We re-analysed three different data sets on thermal acclimation of swimming performance (critical sustained swimming speed, U_crit) in mosquitofish (Gambusia holbrooki; Seebacher et al., 2015; Loughland and Seebacher, 2020). In each data set, individual fish were acclimated twice, to a cool (18 or 20°C) and to a warm (28°C) temperature in random order; acclimation temperatures corresponded to spring and summer temperatures at the capture site (Seebacher et al., 2014). All fish were sourced from the same wild population (Manly Dam, Australia 33°78′S; 151°26′E) and were acclimated for 3–4weeks to the different temperatures. After acclimation, U_crit of each fish was measured at different acute test temperatures. Experimental fish were of mixed sex (184 females, 424 males) and were sexually immature at the start of acclimation treatments. Fish were separated from each other during acclimation treatments to ensure that females were not pregnant at the time of U_crit measurements.

In the first data set (data set 1), n=48 wild-caught fish were each acclimated to 20 and 28°C in random order, and after each acclimation treatment, U_crit was measured at 20, 28 and 32°C acute test temperatures in each fish (Seebacher et al., 2015). In the second data set (data set 2), U_crit of 416 double-acclimated (to 18 and 28°C), wild-caught fish was measured only at the acute test temperature that coincided with acclimation temperatures (Loughland and Seebacher, 2020). The third data set (data set 3) from the same publication (Loughland and Seebacher, 2020) was collected from third- or fourth-generation offspring (total n=144 fish) bred in outdoor mesocosms from parents collected at the study site. Each of these fish was acclimated to 20 and 28°C, and U_crit of each fish was measured at 20 and 28°C acute test temperatures after each acclimation treatment.

Double acclimation permitted calculation of an index of acclimation capacity for each fish (Seebacher et al., 2015): Acclimation capacity = 1 − ( P 28 − P 20 ) [ ( P 28 + P 20 ) / 2 ] , where P₂₈ is the U_crit of a fish that is acclimated to 28°C and measured at 28°C acute test temperature, while P₂₀ is the equivalent measure at 20°C (or 18°C in data set 2). The acclimation capacity index indicates relative thermal compensation (i.e. the ability to maintain relatively constant performance across thermal conditions) by contrasting the difference between P₂₀ and P₂₈. Acclimation capacity approaches 1 as P₂₀ approaches P₂₈ and decreases as P₂₀ decreases. If a fish overcompensated for low temperatures and P₂₀>P₂₈, the index will be >1. The index is based on the difference between P₂₀ and P₂₈ and is a dimensionless number that is independent from the absolute values of P₂₀ and P₂₈. More details of experimental procedures are given in the original publications (Seebacher et al., 2015; Loughland and Seebacher, 2020).

Our purpose here was to mimic typical approaches in the literature to assess acclimation of populations to test whether subsampling of populations can reflect true variation and responses of the population. Hence, we randomly subsampled the combined data sets 1 and 3 (n=192 fish, see Supplementary Material for R code) to draw 10 samples of eight replicates each for warm- and cold-acclimated fish. Each of these 10 data sets mimics a fairly typical experiment in the literature, for example as in our study on mosquitofish where we compared acclimation in spring- and summer-caught fish (Seebacher et al., 2014).

We analysed each data set with a permutational analysis [in the R package lmPerm (Wheeler and Torchiano, 2016)] with acclimation temperature and acute test temperature as independent factors. Permutational analyses do not make assumptions about underlying data distributions but use the data per se to infer significant differences (Drummond and Vowler, 2012). Hence, values of p in permutational analyses are not associated with any particular distribution, and there are no test statistics (such as F or t; Ludbrook and Dudley, 1998). The value of p in permutational tests has the practical meaning of denoting the number of randomised permuted data sets for which the treatment effects were as or more extreme than the observed experimental data divided by the total number of permutations.

Our aim here was to model how populations comprised of individuals with different acclimation capacities respond to different environmental conditions. We modelled different phenotypic compositions of populations by selecting subpopulations from the complete data set (data sets 1, 2 and 3 combined; n=608 fish), which were the top 10% of fish with the highest acclimation capacity (high), the bottom 10% with the lowest acclimation capacity (low) and the central 10% (centre); each of these subpopulations was comprised of 61 fish.

