Are acute and acclimated thermal effects on metabolic rate modulated by cell size? A comparison between diploid and triploid zebrafish larvae

The substantial variation in cell size observed among extant animal species is largely mirrored by variation in genome size (Gregory, 2001; Dufresne and Jeffery, 2011). Although cell size and genome size are strongly and positively correlated, the causality is not fully resolved (Gregory, 2005; Czarnoleski et al., 2018). While cell size and genome size may seem esoteric aspects of an organism, there is evidence that it could have important life-history ramifications owing to their influence on body mass, metabolic rate and the mass scaling of metabolic rate (Gregory et al., 2000; Kozłowski et al., 2003; Czarnoleski et al., 2018; Kozłowski et al., 2020). Smaller cells have a higher ratio of membrane surface area to cell volume. Therefore, they will require more energy for phospholipid turnover and to maintain ionic gradients between the cytoplasm and the cell's surroundings (Szarski, 1983; Rolfe and Brown, 1997; Czarnoleski et al., 2013, 2015a). Studies on freshly isolated tissues and cultured cells have indeed found that smaller cells exhibit higher mass-specific metabolism compared with larger cells (Goniakowska, 1970; Monnickendam and Balls, 1973). In addition to higher demand for energy, smaller cells also have a greater capacity for oxygen uptake, as intracellular diffusion distances are shorter (Woods, 1999; Miettinen et al., 2017) and oxygen diffuses more rapidly in lipid membranes than in aqueous cytosol (Subczynski et al., 1989). As a result, risks of oxygen limitation are hypothesized to be greater for tissues composed of larger cells (Atkinson et al., 2006; Czarnoleski et al., 2013; Verberk et al., 2020). Thus, there are distinct metabolic consequences to cell size.

Previous studies investigating the link between metabolic rate and cell size have exploited natural variation in genome size and cell size within species and complexes of closely related species (reviewed by Hermaniuk et al., 2017) or focused on interspecific comparisons (Vinogradov, 1995; Gregory, 2002, 2003; Starostová et al., 2009; Czarnoleski et al., 2018; Gardner et al., 2020). Comparing species with different evolutionary histories is problematic as it is difficult to isolate the effect of cell size (or its proxy genome size) on metabolic rate. Support for a connection between cell size and whole-body metabolic rate was not consistently found in different animal groups. Polyploidy originates almost exclusively in ectothermic animals as a consequence of duplication of entire chromosome sets (Otto, 2007; Choleva and Janko, 2013) and the increase in genome size is associated with increases in cell size across a range of tissues (Fankhauser, 1945; Suresh and Sheehan, 1998; Hermaniuk et al., 2016). Despite the advantages of using variation in ploidy within species, research on the relationship between cell size and whole-body metabolic rate in animals of different ploidy also provided inconclusive results.

A way forward in clarifying the relationship between cell size and whole-body metabolic rate could be the need to include temperature. In ectotherms, warming is known to reduce cell size (Arendt, 2007; Hessen et al., 2013; Walczyńska et al., 2015) for most, but not all, tissue types (Czarnoleski et al., 2016, 2017). Such changes in cell size could be adaptive to reduce risks of oxygen limitation (Szarski, 1983; Walczyńska et al., 2015; Verberk et al., 2020), as warming increases metabolic oxygen demand (Clarke and Fraser, 2004), possibly resulting in a mismatch between oxygen demand and (the capacity for) oxygen supply (Pörtner, 2010; Verberk et al., 2011). Indeed, the concept of understanding thermal responses from an oxygen perspective is gaining traction (e.g. Pörtner, 2010; Czarnoleski et al., 2015b; Hoefnagel and Verberk, 2015; Verberk et al., 2016b), especially for aquatic ectotherms, as oxygen uptake is more challenging in water than in air due to the much lower diffusion rate of oxygen (Woods, 1999; Verberk et al., 2011). Thus, metabolic consequences of cell size in aquatic ectotherms could depend on temperature, with smaller cells being more advantageous in warmer conditions.

