The temperature-size rule emerges from ontogenetic differences between growth and development rates


*Correspondence author. E-mail:


1. The temperature-size rule (TSR) is a widespread phenomenon, which describes the phenotypic plastic response of species’ size to temperature: individuals reared at colder temperatures mature as larger adults than at warmer temperatures.

2. The TSR is driven by an unequal thermal response of growth and development rates. However, we currently lack an understanding of how these rates change through ontogeny and their decoupling. Further, we do not know how this decoupling varies across generations.

3. Using the brine shrimp Artemia franciscana as a model, we examine growth and development rates through ontogeny at different temperatures across two generations.

4. The slopes of natural-logged weight-specific growth rates against temperature are steeper in earlier than later larval stages, indicating their greater temperature dependence, whereas development rates maintain the same temperature dependence across life stages. An inverse TSR is generated in early larval stages; the typical TSR (smaller size at warmer temperatures) is only established later in ontogeny.

5. Phase-specific temperature dependence of growth and development rates is not significantly different across the 1st and 2nd generation, suggesting the TSR is primarily a within-generation outcome.

6. Ontogenetic size responses in Artemia are compared to other crustacean species to identify patterns within this subphylum. Data for a range of crustaceans follow the same ontogenetic pattern: early larval stages show an inverse or no TSR, with TSR being only established in later stages. Adults often, but not always, show the greatest response.


The temperature-size rule (TSR) describes the impact of temperature on intraspecific size of ectotherms: individuals reared at colder temperatures reach maturity at a larger size than when reared at warmer temperatures (Atkinson 1994). This phenotypically plastic response has been observed in >83% of ectothermic species studied, including plants, bacteria, protists, invertebrates and vertebrates (Atkinson 1994). Adult size is in effect a product of growth rate (increase in weight per time) and development rate (increase in life stage per time), and the TSR signals that these rates must be decoupled (Sibly & Atkinson 1994; Van Der Have & De Jong 1996; Kingsolver & Huey 2008). Much of the focus of the TSR has been on explaining the ultimate reason for size change (Walters & Hassall 2006; Kingsolver & Huey 2008; Arendt 2011). However, to understand why size changes, we first need a clearer understanding of how size changes are generated. In particular, there remains a lack of data available on growth and development rate decoupling through ontogeny. Decoupling of these rates is most often inferred from differences between size at some final life stage and the time taken to reach this mature stage (Partridge et al. 1994; Blanckenhorn 2000; Stillwell & Fox 2005), without considering the ontogenetic timing of these size changes at a higher resolution. Previous work on ectotherms has found temperature-acclimated adults show a greater temperature-size response than do acclimated progeny, such as eggs (Forster, Hirst & Atkinson 2011a). A recent analysis of marine pelagic copepod data has shown development rates to have greater temperature dependence across all life history stages than growth rates (Forster, Hirst & Woodward 2011b). Further, there was weak support for growth rate being more temperature dependent at smaller, early life stages than later stages (i.e. slopes of natural-logged growth rates against temperature were steeper at early life stages), suggesting a size-dependent or ontogenetic component in these crustaceans. Similarly, analyses of the interaction between growth and development during ontogeny for the tobacco hornworm moth Manduca sexta have shown the TSR to emerge only during later larval stages (Davidowitz, D’amico & Nijhout 2004; Davidowitz & Nijhout 2004; Nijhout, Davidowitz & Roff 2006; Diamond & Kingsolver 2010). However, we still lack experimental data that focus on the timing of ontogenetic temperature-size changes at a high resolution. Mortality rates are significantly higher during early larval stages of many species (McConaugha 1992; Cornell & Hawkins 1995; Hirst & Kiørboe 2002): this is an intense period of selection with important fitness consequences for species. It is therefore essential that we gain a better understanding of how growth and development rates, along with size, change during ontogeny and the degree to which they are temperature dependent. In this study, we test these ideas in detail using Artemia franciscana (Kellogg) as a model crustacean. We examine size, growth and development rates at a high temporal resolution through ontogeny.

The majority of studies investigating the TSR through growth and development rates simply expose eggs or early larval stages to novel temperatures and observe the effect on adult size (Smith-Gill & Berven 1979; Partridge et al. 1994; De Jong 2010). However, organisms are capable of acclimating to different temperature regimes; for example, the metabolic rate of fishes transferred from a low to high temperature initially increases substantially but then reduces towards an acclimatory rate (Johnston & Dunn 1987). A previous meta-analysis examining the TSR in organisms acclimated to their thermal environments found body size of egg and early larval stages to be less temperature dependent than adult stages (Forster, Hirst & Atkinson 2011a). Further, data on size changes through ontogeny in two copepod species, acclimated to different temperatures for multiple generations, showed size to effectively ‘reset’ at the beginning of each generation, i.e. eggs showed no temperature-size response yet adult stages followed the TSR (see Forster, Hirst & Atkinson 2011a). This suggests that the drivers of the TSR, growth and development rates remain decoupled, although the rates themselves have not been directly measured across multiple generations. This study aims to test this directly by measuring these rates over two generations in A. franciscana and to determine whether the degree to which these rates are decoupled changes.