From the P₂₀ and P₂₈ values of each fish, we determined the slope of change in U_crit between these temperatures to estimate U_crit at intermediate temperatures assuming that acclimation has taken place. We simulated environmental conditions within the measured range (20–28°C) by either assuming constant conditions of 20 or 28°C or letting temperatures vary between 20 and 28°C. To model variable temperatures, we assumed sinusoidal temperature variation with a mean of 24°C and an amplitude of 4°C. We randomised the phase of the sinusoidal temperature fluctuation 100 times for each fish and recalculated U_crit (from the slope between P₂₀ and P₂₈) each time. We thereby ‘exposed’ each fish to the complete temperature variation. From the simulated data set, we calculated mean U_crit and 95% confidence intervals for each subpopulation (using n=61 for CI calculations to represent the number of fish in the simulation rather than the number of simulated values).

Mean values for each subpopulation may mask the underlying distributions of U_crit, which may be important for ecological responses. For each subpopulation (low, centre, high acclimation capacity), we therefore modelled U_crit distributions in the variable environment (as determined above) as the per cent of U_crit values that fell above a given fraction of maximum speed. The low acclimation capacity subpopulation had the highest maximum U_crit, which we used as maximum U_crit in all simulations. However, rather than using the single highest U_crit value, which is not representative of most fish, we defined the 90th percentile of the low acclimation capacity subpopulation as the maximum U_crit. We then determined the percentage of U_crit values that fell above a given fraction of this maximum U_crit for each subpopulation, which we defined as ‘achievable’ U_crit. See Supplementary Material for R code.

Thermal performance curves varied considerably between individuals (Figure 2). In both cold- and warm-acclimated fish from data set 1 (Figure 2A), U_crit tended to increase from 20 to 28°C and decrease at 32°C. However, this pattern was not consistent among individuals, and frequently U_crit increased between 28 and 32°C. In fish from data sets 1 and 3 (n=192; Figure 2B), U_crit mostly increased between 20 and 28°C at both acclimation temperatures, but there was considerable variation in thermal sensitivity (i.e. Q₁₀ values). Similarly, responses to acclimation differed considerably between individuals (Figure 2C). At both 20 and 28°C test temperatures, acclimation to 28°C led to either increased or decreased performance in different individuals (data from data sets 1 and 3).

Similar to the patterns of thermal sensitivity described above, there was pronounced variation in acclimation capacity within the total population (data sets 1–3, n=608 fish; Figure 3). Fish phenotypes ranged from having the capacity to fully compensate (and even overcompensate) for the 8–10°C temperature difference following acclimation, to having no capacity for acclimation at all so that U_crit changed passively (thermodynamically) with an acute change in temperature. Interestingly, there was a trade-off in capacity for acclimation with performance at warm conditions (P₂₈): P₂₈ decreased significantly (regression: Y=12.15−3.39x; R²=0.22, p<0.0001) as P₂₀ increased (Y=3.97+4.67x; R²=0.54, p<0.0001) with increasing acclimation capacity. Across both acclimation treatments, mean performance of fish increased slightly but significantly with increasing acclimation capacity (Y=8.06+0.68x, R²=0.014, p=0.0032; Figure 3). However, the persistence of both plastic and fixed phenotypes in the population indicates that on average, plasticity is not necessarily advantageous over fixed phenotypes, but also that plasticity does not carry a cost that would select against it.

Increased variation of phenotypes can render the populations as a whole more resilient to change if different phenotypes are advantageous under different environmental conditions (Schindler et al., 2015; Blondel et al., 2021). Plasticity is advantageous by buffering performance under cooler conditions, such as in winter, while fixed phenotypes perform better at high temperatures during summer. This evolutionary bet-hedging can increase population persistence (Schindler et al., 2015). Mechanistically, it is an interesting question of what mediates the observed variation between individuals. Reactive oxygen species (ROS) and heat shock proteins can be induced by acute heat or cold exposure (Liu et al., 2018). Cold acclimation also increased ROS production in mosquitofish (Loughland and Seebacher, 2020) and grass snakes (Bury et al., 2018), and salmon acclimated to 20°C had greater rates of oxidative phosphorylation but reduced ROS production compared to 12°C acclimated fish (Gerber et al., 2020). Mosquitofish with high acclimation capacity also have greater antioxidant capacities (Loughland and Seebacher, 2020). Increased ROS as a result of reduced antioxidant capacity can decrease swimming performance (Ghanizadeh-Kazerouni et al., 2016) and may at least partly explain the observed patterns in our study. Alternatively, the dynamics of signalling pathways, from endocrine (e.g. thyroid hormone) to epigenetic (e.g. histone acetylation and DNA methylation), may differ between individuals and thereby cause individual variation (Simmonds and Seebacher, 2017; Little, 2021; Loughland et al., 2021).