A second key ingredient could be to include test conditions that are pushing animals to their limits. Previous work comparing metabolic rate in animals of different ploidy focused on either standard or routine metabolic rate, which relates to energetic costs of maintaining basic life function. Under these conditions, the large cells of polyploids are less likely to constrain metabolic rate – a pattern observed in many studies (invertebrates: Ellenby, 1953; Shpigel et al., 1992; fish: Sezaki et al., 1991; Parsons, 1993; Hyndman et al., 2003; amphibians: Licht and Bogart, 1990; Lukose and Reinert, 1998). However, capacity limitations for oxygen uptake are more likely to be manifested under challenging conditions, such as high temperatures, strenuous activity or low oxygen conditions (Hyndman et al., 2003; Atkins and Benfey, 2008; van de Pol et al., 2020; Rubalcaba et al., 2021).

Given the thermal sensitivity of aquatic ectotherms, we designed a full factorial experiment that investigates the link between cell size and whole-body metabolic rate in fish larvae combining different rearing (developmental) and test temperatures and different levels of oxygen availability. This approach allows us to distinguish between the two temperature effects (acute and acclimated) that are regulated by different mechanisms (Havird et al., 2020). Acute test temperature induces passive responses shaped by the thermodynamics of molecular interactions and subsequent immediate energetic feedbacks while acclimation (developmental) temperature induces active responses through molecular mechanisms including gene expression, membrane composition and enzyme concentrations (Angilletta, 2009; Havird et al., 2020). As a model system we used diploid and artificially induced triploid zebrafish larvae [Danio rerio (Hamilton 1822)], which have previously been shown to differ in genome size and cell size; the 50% increase in DNA content results in a 1.5-fold increase in cell size (erythrocytes were directly measured) in triploids, while their cell number was reduced by the same factor of 1.5 (van de Pol et al., 2020). Hence, triploids were of similar body size, consisting of larger but fewer cells compared with their diploid counterparts. A similar correlation between erythrocyte size as a proxy of cell size of other tissues has been reported for amphibians (Kozłowski et al., 2010). A previous study showed that under non-demanding conditions, diploid and triploid zebrafish larvae were highly similar in terms of gene expression, growth and development (van de Pol et al., 2020). Furthermore, zebrafish larvae rely on cutaneous oxygen uptake, as their gills are not fully developed until 14 days after fertilization (Kimmel et al., 1995; Rombough, 2002). Therefore, they cannot increase their capacity to deliver oxygen by increasing gill ventilation rates. Taken together, this makes zebrafish larvae an excellent model for studying the consequences of cell size variation on thermal physiology of ectotherms.

Although some studies have reported triploid individuals to have reduced aerobic metabolism at elevated developmental temperatures (Stillwell and Benfey, 1996; Atkins and Benfey, 2008), none has looked at the interacting effects of developmental and test temperatures combined with lowering P_O2 on metabolic rate in diploid and triploid animals. We hypothesized that triploids composed of larger cells have a lower capacity for oxygen supply and that this would be manifested as a lower metabolic rate at higher water temperatures and in hypoxic water, when compared with diploid counterparts. We expected that developmental temperature would modulate the effect of test temperature such that animals reared at high temperatures would be better able to meet their elevated oxygen demand at high test temperatures. We also assessed hypoxia tolerance in diploid and triploid fish at different temperatures, by means of determining P_crit, which is the value of P_O2 below which the animal can no longer maintain a stable rate of oxygen consumption. Wood (2018) recently questioned the usefulness of using a single P_crit value as an indicator of hypoxia tolerance and suggested that instead one should consider the entire metabolic response to declining oxygen tensions. Here we compared different methodological approaches to assess the ability of the animal to regulate its metabolic rate across progressively lowering P_O2.

