Links between mitochondrial function, whole-animal metabolic rate, telomere dynamics and swimming performance in minnows

TL;DR

Minnows with higher baseline metabolism and better cellular energy production swim faster, despite no link to mitochondrial efficiency or muscle fiber type.

Contribution

This study links whole-animal metabolic rate and mitochondrial OXPHOS capacity to swimming performance in minnows.

Findings

Critical swimming speed (Ucrit) is positively related to standard metabolic rate (SMR) and OXPHOS capacity.

Mitochondrial ROS production is linked to OxCE but not to telomere length.

Swim performance is not influenced by mitochondrial efficiency or muscle fiber composition.

Abstract

The majority of fish swim by aerobic muscular force, and so there has been considerable interest in the metabolic basis for swimming. Most of this work has measured whole-body oxygen consumption as a metabolic proxy, without any quantification of the actual energy that is produced at the cellular level. In this study, we explored links between organism level metabolic rate [both standard (SMR) and maximal (MMR)], mitochondrial function [the rates of oxygen consumption associated with oxidative phosphorylation (OXPHOS) and offsetting proton leak (i.e. OXPHOS coupling efficiency; OxCE)] and swim performance (Ucrit) using the European minnow (Phoxinus phoxinus). We also measured the relative proportion of aerobic (slow-twitch) and anaerobic (fast-twitch) muscle fibres within the muscle tissue. Lastly, we measured mitochondrial reactive oxygen species (ROS) production rates and the telomere lengths of the minnows (because rates of telomere shortening are known to be influenced by ROS). We found that the critical swimming speed of a fish was unrelated to measures of mitochondrial efficiency (OxCE) or MMR, or to the proportion of aerobic fibres within the muscle mass. However, Ucrit was positively related to individual SMR and OXPHOS capacity, indicating that better swimmers are supported by a higher baseline metabolism and a greater cellular capacity for producing ATP. There was also a significant link between OxCE and rates of mitochondrial ROS production, but this was unrelated to telomere length. This study exemplifies how cellular energy production can influence overall performance.

Summary: Minnows swim better when their baseline metabolism is higher and their cells are better at producing energy.

Figures

Experimental overview. Fish were captured on 19 February 2020, tagged on 9 March 2020, and the Ucrit trials began on 18 August 2020 (the large time gap between tagging and swimming trials being due to Covid-19 lockdown restrictions). See Materials and Methods for further details of each measurement.

Relationships between individual body mass (g) and metabolic variables. (A) Standard metabolic rate (SMR, correlation coefficient=0.78, n=54), (B) maximal metabolic rate (MMR in mg O2 h−1, correlation coefficient=0.56, n=54) and (C) aerobic scope (AS in mg O2 h−1, correlation coefficient=0.53, n=54). Both variables were log10-transformed for the MMR and AS regressions as this improved the normality of the residuals from these analyses. Residuals from these three regression equations were then used to create three new variables: (D) residual standard metabolic rate (rSMR), (E) residual maximal metabolic rate (log rMMR) and (F) residual aerobic scope (log rAS), respectively. See Materials and Methods for statistical analyses.

Relationship between an individual's Ucrit score and its body mass, rSMR and OXPHOS/CS. (A) Body mass (correlation coefficient=−0.55, n=54), (B) residual standard metabolic rate (rSMR, correlation coefficient=0.30, n=54) and (C) muscle mitochondrial capacity for oxidative phosphorylation respiration (OXPHOS/CS, correlation coefficient=0.31, n=54). Data are plotted as individual fish with regression line. Ucrit scores in B and C have been plotted as partial residuals when body mass is held constant. See Materials and Methods for statistical analyses.

Relationship between the OXPHOS coupling efficiency (OxCE) of muscle mitochondria and the production rate of reactive oxygen species, measured at the time of OXPHOS (ROSOXPHOS, correlation coefficient=0.29, n=54). Data are plotted as individual fish with regression line. See Materials and Methods for statistical analyses.

Tables

Table 1.. Summary of the response and predictor variables used in the five statistical analyses

Variable name	Variable description
U crit	A measure of an individual's critical swimming speed, i.e. the speed at which the fish fatigues and can no longer hold position. Measured in body lengths per second (BL s−1). Continuous variable.
Body mass	The mass of each individual measured to the nearest 0.01 g. Measured the day after the metabolic rate measurements. Continuous variable.
CS	The rate of citrate synthase activity. Used as a quantitative proxy of mitochondrial volume. Measured as μmol g−1 min−1.
rSMR	Standard metabolic rate (SMR) is measured when the fish is in a post-absorbative inactive state. Measured as mg O2 h−1. Residual standard metabolic rate (rSMR) is the residual from a regression between log10-transformed SMR and log10-transformed body mass (g). Continuous variable.
rMMR	Maximal metabolic rate (MMR) is measured after exhaustive anaerobic exercise. Measured as mg O2 h−1. Residual maximal metabolic rate (rMMR) is the residual from a regression between log10-tranformed MMR and log10-transformed body mass (g). Continuous variable.
OXPHOS	The oxygen consumption rate of the mitochondria during oxidative phosphorylation (OXPHOS). Calculated per mg of wet mass shredded tissue preparation measured as pmol s−1 mg−1. This oxygen consumption rate was measured in the presence of all substrates (pyruvate, malate, glutamate and succinate) and saturating levels of ADP.
OxCE	OXPHOS coupling efficiency is a calculated ratio that estimates the proportion of oxygen being used for ATP production rather than to offset proton leak. Standardised to range between 0 and 1. Unitless. Continuous variable.
ROSOXPHOS	The rate of reactive oxygen species (ROS) production within the mitochondria, measured during oxidative phosphorylation, i.e. in the presence of all substrates (pyruvate, malate, glutamate and succinate) and saturating levels of ADP. Measured as pmol s−1 mg−1. Continuous variable.
OXPHOS/CS	The oxygen consumption rate during oxidative phosphorylation per mg of wet mass shredded tissue preparation, divided by CS activity in the same tissue. This oxygen consumption rate was measured in the presence of all substrates (pyruvate, malate, glutamate and succinate) and saturating levels of ADP. Measured as O2 s−1 μmol−1 min−1. Continuous predictor variable.
Telomere length	An individual's mean telomere length as determined by telomere restriction fragment (TRF) analysis. Measured as bp. Continuous variable.
Slow muscle proportion	The proportion of muscle consisting of red (aerobic) fibres, determined by histology. Measured as a percentage. Continuous variable.