Whether patterns in temperature-size responses are similar across closely related species needs to be addressed. Artemia franciscana are within the Anostracans; Anostracans are the closest example to primitive crustacean morphology and provide the nearest case of the presumed ancestral state (Browne, Sorgeloos & Trotman 1991). It may be that other crustaceans exhibit similar size changes through ontogeny as Artemia. We, therefore, need to compare the ontogenetic temperature-size data across crustaceans more broadly.

We address the following questions: (i) Do the temperature dependence of growth and development rates in A. franciscana vary through ontogeny, and how does this impact the temperature-size response? (ii) Does the impact of temperature on these rates differ between the first and second generation? and (iii) Is the ontogenetic basis of the temperature-size response similar across crustaceans?

Materials and methods

Batch cultures of A. franciscana were established using decapsulated cysts. These cysts had been collected from the Great Salt Lake (Utah, USA), disinfected and decapsulated (provided in this state by the company Waterlife). A minimum of ∼300 cysts were placed in 1-L beakers containing 900 mL GF/F filtered sea water with a salinity of 30 at a range of constant temperatures (20, 22·5, 25, 27·5, 30, 32·5 and 35 °C). Air stones were used in each beaker to ensure the water was sufficiently aerated. Cultures were maintained at fixed temperatures using Grant SUB Aqua 26 water baths (held within ± 0·2 °C of control temperature). Upon hatching, stage 1 nauplii were transferred from the batch cultures into a minimum of two separate replicates at each temperature (with n = 50 per replicate), initiating the first generation of the experiment. As cultures did not survive for long at 35 °C, these were not continued. After hatching, nauplii were provided ad libitum with the algae Arthrospira plantensis. Replicates were fed a minimum of 10 mL of saturated A. plantensis solution (where no more algae could be suspended in solution) every day such that a green coloration was visible and was maintained in the cultures at all times. All individual A. franciscana were staged, and their total body length was measured on a daily basis using a light microscope. Post-embryonic larvae were staged in a similar manner to Weisz (1946), using number of segments and limb mobility (see Appendix S1 in Supporting Information), with development being divided into 17 stages. Beyond stage 17, measurements of development ceased, as without the addition of further segments or appendages, accurate assessment of stage could not be achieved. Once animals reached the adult stage in the cultures, we inspected for the presence of their nauplii on a daily basis. A total of 50 nauplii were removed from each temperature replicate with a Pasteur pipette and placed into new beakers; this was the initiation of the second generation. In a small number of cases (3/12 cultures), there were insufficient nauplii from the experimental culture; in these cases, we supplemented with nauplii from other batch cultured adults, maintained at the same temperature as the replicate in question and at saturated food conditions. Development and growth experiments were started for the second-generation replicates, the time at first appearance of nauplii defining = 0. Body lengths and stage were determined daily on live individuals, as conducted for the first generation, and the same feeding regime was used also. Across both generations, water in all replicates was changed weekly; between these changes, any water loss via evaporation was replaced using distilled water to maintain salinity.

Calculating Growth and Development Rate

Individual dry weights were estimated from length measurements using equations calculated from Reeve (1963, see Appendix S1), and average weight calculated for each replicate at each observation point. Weight-specific growth rates (g, day−1) were determined as:

image(eqn 1)

where DWt = dry weight at time t, DW0 = dry weight at previous observation point and = time between observations (days).

Although measurements were taken daily, as in some instances >3 stages can pass in a single day, growth rates were calculated during ontogeny by combining data for 3 stages together: specifically stages 3–5, 6–8, 9–11, 12–14 and 15–17. We term each of these ‘phases’. As the two initial stages have very rapid development (<24 h for transition through both stages at most temperatures), we did not include these.

Development times were calculated as median stage-specific development times, i.e. from initial nauplii introduction to the point at which 50% of individuals reached stage 2, then from this point to the point at which 50% of individuals reached stage 3, etc. These median development times were calculated (for each replicate) from stage frequency data, following the methods of Campbell et al. (2001). Median development times were calculated for each stage from 1 to 17 inclusively. These development rates were then grouped into the phases, to allow direct comparison with growth rates.