The sources of parental zebrafish and the induction of triploidy have been described in detail elsewhere (van de Pol et al., 2020). Briefly, parental zebrafish [wild-type strain supplied by Zebrafish International Resource Center (ZIRC), ZFIN ID: ZDB-GENO-960809-7] were kept in tanks (4 liters volume) with recirculating tap water (temperature 27°C, pH 7.5–8) containing about 30 individuals under 14 h:10 h light:dark photoperiod. Once a week, one tank was chosen to perform in vitro fertilization (Westerfield, 2000; see van de Pol et al., 2020 for an in-depth description of the procedure) using randomly selected males and females. To ensure good quality eggs, zebrafish females were given about 20 days rest after spawning; therefore, care was taken not to re-use the same tank within 3 weeks. The procedure was repeated for eight consecutive weeks (i.e. eight rounds of fertilization). To induce triploidy, we used a cold shock treatment by immersing eggs 3 min after fertilization in a tank with 4°C E2 medium (standard medium for zebrafish care containing 5 mmol l⁻¹ NaCl, 0.17 mol l⁻¹ KCl and 0.33 mmol l⁻¹ MgSO₄) for 20 min. For each round of fertilization, eggs were obtained from multiple females: three to four females were used to produce diploid progeny, and three to five females were used to produce triploid progeny, except for round eight, where we used only two females to produce diploids and triploids, one in each group. To secure the heterogeneity of the offspring, all eggs were fertilized using the pooled sperm suspension of eight to 12 males. Egg quality was visually assessed by checking their colour, transparency and shape. We only used eggs that appeared yellowish and translucent, with a regular round shape. Eggs that did not meet the accepted criteria were removed. Before transferring the fertilized eggs to the rearing tanks (see below), the progeny of different females of a given ploidy level were pooled.

Within 2 h after fertilization, diploid and triploid embryos were divided over separate 48-well plates (three embryos per well, approximately 1 cm³ volume) with a mesh bottom and placed in a rearing tank with E3 medium (E2 medium with addition of 10⁻⁵% Methylene Blue). During development, these rearing tanks were kept in three water baths at constant temperatures of 23.5, 26.5 and 29.5°C (±0.5°C), each water bath holding one rearing tank with diploids and one with triploids. The E3 medium in the rearing tanks was constantly aerated to ensure full oxygen saturation. Larvae were reared until metabolic rate was measured, which was on the fifth day post-fertilization for larvae reared at 26.5°C. For larvae reared at 23.5 and 29.5°C, measurements were conducted 1 day later and earlier, respectively. This ensured that these larvae had reached the same developmental stage in physiological time, when their yolk sack was almost fully resorbed and larvae were able to produce a swift escape response. Survival and hatching rates were scored two times a day, and dead embryos or larvae were removed. A mixed effects model showed that average survival rate at 5 days post-fertilization for eight rounds of fertilization did not depend on rearing temperature (P=0.6611), but differed between diploids (33.8% mortality) and triploids (48.4% mortality) (P<0.0001).

A full factorial design was employed in this study combining two ploidy levels (diploid and triploid), three rearing temperatures (23.5, 26.5 and 29.5°C) and three test temperatures during metabolic rate measurements (23.5, 26.5 and 29.5°C). This allowed us to separate between developmental plasticity and acute thermal effects. For any given combination of rearing and test temperature (nine conditions in total), larvae originated from two different rounds of fertilization. Before each set of metabolic rate measurements, all experimental larvae were photographed with a dissection microscope (Leica MZ FLIII; Leica Microsystems, Germany) to assess body length in millimeters. To avoid effects of handling stress from photography during the metabolic rate experiments, the pictures were taken about 3 h before each respirometry trial. The length measurements were made using the segmented line tool of the Fiji image processing package (ImageJ open-source software: https://imagej.net/ImageJ).

This study was conducted at the Institute of Water and Wetland Research (IWWR) of the Radboud University, Nijmegen (Netherlands), in accordance with the Dutch Animals Act (https://wetten.overheid.nl/BWBR0003081/2019-01-01), the European guidelines for animal experiments (Directive 2010/63/EU; https://eur-lex.europa.eu) and institutional regulations. Because the experiments were performed with larvae up to a developmental stage of 5 days post-fertilization, a stage where larvae are not yet dependent on external feeding, no ethical approval was required.