Table 2.. A summary of all the measured/calculated traits

	Units	N	Min.	Max.	Mean	s.d.
U crit	BL s−1	54	10.11	14.00	11.86	0.940
Body mass	g	54	0.84	2.74	1.66	0.451
CS activity	μmol g−1 min−1	54	3.25	9.20	5.84	1.460
SMR	mg O2 h−1	54	0.15	0.41	0.25	0.061
MMR	mg O2 h−1	54	0.91	6.96	2.84	1.247
OXPHOS	pmol s−1 mg−1	54	4.36	12.84	8.10	2.058
OxCE	unitless ratio	54	0.77	0.94	0.86	0.038
ROSOXPHOS	pmol s−1 mg−1	54	0.002	0.020	0.008	0.003
Telomere length	bp	54	10.02	18.86	14.33	2.171
Slow muscle proportion	%	54	0.24	3.07	1.20	0.669

Table 3.. Statistical summary of the five linear mixed-effects models after undergoing model selection

Model	Response	Predictors	Estimate	s.e.	t / z	P
1 (n=54)	U crit	Intercept Scale (body mass) Scale (rSMR) Scale (OXPHOS/CS)	11.85889 0.54204 0.22270 0.23728	0.1065 0.0971 0.0977 0.0988	111.39 5.58 2.28 2.40	<0.001 <0.001 0.023 0.016
2 (n=54)	rSMR	–	–	–	–	–
3 (n=54)	rMMR	–	–	–	–	–
4 (n=54)	ROSOXPHOS	Intercept Scale (OxCE)	0.00790 0.00103	0.0005 0.0005	15.66 2.24	<0.001 0.025
5 (n=54)	Telomere length	–	–	–	–	–

Table 4.. Summary of the random effect (cohort ID) from our five linear mixed-effects models (n=54 for each model) after model selection

Model	Response	R2m	R2c	Random effect	Var.	s.d.
1	U crit	0.427	0.479	Cohort ID Residual	0.04702 0.47096	0.2168 0.6863
2	rSMR	–	–	–	–	–
3	rMMR	–	–	–	–	–
4	ROSOXPHOS	0.088	0.200	Cohort ID Residual	<0.00001 <0.00001	0.0012 0.0031
5	Telomere length	–	–	–	–	–

Full text

INTRODUCTION

A fish's swimming capacity underpins its ability to perform ecologically relevant tasks such as migrate, escape predators and overcome adverse ecological conditions, and is therefore a strong predictor of overall performance (Walker et al., 2005; Langerhans and Reznick, 2010; Eliason et al., 2011; Domenici et al., 2019). As such, intraspecific differences in swimming performance may help to determine longer-term patterns of fitness and longevity. Most fish swim by muscular force (McKenzie, 2011) and there are two types of skeletal muscle associated with swimming: aerobic (slow-twitch; type I) red muscle and anaerobic (fast-twitch; type II) white muscle. White muscle makes up the majority of the muscle mass and supports anaerobically fuelled swimming bursts, e.g. for escaping predators (Marras et al., 2011; Domenici et al., 2019). In contrast, although red muscle constitutes less by way of total muscle mass, it is mitochondrially rich and fuels the majority of the routine aerobic steady swimming activity among fishes (Videler, 1993; Hammer, 1995; Teulier et al., 2019). Thus, energy supply within the red muscle depends on a steady supply of metabolic fuel and oxygen within the blood (Weber et al., 2016), and this has led to the suggestion that an individual's metabolic rate may act to limit its locomotive performance (Metcalfe et al., 2016).

Whole-organism metabolic rates are typically measured in fish at their lower (standard metabolic rate, SMR) and/or upper (maximal metabolic rate, MMR) limits, with the former being measured when the fish is in a post-absorptive inactive state, and the latter being measured after exhaustive exercise (Metcalfe et al., 2016). Meanwhile, aerobic scope (AS) is defined as the difference between SMR and MMR and is thought to represent the metabolic energy available to a fish beyond its basic maintenance costs (Norin and Clark, 2016). Assuming that metabolic fuels are present in excess, fish with a higher MMR or AS are presumed to have a greater capacity for energy production and hence locomotion – a hypothesis that has found support from studies linking metabolic rate to swim performance (Reidy et al., 2000; Marras et al., 2013; Rubio-Gracia et al., 2020; Pang et al., 2021; Fu et al., 2022). When measuring swim performance, perhaps the most commonly used and standardised protocol is the critical swimming speed test (Ucrit), which involves stepwise increases in water velocity in a swim tunnel until the fish fatigues and can no longer hold position (i.e. swim) in the water column. This test initially prompts the fish to perform steady aerobic swimming, but at the highest water velocity the fish transitions to bursts of anaerobic swimming prior to fatigue. Even so, an individual's maximum sustainable swimming speed (Ums) occurs entirely aerobically and is thought to correspond to about 60–80% of Ucrit (McKenzie et al., 2021), and a number of studies have specifically identified links between Ucrit performance and whole-organism metabolic rate (e.g., Rubio-Gracia et al., 2020; Fu, Dong & Killen, 2022).

Protocols for measuring metabolic rate typically use whole-body oxygen consumption as a metabolic proxy (Killen et al., 2021), without any quantification of the actual energy that is then produced at the cellular level (Metcalfe et al., 2023). The principal energy source for most biological functions is adenosine triphosphate (ATP), and the majority of ATP (>90%) is generated within the mitochondria (Pizzorno, 2014). It is now known that individuals can vary in (1) their mitochondrial capacity for ATP production, and (2) the efficiency with which their mitochondria consume oxygen and substrates to produce ATP. Among-individual differences in the capacity for mitochondrial ATP production can arise owing to variation in the fractional volume of mitochondria within a given tissue (Hood, 2024) or owing to differing rates of enzymatic activity and/or the density of the mitochondrial complexes (Heine et al., 2023). Variation in the efficiency with which mitochondria produce ATP can also arise for a number of reasons (Brand, 2005; Salin et al., 2015; Metcalfe et al., 2023), including differing rates of proton leakage across the inner mitochondrial membrane. Elevated proton leakage requires increased oxygen and substrate consumption to compensate, without contributing towards ATP synthesis (Brand, 2005).

