03, R²=0.34). The interactions between test temperature and integrated cost reflect the different slopes of that increase and the lack of a relationship at 24°C during the 10th minute.

We have shown that fish do not adjust their voluntary speed to minimise aerobic metabolic cost, except when moving in a familiar environment at their long-term acclimation temperature; under these conditions, energy expended during voluntary movement does not increase with increasing integrated cost of locomotion as it does under all other conditions. The implications of these data are that the energetic costs of exploring and dispersal into novel environments are greater than those of movement within familiar home ranges. Search behaviour is common to most animals, and there is a trade-off between exploiting known environments and exploring novel environments (Hills et al., 2015). Our data show that in addition to a cognitive dimension to search behaviour (Hills et al., 2015), there is also a shift in the way animals move. In known environments, animals adjust movement speeds to minimise metabolic cost, but the motivation to explore unfamiliar environments seems to override energetic optimisation of movement speeds. The risks of exploring new environments in terms of failing to find resources and unknown predation pressures, for example, are therefore exacerbated by increased energy expenditure.

The level of familiarity with the environment elicits different cognitive processes, and a trade-off between exploration of novel environments and exploitation of known environments influences spatial behaviour (Hills et al., 2015; Wong et al., 2010). In an unfamiliar environment, animals move to collect information, which is converted into a spatial map that subsequently directs movement (Baumann and Mattingley, 2014). Hence, movement patterns may be determined by neural physiology and information processing (Baumann and Mattingley, 2014) as well as by muscle physiology. Nonetheless, voluntary speed in unfamiliar environments was still lower than U_crit, which begs the question of what determines voluntary speed under these circumstances? Voluntary speed may be influenced by the dynamics of calcium release and re-sequestration into the sarcoplasmic reticulum. Ca²⁺ cycling that mediates contraction and relaxation of muscle is associated with both maximal locomotor performance and voluntary speed. For example, inhibition of excitation–contraction coupling in zebrafish muscle also reduced voluntary speed even though it was not constrained by maximal capacities (Seebacher et al., 2015b). Animals moving voluntarily over periods longer than a short sprint should move at a speed where Ca²⁺ handling by SERCA (sarco/endoplasmic reticulum Ca²⁺-ATPase) and cross-bridge detachment do not compromise subsequent contraction (Allen et al., 2008; Barclay, 2012). Fatigue induced by slower relaxation rates would mean severe reductions in locomotor ability and therefore increased vulnerability by slowing escape from detrimental environmental conditions. Interestingly, fish responded to both spatial and thermal familiarity. It is possible that novel thermal environments elicit an exploratory response in a similar way to novel spatial environments. Hence, similar to exploration in novel spatial environments, fish may explore thermal opportunities, which overrides other physiological factors.

Other environmental factors may also induce differences in movement dynamics similar to the differences between unfamiliar and familiar environments we observed. For example, perceived predation threat such as chemical cues from predators or from injured conspecifics may alter the energetics of voluntary movement, even if animals are not directly escaping from predators (Irschick and Garland, 2001; Wilson et al., 2015). In group-living animals such as zebrafish, the benefits of staying with the group (Krause and Ruxton, 2002) may override energetic optimisation of movement. We have shown that the energetic cost of movement differs considerably between individuals, which raises the possibility that the energetic cost of staying within a shoal differs between individuals. The interaction between ecological context and energetics of movement is important to pursue in future studies because it could influence the fitness of individuals under natural conditions.

We hypothesised that voluntary speed is associated with U_crit because mechanical efficiency is optimised at a fraction of maximal power output for each individual. We accept this hypothesis only for fish swimming in unfamiliar environments at high temperatures. Voluntary movement speed can be associated with maximal speed (O'Steen and Bennett, 2003; Plaut, 2001). However, this relationship can change across temperature, because of the different physiological mechanisms underlying the two forms of swimming (O'Steen and Bennett, 2003). Limited energy (ATP) supply may constrain muscle contraction and relaxation, and it may thereby constrain maximal locomotor performance. However, this is more likely to occur in fast muscles rather than in slow muscles where ATP concentrations are well buffered by the phosphocreatine system (Allen et al., 2008). Hence, limited ATP supply is less likely to affect sustained locomotion unless there is a substantial component of fast muscle that contributes to sustained locomotion. Even though U_crit in our zebrafish was not constrained by metabolic scope, glycolytically produced ATP can contribute to powering swimming, particularly at high U_crit when animals recruit fast muscle fibres (Hammer, 1995). Hence, the slight decreases in Ṁ_O2 at high swimming speeds we observed in zebrafish may reflect a shift from mitochondrial to glycolytic ATP for a brief period before fish stopped swimming. Our measures of energetic cost include aerobic costs only. Unlike in other species of fish (Claireaux et al., 2006; Tudorache et al., 2008), the increase in metabolic rate with swimming speed in our zebrafish was logarithmic rather than exponential. Earlier studies on zebrafish swimming showed similar results in that the increase in metabolic rate with increasing swimming speed was relatively flat (Plaut and Gordon, 1994). These results indicate that in zebrafish ATP is supplied by glycolytic pathways at higher speeds, which is concomitant with the relatively frequent burst swimming episodes at high speed. Hence, our estimates of cost could be an underestimate of the total metabolic cost.

A striking aspect of our data was the considerable variation between individuals in voluntary speed, U_crit, and integrated cost of locomotion. Locomotor performance is closely related to fitness and it could therefore be expected that purifying selection would decrease variation (Le Galliard et al., 2004). However, when corrected for body length, there was more than an order of magnitude difference in voluntary speed and more than a threefold difference in integrated cost of locomotion between individuals within test temperatures. This variation may be due to trade-offs between muscle phenotypes. Fish may be constrained in optimising both sprint and endurance at the same time, but both forms of locomotion may be advantageous at different times (Wilson and James, 2004). Hence, selection does not favour a particular locomotor phenotype. The consequences are that individuals within populations may show different tendencies for movement-related behaviour such as dispersal. In group-living animals, it may also mean that groups assort by locomotor phenotypes, which would facilitate cohesion of the group.