Oxygen consumption was measured in a closed respirometry system using a 24-well glass microplate equipped with oxygen sensor spots glued onto the bottom of 200 µl wells (Loligo Systems, Viborg, Denmark) integrated with a 24-channel fluorescence-based oxygen reading device (SDR SensorDish Reader; PreSens, Regensburg, Germany). The sensor spots measure the partial pressure of oxygen (in kPa), which we combined with the temperature-dependent solubility to calculate the oxygen concentrations at different temperatures. Depletion of oxygen over time was used to calculate oxygen consumption rates per individual (in nmoles O₂ per hour) as well as the critical value in P_O2 (in kPa).

In each run, diploid and triploid larvae without morphological abnormalities were transferred into the 24-well microplate with E2 medium, and these wells were then sealed using an adhesive optical PCR sealing film (Microseal ‘B’ PCR Plate Sealing Film; Bio-Rad, Hercules, CA, USA), making sure to avoid air bubbles inside the wells. In most cases, 10 diploids and 10 triploids were tested in one run, leaving four wells of our 24-well microplate empty, which were used as a control. The microplate with animals was placed in a flow-through water bath connected to a cooling/heating circulating bath (Grant LT ecocool 150) which allowed for temperature stabilization at 23.5, 26.5 and 29.5°C (±0.1°C), respectively. Oxygen concentrations inside the wells were recorded every 30 s with automatic temperature and pressure correction using MicroResp version 1.0.4 software (https://www.loligosystems.com/). All measurements on metabolic rate were carried out in darkness due to sensitivity of the sensor spots to ultraviolet light. Spontaneous activity of larvae was possible during measurements, so their metabolic rate in normoxia was defined as the routine metabolic rate. In each run, at least four randomly selected wells with E2 medium but without animals were used the assess background respiration. The entire trial lasted at least 16 h (between 16:00 and 08:00 h), during which all larvae completely depleted the available oxygen from the wells (Fig. S1). After each experimental run the dead larvae were removed from the wells. To prevent accumulation of bacteria in the wells, the microplate was bleached using a 35% H₂O₂ solution and rinsed with demineralized water. We conducted a total of 24 runs (8 rounds×3 temperatures), measuring oxygen levels in the wells until they were fully depleted in 422 larvae (208 diploids and 214 triploids). The first 20 min of each run was discarded from further analysis as the temperature equilibrated during this period and the larvae adjusted to being transferred to the microplate. Rates of oxygen consumption were calculated from the declines in oxygen levels over time as the slope of a linear regression using a moving time window of 5 min. We also evaluated in separate trials the extent of oxygen ingress for each test temperature by incubating the microplate sensors under hypoxic conditions and measuring how fast oxygen levels increased over time. The larval oxygen consumption rates were adjusted for both background respiration and oxygen ingress although both influences were negligible compared with the respiration rates of the larvae (e.g. background respiration rates as assessed with blanks never exceed 3% of the measured respiration rates in larvae).

Triploidy induction was verified by measuring the amount of DNA in cell nuclei of homogenized larvae using flow cytometry. This procedure was performed according to van de Pol et al. (2020) with minor changes. For each round of fertilization, triploidy induction efficiency was calculated from three pooled samples (one for each of the three rearing temperatures) of cold shocked larvae with one internal diploid control. Larvae without morphological abnormalities were selected for the metabolic rate measurements. The total number of larvae measured for each round ranged between 20 and 30, depending on how many high-quality larvae were available. The amount of lysis buffer added was adjusted to the number of larvae in the pool. The three pooled samples for each batch were processed simultaneously, by homogenizing the larvae and consequently staining the DNA with Propidium Iodide (Sigma-Aldrich). All samples were analysed with a Beckman Coulter FC500 5-color flow cytometer, and triploidy induction efficiency was calculated using the R package ‘flowPloidy’ (Smith et al., 2018). For each round of fertilization, we obtained the following percentages of triploidy induction efficiency: (1) 95.8, (2) 96.3, (3) 100, (4) 63.3, (5) 100, (6) 100, (7) 100 and (8) 100.