The obvious beneficial nature of improved mitochondrial function therefore raises the question as to why such variation in mitochondrial phenotypes persists within populations (Salin et al., 2015; Metcalfe et al., 2023). One potential theory is that having a higher mitochondrial volume and/or having mitochondria with greater respiration rates can both result in a greater production rate of reactive oxygen species (ROS) as a metabolic by-product (Hou et al., 2021; Dawson et al., 2022); however, this theory is not always empirically supported (Christen et al., 2018). Meanwhile, a greater mitochondrial efficiency (i.e. a low rate of proton leakage across the inner mitochondrial membrane) can result in high rates of ROS production owing to the greater proton gradient across the membrane (Brand, 2000; Metcalfe and Olsson, 2022). Although ROS have many functions in cells, if allowed to build up unquenched they can cause oxidative damage to biomolecules including proteins, lipids and DNA (Beckman and Ames, 1998; von Zglinicki, 2002; Reichert and Stier, 2017). Therefore, among-individual variation in mitochondrial content may have evolved by way of a trade-off between cellular energy supply and the suppression of cellular damage (Salin et al., 2015).

One way of assessing the extent of DNA damage caused by ROS is to measure the length of the telomeres that act as protective caps at the ends of chromosomes. Although telomeres shorten in length during cell division owing to the ‘end replication problem’ (Chan and Blackburn, 2004; Shay and Wright, 2019), this rate of telomere shortening is accelerated by ROS damage (Angelier et al., 2018; Barnes et al., 2018; Chatelain et al., 2020), in part because telomeres are thought to be particularly sensitive to oxidative damage caused by ROS owing to their high guanine content (Haussmann and Marchetto, 2010; Monaghan and Ozanne, 2018). Although telomere repair mechanisms do exist (such as the expression of the enzyme telomerase), such mechanisms are often downregulated in post-embryonic somatic tissues (Gomes et al., 2010; Tian et al., 2018) and therefore often do not counterbalance any telomere loss that may arise from ROS damage. Eventually telomeres may shorten to such an extent that the central coding region of a chromosome becomes vulnerable, which can then trigger the senescence or death of that cell (Victorelli and Passos, 2017; Zhu et al., 2019).

In summary, there are clearly complex trade-offs between whole-organism metabolic rate, mitochondrial functioning and telomere dynamics, which could have potential outcomes for overall performance and cellular senescence (Metcalfe and Olsson, 2022). Here, we explored the links between metabolic rate (SMR and MMR), mitochondrial function [the rates of oxygen consumption associated with oxidative phosphorylation (OXPHOS) and offsetting proton leak (LEAK)] and swim performance (Ucrit) in the European minnow (Phoxinus phoxinus). Minnows are widespread throughout large parts of Europe (Kottelat and Freyhof, 2007) and are classified as invasive in many water bodies. We also measured the mitochondrial ROS production and telomere lengths of the fish, to test whether a greater mitochondrial capacity and/or efficiency comes at the cost of long-term cellular damage. Lastly, we measured the relative proportion of aerobic (slow-twitch) and anaerobic (fast-twitch) fibres within the muscle mass of the minnows, to test whether shifts in muscle fibre type affected our metabolic measurements. We hypothesised that individuals with a greater mitochondrial capacity and/or efficiency would benefit from a greater supply of energy, which in turn would confer a greater swimming performance, but at the cost of greater ROS production and associated cellular damage.

MATERIALS AND METHODS

Experimental overview

This experiment was carried out under the jurisdiction of a UK Home Office project licence (PP6899400). We chose to conduct this study on wild Eurasian minnows [Phoxinus phoxinus (Linnaeus 1758)], because they adapt well to captive conditions (authors’ personal observations), are relatively good swimmers and perform well in swim tunnel protocols (Holthe et al., 2009; Rubio-Gracia et al., 2020). The wild minnows were captured by dip-netting on 19 February 2020 in the River Endrick, Scotland (56°03′41″N, 4°22′28″W). A total of 238 suitably sized minnows were transferred to aquarium facilities at the University of Glasgow, where they were held in circular tanks (circumference 283 cm, total volume 325 litres) and supplied with a constant turnover of aerated, copper-free water that had undergone sediment filtration, de-chlorination and UV sterilisation. During this time, minnows were fed daily ad libitum using frozen bloodworm; initially being held at ∼11°C, but then gradually increased to ∼14°C by the start of the experiment in August 2020. On 19 February 2020, all minnows were anaesthetised (with a 25 mg l^−1^ benzocaine solution) and measured for body fork length (±0.1 mm), with the majority of individuals (n=126) falling within a size class of 25–35 mm, which, according to Frost (1943), would age them at roughly 1 year old at that time. Therefore, we used individuals in this size class for our swim performance trials, so on 9 March 2020, 81 of these minnows (selected at random from the 126 individuals) were again anaesthetised (25 mg l^−1^ benzocaine solution) and were tagged on their dorsal surface (in the area between the gill plate and the dorsal fin) using Visible Implant Elastomer (VIE; Northwest Marine Technology Inc.), with each fish being given a unique four-tag combination using three colours in total. The tagged minnows were then held in the same circular tanks until the start of the experiment, which began on 18 August 2020 (the delay was caused by the lockdown restrictions surrounding the Covid-19 pandemic). In total, 54 of the VIE tagged minnows were used in this experiment and were randomly assigned to groups of three fish. Each minnow spent a total of 29 days in the experiment; see Fig. 1 for a summary of the experimental timeline. In groups of three, minnows were tested twice for critical swimming speed (Ucrit) on day 1 and then 1 week later on day 8. Fish were then allowed to recover for ∼14 days before their standard (SMR) and maximum metabolic rates (MMR) were measured between days 21 and 23. Then after a further week of rest, fish were euthanised on day 29 and tissues were sampled for mitochondrial assays, muscle histology and telomere measurement. When not undergoing trials to measure swimming speed or metabolic rate, fish were held in mesh baskets (one basket per group of three fish) that were suspended within a single tank (100×100×40 cm) that was held at a constant temperature of 14°C and were fed ad libitum on frozen bloodworm. There was no mortality of the minnows throughout the experiment.

Experimental overview. Fish were captured on 19 February 2020, tagged on 9 March 2020, and the Ucrit trials began on 18 August 2020 (the large time gap between tagging and swimming trials being due to Covid-19 lockdown restrictions). See Materials and Methods for further details of each measurement.

Measurement of critical swimming speed

To quantify prolonged swimming performance, each minnow was tested for Ucrit on its experimental day 1 and again on day 8. Ucrit tests were conducted in a 90 l Steffensen-type swim tunnel (Loligo Systems, Tjele, Denmark) at 14°C. The working section of the tunnel measured 66×20×20 cm (length×height×width) and an external chiller was used to maintain the water temperature at 14°C (the same as in the holding tanks). Fish were tested in groups of six, with three fish undergoing their first trial (day 1) and three fish undergoing their second trial (their day 8). Testing the swimming performance of shoaling species individually versus in a group can affect their Ucrit score, partly because of the stress of being held individually (Remen et al., 2016; Zeng et al., 2024); therefore, three random non-experimental minnows from the stock population were used to make up numbers if only three focal fish were being tested, e.g. for the first and last trials. Therefore, fish were always tested in a group of six, whereas the Ucrit scores were calculated for each fish individually.