Model Fitting

We used allometric models to describe the effect of temperature on both growth and development rates. These models have previously been shown to be the most appropriate for other crustaceans, i.e. marine copepods (Forster, Hirst & Woodward 2011b) and planktonic larval species (O’connor et al. 2007). Further, applying exponential and Arrhenius functions to the data and comparing their fit using an information-theoretic approach (Akaike Information Criteria, Akaike 1974; Burnham & Anderson 2002) did not produce a better fit (see Appendix S2 in Supporting Information). Growth rates (g, day−1) and development rates (D, day−1) were modelled as functions of temperature (T, °C) using the allometric function:

image(eqn 2)

Where = rate, a and b are constants and ε is the error term. We centred the data for both growth and development rates around the mid-temperature of the experiment (25 °C). This improves the interpretability of each model (e.g. the intercept becomes the rate at 25 °C) and reduces the correlation between parameters (O’connor et al. 2007), thus lnT became [ln− ln(25)], we term this centred temperature as TC. Growth rates and development rates were available for different phases (e.g. stages 3–5, 6–8) and for 1st and 2nd generations. A linear mixed effect model was used to account for these differences in phase and generation number, incorporating phase as a random effect and generation number as a fixed effect (see Appendix S2). We applied multiple variations of the models in eqn 2 for growth and development, which allowed parameters a and b to vary with phase and/or generation number. Using a log-likelihood ratio test, we then discerned which variation of these equations best modelled data for each rate (see Appendix S2 for further details). These procedures were followed to determine whether ontogeny impacts the intercept of growth and development (lna, eqn 2) and/or the slopes (b, eqn 2). Similarly, these procedures were used to determine whether the two generations show differences, both in their intercepts and in their slopes.

Comparison of Weight within Stage

To estimate the temperature effect on weight at stage and to discern the importance of generation number on organism size, we described the effect of temperature on dry weights for each stage using a linear mixed effects model and an exponential equation form. A previous analysis of multicellular species data for weight vs. temperature has shown this equation form to best describe temperature-size data (Forster, Hirst & Atkinson 2011a). Further, we confirmed here, using Akaike weights, that other model forms (allometric and Arrhenius) did not provide a better fit to the data. We followed the same method used for modelling growth and development rates, except using individual stages (3, 4, 5 etc.), rather than phases. The equation applied was of the form:

image(eqn 3)

where DW = dry weight (μg) and TC = the experimental temperature (°C) minus 25. Having calculated the stage-specific slopes for all larval stages of A. franciscana, we converted these slopes to % change in weight per °C using the formula (exp(slope) − 1)*100. We collected the eggs produced as the second generation at each temperature (>50 eggs for each temperature) and measured their diameter (E). Eggs were near-spherical and volume calculated as 4/3*π(E/2)3. The slope of egg volume against temperature was calculated using generalized least squares regression and eqn 3.

Comparison of Ontogenetic Temperature-Size Response for Crustaceans

To compare the temperature-size response of A. franciscana with other species, we searched the wider literature for data on weight at stage vs. temperature in other crustaceans, measured through ontogeny. We searched the ISI Web of Knowledge using the search terms ‘larva* AND temperature AND (mass OR size OR weight)’ along with previous data collected for copepod species (Forster, Hirst & Woodward 2011b) and other multicellular organisms (Forster, Hirst & Atkinson 2011a). We included data where size had been measured for ≥3 temperatures at ≥2 larval stages. We applied a linear mixed effects model to each species in turn, using eqn 3 and following the same methods as applied to the A. franciscana data.


Artemia were successfully reared over two generations at fixed temperatures ranging between 22·5 and 32·5 °C. Although those reared at 20 °C reached adulthood and reproduced at the end of the first generation, the 2nd generation did not reach maturity. Growth rate changed during ontogeny, with two distinct trajectories: the first being early larval growth during the formation of thoracic segments (stages 1–11) and the second trajectory during the formation of abdominal segments (stages 12–17); growth was distinctly faster during this second period (as demonstrated by the steeper slope of ln weight vs. time, Fig. 1a). Our phases were defined so as not to combine stages across this division. Such marked shifts through ontogeny were not present in development rates (Fig. 1b).

Figure 1.

 (a) Artemia franciscana progression of ln(dry weight) over time (days) at different temperatures. Early larval stages (where thoracic segments are added) show a shallower slope than later larval stages (abdominal growth), i.e. growth rate is lower in the earlier stages. Data are from individual replicates during the 1st generation. (b) A. franciscana increase in development stage (1–17) with time. Data represent the median development times for the same replicate from the 1st generation (as in a). Development rates do not show two distinct trajectories during ontogeny as does growth.