All statistical analyses were conducted in R version 3.5.3 (https://cran.r-project.org/). We assessed the congruence between the P_crit derived from the piece-wise linear regressions and the P₅₀ derived from the Michaelis–Menten equation (under h=1, h=2 and h=3). This analysis showed that both approaches yielded similar values when h=2. P_crit values were sometimes over-estimates when initial oxygen consumption rates were very high. At lower h values (h=1), P₅₀ values tended to be under-estimated, whereas higher h values (h=3) resulted in over-estimates. Estimates for the Ṁ_O2,routine under normoxia varied in tandem with P₅₀ values, being higher for the higher P₅₀values associated with h=1 and lower for the lower P₅₀ values associated with h=3. Estimates for the Ṁ_O2,routine based on the Michaelis–Menten equation under h=2 also corresponded to the mean oxygen consumption rate at oxygen levels above 60% saturation (i.e. well above the oxygen level below which oxygen reduces respiration rates) (Fig. 1).

We performed a linear mixed effects analysis of the relationship between oxygen consumption rates (Ṁ_O2) of diploid and triploid larvae and different rearing and test temperatures using R package ‘lme4’ (Bates et al., 2015). Both rearing and test temperatures were treated as continuous variables to describe the relationship between temperature and Ṁ_O2. To account for differences in the efficiency of triploidy induction, ploidy level for each individual was coded as the induction efficiency obtained for a batch of that individual. The induction efficiency ranged between 63.3 and 100% across batches (see also above). Diploid larvae were encoded as having a 0% efficiency. As random effect, we used round of fertilization. In preliminary analyses we examined the larvae body length among experimental treatments. Body length differed with rearing temperature (F=27.650, P<0.0001) but not with ploidy level (F=0.155, P=0.6937). Body length was highest at 26.5°C, and lowest at 29.5°C (Fig. S2). Although the smaller length at high temperatures is consistent with the temperature–size rule, the difference was very small and temperature explained 9.7% of the variation in measured body size. We also tested for an effect of body mass (estimated as length cubed) on Ṁ_O2,routine and P₅₀. We found no consistent effect of body mass. Instead, body mass appeared as a significant factor in diploids and effects of body mass were especially pronounced in larvae reared at 29.5°C. While a positive relationship between Ṁ_O2,routine and body mass was observed for diploid larvae, no such relationship was found for triploids. Given a possible body size influence on the modelled relationships on Ṁ_O2,routine, we included body mass as a covariate. We found no qualitative differences when comparing the model on these Ṁ_O2,routine values corrected for body mass with a model based on uncorrected values and which did not include body length as a factor (Tables S1 versus S2). Hence, we chose to present the model based on uncorrected values in our paper. In our analysis of the P₅₀ values we similarly ran preliminary analyses but here we did not find any influence of body mass on the P₅₀ value. All tests were based on the restricted maximum likelihood (REML) approach. The Akaike information criterion (AICc) and its variants (ΔAICc, AICcWt) were used to find best-fitting model for observed data.

A comparison of two models determining the critical value of P_O2 during oxygen consumption measurements showed that P₅₀ (derived from the Michaelis–Menten equation) correlates best to P_crit (breakpoint in piece-wise linear regression) under h=2 (Hill number) (Fig. 1B). Similarly, the fitted Ṁ_O2,routine based on the Michaelis–Menten equation under h=2 also corresponded best to the mean oxygen consumption rate displayed by the larvae at oxygen levels above 60% saturation (Fig. 1E). This indicates that the Ṁ_O2 versus P_O2 profile in zebrafish larvae approximates a sigmoidal relationship. As we found both methodological approaches highly comparable, in further analyses we used the constants (P₅₀ and Ṁ_O2,routine) derived from the Michaelis–Menten equation with h=2 for each of the 422 larvae.