Fish were fed in the morning of the day prior to a trial, meaning that they had been fasted for ∼24 h. The group of six fish was added to the swim tunnel in the afternoon prior to the trial and allowed to acclimate overnight at a slow water velocity of 2.5 cm s^−1^, which approximately corresponded to 0.5 body lengths per second (BL s^−1^), as the mean±s.e.m. fork length of the tested minnows was 5.18±0.60 cm. Minnows are naturally gregarious fish, and so the six fish formed a school and oriented into the flow. On the day of testing, room lighting was switched on at least 1 h prior to a trial starting (with each trial commencing at 10:30 h ±20 min). During the trial, water velocity was incrementally increased by 2.5 cm s^−1^ (i.e. 0.5 BL s^−1^) every 5 min until each fish was exhausted to the point of no longer being able to maintain position within the water column and thus ceasing swimming. These 5 min increments have been routinely used in other studies on small fish (Seebacher et al., 2016; Frenette et al., 2019; Thambithurai et al., 2019). A fish was identified as having reached this point if it fell sideways against the back grid of the swim tunnel and remained there for at least 5 s. Trials were observed by two scorers through a slit in a sheet of opaque black material, with one observer recording the time and water velocity (BL s^−1^) at which the individual had reached the point of exhaustion, while the other observer quickly removed each exhausted fish via a hatch at the rear of the swim tunnel and transferred them to a bucket of aerated water. This process was repeated until all six of the minnows had reached the point of exhaustion. The fork length of each fish was then measured (±0.1 mm) – using a small plastic bag filled with water to gently restrain the fish – to calculate a specific BL s^−1^ measurement for each individual. The two separate measurements of fork length (on the two separate days that Ucrit was measured) were highly correlated (Spearman's rho=0.97, N=54, P<0.001), thus showing that this was a suitable method for obtaining fork length without the need for anaesthesia. The six fish then remained in the bucket of aerated water (for no more than 15 min) to ensure that they had resumed normal swimming behaviour prior to being returned to their respective mesh holding basket.

Critical swimming speed was then calculated as per Dalziel et al. (2012) using the following formula: Ucrit=Ui+(Uii×ti/tii), where Ui is the highest speed (BL s^−1^) at which an individual was able to swim for a full 5 min increment, Uii is the incremental speed increase (0.5 BL s^−1^), ti is the amount of time that a fish was able to swim at its final speed prior to reaching exhaustion (min), and tii is the amount of time spent at each speed (i.e. 5 min). There was good repeatability between the first and second Ucrit measurements (Spearman's rho=0.61, N=54, P<0.001); therefore, the mean of both measurements was used for the subsequent analyses.

Estimation of metabolic rates

The maximum metabolic rate (MMR) and standard metabolic rate (SMR) (see Table 1) of each individual was measured using automated intermittent-flow respirometry (Killen et al., 2021). The respirometry setup consisted of nine 80 ml glass cylindrical chambers submerged within an aerated ambient water bath, held at 14°C (13.97±0.10°C, mean±s.d.). Each chamber had its own mixing circuit, consisting of (1) the chamber, (2) a probe vessel containing a single oxygen probe and (3) a circuit of gas-impermeable tubing. Each of the nine oxygen probes were attached to a Firesting O_2_ 4-channel oxygen sensor (FSO2-4, PyroScience, Aachen, Germany; up to four probes per sensor) and were set to measure oxygen content every 2 s. A peristaltic pump (set to 100 rpm) was used to circulate water through the mixing circuits. Throughout the measurement period, each circuit alternated between flush cycles (lasting 3 min) and closed cycles (lasting 8 min). During the closed cycles, the automated flush pumps were switched off to prevent fresh aerated water from entering the circuits, thus allowing the rate of oxygen depletion (i.e. the rate of oxygen uptake by the fish) to be measured.

In preparation, fish were not fed on the day on which they were transferred to the respirometry chambers (so had not been fed since the previous morning, to ensure that their gut would be empty). At around 16:00 h, fish were exposed to a 3-min exhaustive chase protocol, prior to being transferred to their respective respirometry chamber. This protocol assumes that the subsequent rate of oxygen uptake by the fish is at its maximum, and can be captured after transfer to the chamber (Killen et al., 2015); it has been shown to yield the same results as when MMR is measured by an incremental swimming test (Killen et al., 2017b; Zhang et al., 2020; but see also Raby et al., 2020). Therefore, the highest rate of oxygen uptake by the fish during the first closed cycle of the respirometer was used as our estimate of MMR (mg O_2_ h^−1^) and was calculated based on a 1 min rolling regression of the rate of oxygen depletion; see Table S1 for further details. Fish then remained in their chambers overnight until around 09:30 h the following morning in order to calculate SMR. During this time, the rate of oxygen uptake (mg O_2_ h^−1^) was calculated separately for each of the closed cycles, and SMR was estimated as the mean of the lowest 20th percentile of these rates (with the first 5.5 h of oxygen measurements being excluded from the SMR calculations). The AS for each fish was then calculated as the difference between its MMR and SMR. Fish were then removed from their chambers and anaesthetised to allow accurate measurement of body mass (±0.001 g), which was then used in the subsequent calculations of mass-corrected residual metabolic rates (see Fig. 2). The two previous measurements of body fork length (taken on the same days as the Ucrit measurements) were for the purpose of the Ucrit (BL s^−1^) calculations, but only this final measurement of body mass was used for the subsequent analyses.

Relationships between individual body mass (g) and metabolic variables. (A) Standard metabolic rate (SMR, correlation coefficient=0.78, n=54), (B) maximal metabolic rate (MMR in mg O2 h−1, correlation coefficient=0.56, n=54) and (C) aerobic scope (AS in mg O2 h−1, correlation coefficient=0.53, n=54). Both variables were log10-transformed for the MMR and AS regressions as this improved the normality of the residuals from these analyses. Residuals from these three regression equations were then used to create three new variables: (D) residual standard metabolic rate (rSMR), (E) residual maximal metabolic rate (log rMMR) and (F) residual aerobic scope (log rAS), respectively. See Materials and Methods for statistical analyses.