Acclimation did not have a significant main effect in any of the ten random samples (all p>0.3), but test temperature was significant in all samples, and on average, U_crit increased with increasing test temperature (all p<0.003; Figure 4A). There was a significant interaction between test and acclimation temperatures in samples 1 and 7 (p<0.05) and at a one-tailed significance level in sample 3 (p=0.083); the interaction was not significant in any of the other samples (all p>0.24). Knowing the true acclimation capacity of individual fish showed that means mask the variation in acclimation capacity between individuals, and in many samples, individual acclimation capacity ranged from close to no capacity for acclimation (values→0) to perfect compensation for the acclimation temperature difference between treatments (values→1; Figure 4B).

The relatively small sample size simulated here can lead to fundamentally different conclusions about the population as a whole: in most cases, the data showed that there was no acclimation response, but three samples indicated that there was. These results serve as a cautionary note to avoid undersampling of populations and presenting means in the absence of individual values. However, even if larger samples were collected, sample means would mask the underlying variation and obscure the bet-hedging dynamics discussed above. Experimental approaches that compare individuals from different acclimation treatments may not be sufficient to test for costs or trade-offs associated with plasticity, which would require knowledge of within-individual acclimation capacities. Understanding individual variation in acclimation capacity would be necessary to predict how populations can respond to climate variability, where diversity of phenotypes may be important to increase resilience. Sample means can show population trends across time or contexts.

Our simulations showed that in the fluctuating environment, the average U_crit was similar in all three subpopulations, indicating that capacity for acclimation did not affect mean performance under these circumstances (Figure 5). However, performance in the stable cool environment decreased in the centre and low acclimation subpopulations, but it increased in the stable warm environment in those subpopulations. In contrast, U_crit of fish with high acclimation capacity did not vary significantly (95% CI) between either the stable or the fluctuating environments.

The achievable U_crit in a variable environment was defined as the per cent of U_crit (= achievable U_crit) that was greater than a given fraction of the maximal (90th percentile) U_crit. In other words, the achievable U_crit is synonymous with the per cent of time fish could swim faster than a given fraction of maximum in an environment that varied over time (Figure 6). Fish with low acclimation capacity had the highest maximum U_crit (90th percentile=11.7 body lengths s⁻¹), but in a variable environment, their achievable U_crit declined rapidly. This result is not surprising because in fish with low acclimation capacity, U_crit changes thermodynamically in proportion with the acute temperature change. In contrast, fish with central or high acclimation capacity compensated at least partially (centre) for the decline in environmental temperatures. Hence, even though these groups had lower maximum U_crit (90th percentile: 10.3 and 10.4 BL s⁻¹, respectively), their achievable U_crit was higher than in the low acclimation subpopulation below approximately 0.8 of maximal U_crit. The high acclimation capacity subpopulation had the highest achievable U_crit (Figure 6).

These nonlinear distributions of achievable U_crit could present a selective advantage for plastic phenotypes if the ecological outcomes of movement are maximised at a lower than maximal speed. Animals rarely move at their maximum speed, and the functional outcomes of responses such as escaping a predator may be optimised at fractions of maximal speed (Wilson et al., 2015). Success in escaping from a predator may be highest at a submaximal speed because at maximal speeds, precision of movement, information processing and endurance decline, while energetic costs increase (Wilson et al., 2015; Wynn et al., 2015). Conversely, speeds below a given threshold may simply be too slow for the prey to escape (Husak and Fox, 2006). Hypothetically, if it is assumed that a fraction of 0.7 maximal U_crit is necessary for escape from predators (Husak and Fox, 2006), the different distributions of achievable U_crit in our subpopulations will influence the success of escaping. Only 46% of U_crit values in the low acclimation capacity subpopulation fall above 0.7 maximal speed, while 71% of U_crit values were above 0.7 of maximal speed in the high acclimation capacity subpopulation. Hence, fish with high acclimation capacity would be vulnerable around 30% of the time, while those with low acclimation capacity would be vulnerable more than half the time. This hypothetical example demonstrates that the buffering of U_crit for low temperatures may translate to a selective advantage across acclimation conditions.