As expected, there was a general increase in Ṁ_O2,routine with test temperature for both diploid and triploid larvae. Nevertheless, diploid larvae revealed a different response to test temperature, showing a larger increase in metabolic rate at 29.5°C and a larger decrease at 23.5°C compared with triploids. This difference between ploidies in their thermal response of Ṁ_O2,routine increased with rearing temperature and was most pronounced in larvae reared at 29.5°C (three-way interaction between ploidy, test temperature and rearing temperature: P=0.0131; Table 1; Fig. 2). The thermal sensitivity of oxygen consumption rates expressed as activation energy (E_a) or Q₁₀ value (Table 2) decreased with rearing temperature in triploids, indicating an increasingly lower thermal sensitivity. In contrast, thermal sensitivity in diploids increased with rearing temperature (Table 2).

We found opposite effects of rearing and test temperature: values of P₅₀ increased significantly with test temperature (P<0.0001; Table 3; Fig. 3), while there was a trend for them to decrease with rearing temperature (P=0.0644). As expected, P₅₀ values were significantly higher for the triploids, but the difference was small: on average, triploids had P₅₀ values which were 1.3 kPa greater (Table S3; Fig. 3). We did not find significant interactions between ploidy level and temperature, suggesting that effects of rearing and testing temperature on P₅₀ values were similar for triploids and diploids.

In this study we compared the thermal sensitivity of metabolism between diploid and triploid zebrafish larvae. We compared both the metabolic rate under normoxia as well as the critical values of P_O2 (P_crit) below which oxygen consumption becomes dependent upon the ambient oxygen partial pressure. A common approach is to establish the P_crit by means of fitting two lines to the Ṁ_O2 versus P_O2 relationships using a piece-wise linear regression model (Marshall et al., 2013; Reemeyer and Rees, 2019). Wood (2018) recently recommended fitting a Michaelis–Menten equation to the Ṁ_O2 versus P_O2 profile. We compared both the Michaelis–Menten approach and the piece-wise regression approach to delineate P_crit. Both approaches were highly comparable when we used a modified Michaelis–Menten equation with a Hill number of 2, which approximates a sigmoidal relationship for Ṁ_O2 versus P_O2. Hence, when one uses a Michaelis–Menten equation to fit a curve to the entire Ṁ_O2 versus P_O2 profile, we recommend checking whether a hyperbolic or sigmoidal relationship is more appropriate, and use the corresponding Hill number.