Tissue sampling

When a given individual had reached experimental day 29 (see Fig. 1), it was euthanised with a 50 mg l^−1^ solution of benzocaine and its muscle tissue was sampled for various measurements. The muscle mass from one side of the fish was sampled (tissue mass range: 47–219 mg) for mitochondrial analysis (see ‘Measurement of mitochondrial respiration and ROS production rates’) and immediately transferred to 2 ml of ice-cold mitochondrial respiration buffer MiR05 (0.5 mmol l^−1^ EGTA, 3 mmol l^−1^ MgCl_2_, 60 mmol l^−1^ lactobionic acid, 20 mmol l^−1^ taurine, 10 mmol l^−1^ KH_2_PO_4_, 20 mmol l^−1^ HEPES, 101 mmol l^−1^ d-sucrose, 1 g l^−1^ essential fatty acid free BSA; pH 7.3). A subsample of muscle mass from the other side of the fish was taken for the histological analysis of the relative proportion of aerobic (slow-twitch) and anaerobic (fast-twitch) muscle fibres within the muscle mass of that individual (see ‘Estimation of aerobic versus anaerobic muscle’). To do so, a transverse cross-section of myotome muscle was sampled, ensuring that it contained both the dorsal (epaxial) and ventral (hypaxial) muscles. Muscle sections were first fixed in a 10% formaldehyde solution for 3 days and then transferred to 70% ethanol for long-term storage. The remaining tissue from the second side of the fish was then flash-frozen and transferred to a −70°C freezer for long-term storage, for the subsequent measurements of citrate synthase (CS) activity and telomere length (see below).

Measurement of mitochondrial respiration and ROS production rates

Preparation of the mitochondrial tissue samples (a combination of red and white muscle) followed the same protocol as outlined in Dawson et al. (2022). Following dissection and the immediate transfer to MiR05 buffer (as outlined in ‘Tissue sampling’), the tissue was first minced in the buffer with microsurgical spring scissors, followed by six rounds of gentle homogenisation using a Dounce homogeniser (Cole-Parmer PTFE Tissue Grinder, Cambridgeshire, UK) at 100 rpm.

Mitochondrial function was analysed using high-resolution respirometry (Oxygraph-2k with O2k-Fluorescence module, Oroboros Instruments, Innsbruck, Austria), following modified protocols by Dawson et al. (2022) and Dawson and Scott (2022). The protocols were conducted at 14°C with continuous stirring, with samples consisting of 30 mg of tissue in a final volume 2 ml of respiration buffer. The rate of ROS production was measured following the protocol of Dawson et al. (2018), using the O2k-FluoRespirometry module in the presence of Ampliflu Red (15 µmol l^−1^), and after the addition of a known quantity of exogenous H_2_O_2_ in the presence of horseradish peroxidase (3 U ml^−1^).

The respiration rates of the mitochondria were measured when oxygen consumption had stabilised for at least 5 min, following the addition of the substrates malate (2 mmol l^−1^) and pyruvate (5 mmol l^−1^), followed by ADP (5 mmol l^−1^), then glutamate (10 mmol l^−1^) and finally succinate (25 mmol l^−1^). This rate of respiration in the presence of all substrates represents the maximum rate of oxidative phosphorylation (hereafter termed OXPHOS capacity), as it is occurring under conditions where both Complex I and Complex II are able to operate with an excess of substrates and ADP. Therefore, OXPHOS capacity can be used as a proxy of the maximum capacity for mitochondrial ATP production. Cytochrome c (10 mmol l^−1^) was then added to assess the quality of the mitochondrial preparations, because respiration in the presence of cytochrome c is a measure of mitochondrial membrane integrity (Kuznetsov et al., 2004).

Next, oligomycin (an inhibitor of F_1_F_0_ ATP synthase) was added to measure leak state respiration (the oxygen consumption that is used to offset the leakage of protons across the inner mitochondrial membrane rather than to generate ATP; hereafter termed LEAK). Finally, antimycin A (an inhibitor of complex III) was added to assess non-mitochondrial (i.e. background) oxygen consumption, which was then subsequently subtracted from all respiration measurements.

Measurement of citrate synthase activity

CS activity is a commonly used biomarker for mitochondrial volume (Larsen et al., 2012; McLaughlin et al., 2020), thereby allowing us to better control for mitochondrial volume in the measurements of OXPHOS capacity. The maximal rate of CS activity was measured in our tissue samples at 14°C using a SpectraMax Plus 384 spectrophotometer (Molecular Devices), following a protocol similar to that of Dawson et al. (2020). The sub-sample of frozen muscle tissue was ground in liquid nitrogen and then homogenised using a Power Gen 125 homogeniser (Fisher Scientific) in 20 volumes of cold homogenisation buffer [100 mmol l^−1^ KH_2_PO_4_, 1 mmol l^−1^ EGTA, 1 mmol l^−1^ EDTA, 0.1% Triton X-100 at pH 7.2 and 1 mmol l^−1^ phenylmethylsulfonyl fluoride (PMSF)]. These diluted homogenates were then centrifuged at 1000 g for 5 min at 4°C and the supernatant was collected. The change in absorbance was measured in triplicate at 412 nm within the linear range of the assay over time, using the following assay conditions: KH_2_PO_4_ 100 mmol l^−1^, pH 8.0, acetyl-CoA 0.15 mmol l^−1^, 5,5′-dithiobis-2-nitrobenzoic acid 0.15 mmol l^−1^ and oxaloacetate 0.5 mmol l^−1^. Our control was measured in the absence of oxaloacetate to confirm that our measured activity rate was specific to CS. Using an extinction coefficient (ε) of 14.15 mmol^−1^ cm^−1^, the enzyme activity was expressed in µmol per gram of tissue per minute. The among-triplicate repeatability (%CV) of the CS measurements was 3.87±0.4%, indicating a high degree of repeatability.

Calculation of mitochondrial traits

Our measurements of OXPHOS capacity were divided by CS activity (hereafter OXPHOS/CS) in order to standardise for mitochondrial volume in the samples (Dawson et al., 2018). The respiratory control ratio (RCR) was calculated as RCR=(OXPHOS capacity/LEAK); from this we then calculated the OXPHOS coupling efficiency (OxCE) as OxCE=1–(1/RCR). This represents the proportion of consumed oxygen that is devoted to OXPHOS (rather than to offset the proton leak) and can thus be used as a proxy for the efficiency with which the mitochondria are using oxygen to convert fuel into ATP. It has the advantage over RCR as being a more intuitive and constrained variable since it ranges between 0 (least efficient) and 1 (most efficient). Note that OxCE is also known as P–L control efficiency (Gnaiger, 2020; Roussel et al., 2023) and net phosphorylation efficiency (Shama et al., 2016; Dawson et al., 2024), and is sometimes considered an abbreviation of oxidative control efficiency. Therefore, we ended up with two final measures of mitochondrial respiration that were used in subsequent analyses: OXPHOS/CS (i.e. a measure of the mitochondrial capacity for ATP production, correcting for mitochondrial volume) and OxCE (i.e. the efficiency with which the mitochondria use oxygen when producing ATP).