We found that the routine metabolic rate of diploid zebrafish larvae reared under the highest temperature (29.5°C) showed a different response to the test temperature than that of triploid larvae. At a test temperature of 29.5°C, the metabolic rate of diploids was higher than that of triploids, but at 23.5°C we found the opposite, with diploids having lower metabolic rates than triploids. Previous studies investigating metabolic rates at relatively high temperatures also reported higher aerobic metabolism in diploids compared with polyploids (Stillwell and Benfey, 1996; Atkins and Benfey, 2008; Maciak et al., 2011; Hermaniuk et al., 2017). The lower metabolic rate of triploids could indicate that for them, oxygen becomes limiting at high temperatures. With increasing temperature, maximum rates of oxygen diffusion can be slightly enhanced as the diffusion coefficient of oxygen in water increases with temperature more so than the concomitant decrease in the solubility of oxygen in water (Dejours, 1981; Verberk et al., 2011). However, because oxygen consumption increases more than rates of oxygen diffusion, boundary layers may become oxygen depleted (Verberk and Atkinson, 2013), and consequently aquatic organisms may nevertheless experience a shortage of oxygen in their environment (Pörtner, 2010; Verberk et al., 2016b; Kielland et al., 2019). Although this situation would be similar for triploids and diploids, diploids may be able to maintain higher metabolic rates for several reasons. First, being composed of more, but smaller cells, they also have more, smaller erythrocytes which may confer them with higher oxygen transport capacities. Second, being composed of smaller cells results in shorter diffusion distances which may help individual cells to take up more oxygen and thus generate a larger pressure gradient between the tissues and their environment, promoting diffusive oxygen uptake (Woods, 1999; Atkinson et al., 2006; Czarnoleski et al., 2013). While such effects are probably small, our study showed that hypoxia tolerance in zebrafish was slightly different between triploids and diploids, with diploids having lower P₅₀ values. We also found that P₅₀ values increased with test temperature. In their meta-analysis, Rogers et al. (2016) found a similar effect of temperature and these findings fit well with the idea of a mismatch between oxygen supply and oxygen demand in warmer waters. A recent study on water fleas (Daphnia magna) with a similar experimental set-up to ours likewise found a higher P_crit in warmer water and concluded that phenotypic plasticity in oxygen supply was insufficient to fully compensate for the increase in oxygen demand (Kielland et al., 2019). Their study focused on the combined effect of developmental and test temperature, but in our study design these could be disentangled. While our results indicate that test temperature had a larger effect (increasing P₅₀ and Ṁ_O2,routine), developmental temperature modulated the effects of test temperature (Ṁ_O2,routine) or had opposite effects (P₅₀). The opposing effects of test temperature and development temperature on P₅₀ suggest that there was some developmental acclimation, such that warm-reared larvae were more tolerant to hypoxia. The between-ploidy difference in hypoxia tolerance was not prominent although, as expected, triploids displayed a lower P₅₀ in oxygen-limited conditions. Hypoxia tolerance has recently been argued to be linked to maximum metabolic rate and aerobic scope (Seibel and Deutsch, 2020). A lower hypoxia tolerance could thus indicate a lower athleticism, which could explain the reduced swimming performance in triploids compared with diploids that has been reported previously (van de Pol et al., 2020). Along the same lines, hypoxia decreased thermal performance in triploid Atlantic salmon, manifested in negative effects on survival, growth and feed intake, as well as changes in swimming behavior (Hansen et al., 2015; Sambraus et al., 2017). Other studies, similarly to our results, revealed only small effects of ploidy on hypoxia tolerance above their thermal optima (Benfey and Devlin, 2018; Sambraus et al., 2018). Nevertheless, these results confirm our hypothesis of greater sensitivity to hypoxia in triploids. Consistent with this notion, previous work on triploid frogs has shown that triploids may compensate and increase their capacity to deliver oxygen by developing larger hearts (Hermaniuk et al., 2017). Other work has shown increased gill ventilation rates in triploid trouts (Lahnsteiner et al., 2019), although this would not be applicable to our zebrafish larvae as they do not yet use their gills for gas exchange (Rombough, 2002).

While oxygen limitation mediated by cell size could explain the higher metabolism of diploids in warmer temperatures, it does not explain why warm-reared diploids had lower metabolic rates at the lower test temperature of 23.5°C. Atkins and Benfey (2008) also demonstrated that the routine metabolic rate in triploid fish (Salmo salar and Salvelinus fontinalis) is higher at low temperatures and lower at high temperatures when compared with diploids. A higher metabolic rate can be interpreted both as a burden (high maintenance costs) and as a boon (high energy budget to fuel performance) (see Verberk et al., 2016a; Clarke et al., 2003). Given that we measured routine metabolic rate, rather than standard metabolic rate, and that energetic costs related to maintenance of ionic gradients across membranes are expected to be greater in diploids than in triploids, we tentatively interpret the higher metabolic rate of triploids in the cold as beneficial, reflecting a larger energy budget. Similarly, in a comparison between diploid and polyploid cladocerans (Dufresne and Hebert, 1998; Van Geest et al., 2010) and amphibian larvae (Hermaniuk et al., 2016), faster development was observed in polyploids, but only in the colder test temperatures. Thus, the difference in thermal sensitivity between different ploidies goes beyond oxygen limitation in the warm: there appears to be a genuine advantage to polyploidy in the cold. What could this be? Here, we offer two potential mechanisms, related to homeoviscous adaptation of membrane function (Hazel, 1995) and developmental noise (Woods, 2014).