Estimation of aerobic versus anaerobic muscle

In order to establish what proportion of an individual's muscle section consisted of aerobic (slow-twitch) versus anaerobic (fast-twitch) fibres, we adopted an immunohistochemistry (IHC) approach, which utilised the Anti-Skeletal Muscle Myosin Antibody F59, as this has previously been shown to label aerobic muscle strongly and anaerobic muscle faintly in zebrafish (Devoto et al., 1996; Barresi et al., 2000). The muscle samples were embedded longitudinally in wax, with 2.5 µm sections being cut and incubated overnight at 37°C. The IHC was performed on a DAKO Autostainer Link 48, and heat induced antigen retrieval was achieved using a sodium citrate buffer (pH 6). The primary antibody was F59 (Santa Cruz catalogue ref. sc-32732; diluted 1:200) and the secondary antibody was RTU Dako Envision mouse HRP (catalogue ref. K4001), with both antibodies being incubated for 30 min at room temperature. Visualisation was then achieved using Dako DAB+Substrate (catalogue ref. K3468) with a 10 min incubation. The sections were counterstained using Gill’s Haematoxylin for 27 s, before being dehydrated through graded alcohols, and then cleared and mounted using a synthetic resin.

Images of the IHC slides were captured using a Leica DFC 450C digital camera mounted on a Leica M165 FC stereomicroscope, in combination with LAS v4.4 software (Leica Microsystems). ImageJ software (National Institutes of Health, Bethesda, MD, USA) was used to distinguish between the strongly stained aerobic muscle fibres and the weakly stained anaerobic muscles fibres; see Fig. S2 for an illustrative image. The entire area of the muscle section was outlined and measured (in number of pixels) using the freehand selection tool. In most species of fish, aerobic muscle fibres are located in a wedge-shaped region at the lateral end of the horizontal myoseptum (that separates the dorsal and ventral muscles) (Devoto et al., 1996). Therefore, in order to distinguish between true aerobic muscle in this wedge-shaped region (that was clearly visible in the captured images) and residual F59 staining in deeper regions of the muscle section, the freehand selection tool was again used to outline and measure (in pixels) this wedge-shaped region of aerobic muscle. From that, the proportion of the muscle section that consisted of aerobic (slow-twitch) fibres was then calculated, with the remainder being anaerobic (fast-twitch) fibres.

Measurement of telomere length

Although absolute telomere lengths may differ among tissues, studies have found strong correlations in telomere length between different tissues in a range of vertebrate taxa: birds (Reichert et al., 2013), lizards (Rollings et al., 2019) and fish (Debes et al., 2016). In this study we extracted genomic DNA from each individual using the frozen subsample of ground muscle tissue using the Qiagen Puregene Tissue Kit following the manufacturer's protocol, but with one additional step to remove lipid contamination. For this additional step, following the lysis step of the Puregene protocol, 10 ml of chloroform was added to the lysate, which was then mixed by inversion and centrifuged at 4500 rpm for 10 min. The supernatant (containing the DNA) was carefully pipetted off and the chloroform step was then performed on that supernatant for a second time, before continuing with the precipitation step of the Puregene protocol. Following DNA extraction, a total of 10 µg DNA was then digested overnight at 37°C using the restriction enzymes HinfI, HaeII and RsaI (New England Biolabs, Hitchin, UK). Standardising each sample to a final DNA content of 10 µg required differing volumes of DNA. Therefore, to then standardise both volume and concentration, the digestion step was followed by an ethanol precipitation step (0.1×the final volume of sodium acetate 3 mol l^−1^ and 2.5×the final volume of 100% ethanol). The resulting DNA pellets were resuspended in 13 µl DNA Hydration Solution (provided with the Puregene Tissue Kit) and then stored at −20°C until telomere length measurement.

The mean telomere length of each individual was measured using telomere restriction fragment (TRF) analysis (Mender and Shay, 2015) using a protocol similar to that of Haussmann and Mauck (2008). In brief, samples (i.e. 13 µl digested DNA) were run on a 0.8% agarose gel using a Chef DRII pulsed-field gel electrophoresis system (Bio-Rad) and under the following conditions: 3 V cm^−1^, switch time 0.5 to 7.0 s for 19 h at 14°C in 0.5% TBE buffer (Life Technologie). Following electrophoresis, the gel was dried using a Slab Gel dryer model GD 2000 (Hoefer) pre-hybridised in Church and Gilbert's Hybridization Buffer (VWR) for 1 h at 37°C and then hybridised overnight at 37°C with a radioactive ^32^P-labelled oligonucleotide probe (CCCTAA)4, which binds the 3′ single strand overhang telomere sequence. The gel was then placed on a phosphor screen (Amersham Biosciences) for 4 days. Following this period, the phosphor screen was scanned using a Typhoon Variable Mode imager (Amersham Biosciences). ImageJ software was then used to visualise the telomeres, which could then be measured by comparing them with a radioactively ^32^P-labelled molecular marker (Quick loading 1 kb Extend DNA Ladder, New England Bioscience), which had also been run on the gel.

Statistical analysis

We first examined the relationships between body mass and each of SMR, MMR and AS using regression analyses. Both body mass and the corresponding metabolic variables were log_10_-transformed for the MMR and AS regressions as this improved the normality of the residuals from these analyses. In contrast, log transformation did not improve normality in the body mass/SMR regression and so both variables were left unlogged for that analysis (see Fig. 2). The residuals generated from these three regression equations were then used to create three new variables, residual standard metabolic rate (rSMR), residual maximal metabolic rate (rMMR) and residual aerobic scope (rAS), with these new variables allowing us to differentiate individuals with higher than expected metabolic rates for their body mass (i.e. those with positive residuals) from those with lower than expected metabolic rates (those with negative residuals). rMMR and rAS were highly colinear (r^2^=0.997, P<0.001, n=54), and so only rMMR was used as a variable in the subsequent statistical analyses.

A summary of all the measured traits is presented in Table 2. Data were analysed with a total of five linear mixed-effects models using R (version 4.5.1 in R Studio version 2025.05.1) and the glmmTMB package (McGillycuddy et al., 2025). Model 1 examined whether Ucrit varied as a function of body mass, rSMR, rMMR, OXPHOS/CS, OxCE or the proportion of aerobic fibres within its muscle mass. Model 2 examined whether rSMR varied as a function of body mass, rMMR, OXPHOS/CS, OxCE or the proportion of aerobic fibres within its muscle mass. Model 3 examined whether rMMR varied as a function of body mass, rSMR, OXPHOS/CS, OxCE or the proportion of aerobic fibres within its muscle mass. Model 4 examined whether rates of mitochondrial ROS production (ROS_OXPHOS_) varied as a function of body mass, rSMR, rMMR, OXPHOS/CS or OxCE. Lastly, model 5 examined whether telomere length varied as a function of body mass, ROS_OXPHOS_, rSMR, rMMR, OXPHOS/CS or OxCE. See Table S2 for a summary of the five statistical models. In order to obtain more parsimonious models, we also applied model selection to each model using the Bayesian information criterion (BIC), whereby fixed terms were sequentially removed from the model (providing that this resulted in a relative reduction of the BIC score) until only significant terms remained; see Table 3 for a summary of these slimmed models. All five models included a random intercept term for cohort ID to account for non-independence among individuals of the same experimental group (see Tables 4 and S3 for a summary of the random slopes in the full and slimmed models, respectively).

Continuous variables were mean centred with a standard deviation of 1 using the scale () function, in order to allow comparison between coefficients. Owing to logistical limitations on our sample sizes, interaction terms were not included as they would have made the models over-parameterised and they lacked any obvious biological plausibility. All models were tested for singularity and any models with a singular fit were refitted with a specified gamma prior (Chung et al., 2013) using the glmmTMB package. Residuals from our linear mixed-effects models were tested using a Shapiro–Wilk test, which confirmed normality in all five models.

RESULTS

The critical swimming speed (Ucrit) of a fish was unrelated to measures of mitochondrial efficiency (OxCE), nor was it significantly linked to its MMR or to the proportion of aerobic fibres within its muscle mass. However, Ucrit was significantly influenced by an individual's body mass (model 1: P<0.001) and by its rSMR (model 1: P=0.023; Table 3). Thus, smaller fish reached exhaustion at a relatively higher number of body lengths per second (Fig. 3A) and individuals with relatively higher standard metabolic rates for their size also had higher Ucrit scores (Fig. 3B). Moreover, there was a significant relationship between OXPHOS/CS and Ucrit (model 1: P=0.016; Table 3), indicating that individuals with a greater OXPHOS capacity also had higher Ucrit scores for their body size (Fig. 3C). Variation in rSMR (model 2) and rMMR (model 3) was not explained by any of the variables included in the respective models (Table 3). However, there was a significant link between OxCE and rates of mitochondrial ROS production (model 4: LM: P=0.025; Table 3), as individuals with more efficient mitochondria (in terms of the proportion of oxygen that was used for OXPHOS rather than for offsetting the proton leak) also experienced higher rates of ROS production during OXPHOS (Fig. 4). Lastly, telomere length was unrelated to any of the measured variables (model 5: Table 3).

Relationship between an individual's Ucrit score and its body mass, rSMR and OXPHOS/CS. (A) Body mass (correlation coefficient=−0.55, n=54), (B) residual standard metabolic rate (rSMR, correlation coefficient=0.30, n=54) and (C) muscle mitochondrial capacity for oxidative phosphorylation respiration (OXPHOS/CS, correlation coefficient=0.31, n=54). Data are plotted as individual fish with regression line. Ucrit scores in B and C have been plotted as partial residuals when body mass is held constant. See Materials and Methods for statistical analyses.

Relationship between the OXPHOS coupling efficiency (OxCE) of muscle mitochondria and the production rate of reactive oxygen species, measured at the time of OXPHOS (ROSOXPHOS, correlation coefficient=0.29, n=54). Data are plotted as individual fish with regression line. See Materials and Methods for statistical analyses.

DISCUSSION

We have shown in this study that there can be significant links between an individual's baseline metabolism, its mitochondrial biology and how it performs in a test of its critical swimming speed. When studying critical swim performance in absolute terms (i.e. Ucrit measured as cm s^–1^), larger fish will invariably outperform smaller fish (e.g. Palstra et al., 2020). A commonly used correction for this is to normalise the Ucrit measurements to body size (i.e. expressing speed through the water as BL s^−1^). However, this often results in a significant negative relationship, with larger fish reaching exhaustion when swimming at a relatively lower number of body lengths per second (Hvas and Oppedal, 2019; Thambithurai et al., 2019; McKenzie et al., 2021), and this relationship was found in the present study. Many factors relevant to swimming performance, such as gill area, metabolic rate and the relative level of glycogen stored within the muscle, scale allometrically to fish body size with an exponent (b) less than 1 (Vornanen et al., 2011; Norin and Gamperl, 2018; Skeeles and Clark, 2024), and so may contribute to body size limitations on swimming capacity. Moreover, an individual's body size is intrinsically linked to its rate of growth and it may also be the case that the larger individuals in our population had attained that size owing to a greater energetic investment in growth, which could have theoretically been traded off against investment in swimming capacity (Killen et al., 2014).

We also found a significant effect of SMR on swimming capacity, as individuals with larger rSMR values also had higher Ucrit scores. An individual's SMR is thought to represent its minimal or essential ‘cost of living’, and so individuals with larger and/or more active organs should require an elevated SMR in order to facilitate the maintenance of these essential organs (Konarzewski and Książek, 2013; Metcalfe et al., 2016). For example, a study on brown trout by Norin and Malte (2012) found a significant and positive link between SMR and the activity level of multiple different liver enzymes, thus lending support to rSMR being a good measure of the metabolic machinery of an individual fish. This would suggest that individuals with a higher SMR should also exhibit a greater swimming capacity, as a more active liver will facilitate the faster processing of nutrients and so sustain muscular activity at a higher level. However, it is therefore unclear why we did not find a significant link between maximal metabolic rate and Ucrit, as SMR and MMR are often correlated (Auer et al., 2017) and MMR is thought to reflect an individual's maximal capacity for energy generation (Metcalfe et al., 2023). One suggestion by Raby et al. (2020) is that measuring MMR after an exhaustive chase protocol (i.e. the protocol adopted in the present study) may actually underestimate an individual's true MMR by ∼20%. Even so, the exhaustive chase protocol is still thought to be a good proxy of true MMR, and a meta-analysis by Killen et al. (2017b) found differing MMR protocols to yield similar results so long as these protocols were conducted robustly.