Membrane fluidity is strongly tied to the capacity for transport of oxygen and substrates (Subczynski et al., 1989; Govers-Riemslag et al., 1992; Sidell, 1998; Möller et al., 2016), the activity of many membrane-associated enzymes (Houslay and Gordon, 1982; Ushio and Watabe, 1993), and the permeability of the membrane to cations and water (Singer, 1981). Temperature affects membrane fluidity, but organisms maintain membrane fluidity within some range, by remodeling their cell membranes (e.g. changes in lipid head group composition, acyl chain length and saturation, as well as changes in the cholesterol content of membranes; see Hazel, 1995). Interestingly, ectotherms inhabiting environments characterized by large and rapid thermal fluctuations possess the capacity to rapidly modify (within hours) the lipid composition of their cellular membranes in response to diurnal variations in water temperature (Carey and Hazel, 1989; Williams and Somero, 1996). In our experiments, larvae took 3–5 h to deplete oxygen down to their P₅₀, which should allow for at least some membrane remodeling. While this time period was shorter for animals tested at higher temperatures (see Fig. S3), the higher activity in warmer conditions compensates for this; the rate at which they depleted oxygen down to their P₅₀ had a thermal sensitivity that is comparable to most physiological rates (Q₁₀ of approximately 2). Thus, in thermally unstable environments the larger membrane surface area associated with smaller cells could create limits to membrane remodeling (for review see Czarnoleski et al., 2013; Czarnoleski et al., 2015a). This mechanism would put warm-reared diploids at a disadvantage relative to triploids when tested at a lower temperature; their lower membrane fluidity and lower activity of membrane-associated enzymes could explain their lower metabolic rates. In principle, cold-reared diploids tested at a high temperature would be expected to have higher metabolic rates compared with triploids, as diploids will face the opposite problem of not being able to reduce their membrane fluidity as rapidly as triploids. However, we found no difference in metabolic rates between ploidy levels for cold-reared larvae tested at the highest temperature. Perhaps the speed at which animals can remodel their membrane is lower when faced with colder temperatures (requiring an increase in fluidity) than when faced with warmer temperatures (requiring a decrease in fluidity), but this will require further study. At least under relatively cold conditions, we found that triploids are able to maintain a higher metabolic rate compared with diploids. Other studies have likewise found that larger cells are disproportionately more productive compared with smaller cells in tissues with active transcription and translation machinery (Ginzberg et al., 2015). Both a larger volume of cytoplasm and a larger genome may increase expression of specific genes and increase rates of protein synthesis (Marguerat and Bahler, 2012; Hessen et al., 2013; Doyle and Coate, 2019); during exposure to low temperatures, such effects may boost overall cellular activity as well as the production of membrane-associated enzymes that may additionally shorten the time of membrane remodeling and therefore improve metabolic efficiency. A second mechanism also draws on the importance of gene translation and transcription but focuses on developmental noise that arises from stochasticity in developmental pathways whose regulation arising from random interactions of molecules becomes increasingly unpredictable and variable with reduced absolute numbers of molecules (see Woods, 2014). Such stochasticity potentially depresses performance (Blomberg, 2006; Woods, 2014) and increases if the number of molecules that participate in a reaction are lower or if the reaction rates are slower. Thus, having larger cells with higher absolute numbers of molecules can mitigate the effect of slower reaction rates in the cold. Under cold conditions, a large cell size would therefore provide better control of cell-level stochasticity (Verberk et al., 2020), which may also help explain the generally lower thermal sensitivity of the metabolic rate of the triploid larvae observed in our study.

The results of our study show that metabolic consequences of cell size are dependent on temperature. Both developmental temperature and the temperature to which larvae were acutely exposed affected the metabolic response, and differences between ploidy levels were strongest at the highest rearing temperature. At high temperatures, diploids could maintain higher levels of oxygen consumption rates than triploids, which is consistent with the hypothesis that animals composed of larger cells are more susceptible to oxygen limitation in warm waters. This also fits with our finding of a lower hypoxia tolerance in triploids. In contrast, at low temperatures, triploids could maintain higher levels of oxygen consumption rates, which could be related to membrane remodeling and/or molecular stochastic noise. Our results suggest that being composed of larger cells provides metabolic advantages in cold water, while being composed of smaller cells provides metabolic advantages in warm water.

We thank J. Boerrigter for his assistance with laboratory work. We are also thankful to T. Spanings for husbandry of the zebrafish in our fish facility.