Although Ucrit was not found to be related to MMR, we did find it to be associated with the capacity for OXPHOS by muscle mitochondria, perhaps because OXPHOS is a more direct measurement of the true capacity of the muscle to generate energy for sustained activity. Indeed it has previously been shown that OXPHOS in human muscles is related to muscle performance (Berg et al., 2019; Mau et al., 2023) and increases in response to the intensity of exercise training (Granata et al., 2018; Fiorenza et al., 2019). However, variation in OXPHOS is not the only contributor to variation in cellular energy production, as the latter can also be enhanced via increased mitochondrial volume density, increased surface area of the inner mitochondrial membrane, or via changes in the enzymatic activity of the mitochondrial complexes (Salin et al., 2015; Heine et al., 2023; Hood, 2024). In support of this, there have been a limited number of studies on fish which have shown that exercise training can in fact induce changes in estimated mitochondrial volume (McClelland et al., 2006; Pengam et al., 2021). Individuals may also meet energetic demands by increasing the efficiency with which ATP is produced (Salin et al., 2015; Conley, 2016). A previous study has found positive correlations between the efficiency of ATP production and muscle contraction performance (Distefano et al., 2018). However, our measurement of mitochondrial efficiency (OxCE) was unrelated to Ucrit, suggesting that other measures of energy production were more important.

If there is clearly a benefit for tissues to have a readily available source of energy, it then raises the question as to why variation in mitochondrial function (and thus energy availability) exists among conspecifics. One suggestion is that although it is beneficial for tissues to contain sufficient mitochondria to support their energetic demands, having a higher mitochondrial volume and/or mitochondria with greater respiration rates can result in a greater rate of ROS production (Hou et al., 2021). Therefore, among-individual variation in mitochondrial content may have evolved by way of a trade-off between cellular energy supply and the suppression of cellular damage (Salin et al., 2015). However, we did not find a significant link between OXPHOS and ROS production. Instead, we found a significant link between OxCE and ROS production. Although OxCE was unrelated to swimming performance or to our measures of whole-animal metabolism, individuals with more efficient mitochondria (i.e. mitochondria that contributed a greater proportion of oxygen to ATP production, rather than to offset leakage of protons across the inner mitochondrial membrane) also exhibited greater levels of mitochondrial ROS production. One explanation for this could be the fact that more efficient mitochondria (i.e. those with a lower rate of proton leakage across the inner mitochondrial membrane) may then generate higher rates of ROS production owing to the greater proton gradient across the membrane (Brand, 2000; Metcalfe and Olsson, 2022).

We also hypothesised that individuals with greater rates of ROS production would consequently also have shorter telomeres, as it is now widely accepted that telomere shortening is accelerated by ROS damage (Monaghan and Ozanne, 2018; Chatelain et al., 2020); however, we found no significant links between ROS production and the telomere lengths of the minnows. The minnows in this study were wild caught and it is possible that any stress they experienced up until the point of capture could have affected their telomere length; however, they were captured at the same location, so they were likely exposed to similar environmental stressors. We also now know that the telomere dynamics of ectotherms differ to those of endotherms, where the length of the telomeres in somatic tissues (especially of larger or longer-lived species) tends to shorten throughout post-natal life (Monaghan, 2024). In contrast, the telomere dynamics of ectotherms are less predictable, and there are multiple reports of shortening, lengthening or even a combination of both as an individual ages (Rollings et al., 2014; Peterson et al., 2015; McLennan et al., 2017; Ujvari et al., 2017). One suggested cause of this lack of consistent shortening is the fact that most ectotherms exhibit indeterminate growth, and may thus have evolved telomere elongation processes later into their life in order to balance this sustained proliferative capacity (Gomes et al., 2010). Although there are several different telomere elongation mechanisms, the most common is thought to be via the enzyme telomerase, which has been shown to be active in the post-embryonic somatic tissues of various fish species (Yap et al., 2005; Alibardi, 2015; Peterson et al., 2015; Hatakeyama et al., 2016; Yip et al., 2017). We did not measure telomerase expression in our minnows, and little is known about whether telomeres can lengthen over relatively short time frames (as in this study). Therefore, it remains unclear whether there simply was no effect of ROS production on the length of the telomeres, or whether any potential effects were being buffered, e.g. via telomerase activity.

Although we have shown in this study that there can be significant links between an individual's OXPHOS capacity and its swimming capacity (Ucrit), it is worth mentioning that the Ucrit test may have been unable to account for some of the individual variation in swimming performance. For example, at the highest water velocities, fish typically transitioned for a short period from aerobic swimming to bursts of anaerobic swimming prior to fatigue; however, it was not possible to measure the point of this transition for each individual. Also, minnows are a shoaling species and so we chose to assess the swimming capacity of the minnows while swimming in a shoal, so as both to minimise the stress associated with isolating a social species and to make the trials as ecologically relevant as possible. However, it is not possible to know whether an individual's chosen position within that shoal may have affected its Ucrit score. For example, a previous study of European minnows found that a more anterior position in a shoal is beneficial in terms of access to food (McLean et al., 2018), whereas other studies have shown that individuals holding posterior shoaling positions benefit from a reduced energetic cost due to the flow patterns generated by the fish swimming in front (Killen et al., 2017a,b). It has also been shown that fish in a shoal use less energy than if those same individuals are swimming alone (Marras et al., 2015), although this might partly be due to the stress of isolating individuals of a naturally shoaling species. Therefore, it is difficult to know whether the patterns of swimming in our minnows were influenced by shoal dynamics. Future studies of a similar nature to this one could benefit from using a solitary fish species. It would also be beneficial to establish the causality of the observed relationships, as we currently do not know whether it is a greater OXPHOS capacity that allows an individual fish to be a better swimmer, or whether those fish that are better swimmers (for whatever reason) develop a greater OXPHOS capacity to fuel their high energetic requirements (similar to the effects of exercise training on OXPHOS capacity; Granata et al., 2018). Related to this, future studies would benefit from non-destructive sampling methods (Stier et al., 2019; Quéméneur et al., 2022; Thoral et al., 2024) in combination with measuring mitochondrial performance both pre- and post-exercise. Nonetheless, although it is only relatively recently that there has been a shift in focus towards the ways in which energy is produced at a cellular level, this study exemplifies the close associations between cellular energy production and overall performance. It will now be important in future studies to see whether this relationship holds true under fluctuating environmental variables, such as temperature and oxygen availability.

Supplementary Material

10.1242/jexbio.251517_sup1Supplementary information