Model Fitting

The inclusion of generation number did not improve the fit for either growth or development, and therefore this identifier was removed from the linear mixed effects models. The lack of improved fit suggests no significant difference between 1st and 2nd generation for growth or development rates. For growth rates, the best fit model required the inclusion of phase as a random parameter within both the slope and intercept term. This suggests that growth rates vary between phases (intercept), i.e. some have faster weight-specific growth than others, but also that different phases have a different temperature dependence (i.e. slopes, see Fig. 2). By contrast, the fit of the development model was not improved with the incorporation of phase as a random effect in the slope term, the best fit model required phase to be incorporated within the intercept term only (Fig. 3). This suggests that some phases develop more quickly than others, but they have the same temperature dependence regardless of phase (i.e. slopes are similar).

Figure 2.

Artemia franciscana weight-specific growth rates vs. temperature (°C) across 5 ontogenetic phases (stages 3–5, 6–8, 9–11, 12–14, 15–17). Triangles are mean values for the 1st generation, and squares are means for the 2nd generation. Panel 6 shows the best fit models (i.e. allometric, eqn 2) for each of the phases. All regressions are fitted through data from both generations combined. Regression equations are provided in each panel, where = growth rate (day−1) and TC is temperature T (°C) centred around 25 °C (i.e. TC = [lnT − ln(25)]. Note the log10–log10 scale, error bars represent 95% confidence intervals.

Figure 3.

Artemia franciscana stage-specific development rates vs. temperature (°C). Rates are presented as averages across the 5 phases (stages 3–5, 6–8, 9–11, 12–14, 15–17). Triangles are mean values for the 1st generation, and squares are means for the 2nd generation. Panel 6 shows the best fit models (allometric model, eqn 2) for each of the phases. All regressions are fitted through data from both generations combined. Regression equations are provided in each panel, where = development rate (day−1) and TC is temperature T (°C) centred around 25 °C (i.e. TC = [lnT − ln(25)]. Note the log10–log10 scale, error bars represent 95% confidence intervals.

Comparison of Weight within Stage

We compared the temperature-size response across different stages of Artemia using the linear mixed effects model applied to eqn 3. This revealed a temperature dependence of stage-specific weight, but no significant improvement of the model with the addition of generation number. This was confirmed in the results of the models applied to growth and development: generation number did not appreciably change the thermal response of these rates. To demonstrate this, we present the stage-specific best fit models in Fig. 4, but including the generation number term in the model, to show the slopes of these models are near-identical for the 1st and 2nd generation. The lowest temperature (20 °C) was excluded from this analysis as the reared individuals failed to reach maturity in the second generation. Furthermore, size decreased at this temperature in later stages (Fig. 5); thus, it was excluded to maintain the simplicity of the exponential model (eqn 3). The stage-specific temperature-dependent slopes revealed an inverse temperature-size response at early larval stages (Stages 1–7, Fig. 4). The temperature-size response became flat during intermediate larval stages (Stages 8–10), before establishing the more typical TSR at stages 11 onwards (Fig. 4), with a significant decrease in weight with increasing temperature for stage 12 onwards. We compared final larval size data to egg size in Fig. 5. Final larval size (excluding 20 °C) showed a significant negative slope of dry weight vs. temperature (−2·96%°C−1, 95% CIs ± 0·62%), whereas the slope of egg volume vs. temperature was not significantly different from zero (0·08%°C−1, 95% CIs ± 0·40%).

Figure 4.

Artemia franciscana larval stage dry weights (μg, log10y-axis) vs. temperature (°C). Regression lines represent the best fit to the data using a linear mixed effect model and eqn 3. Early larval stages (1–8) show an inverse TSR, with TSR being established from stage 12 onwards (asterisks denote negative slopes, which are significantly different from zero). These slope values and confidence intervals are shown in the first panel of Figure 6. Best fit lines are given separately for the 1st and 2nd generation. There was no significant effect of generation on dry weight values, as including this within the mixed effects model did not improve the fit.

Figure 5.

 (a) Artemia franciscana dry weight (μg, log10y-axis) vs. temperature (°C) for stage 17, the final larval stage. Triangles represent the 1st generation, and squares represent the 2nd generation. The best fit model (linear mixed effects model, eqn 3) was calculated and shows the TSR equivalent to −2·96% dry weight °C−1. (b) A. franciscana egg volume (mm3, log10y-axis) vs. temperature (°C). The best fit line (generalized least squares regression model, eqn 3) shows no TSR, with a change in volume equivalent to 0·08%°C−1.

Comparison of Ontogenetic Temperature-Size Response for Crustaceans

Along with data for A. franciscana, we collected temperature-size data for 10 other crustacean species including 7 copepod, 2 crab and 1 daphnid species. The % change in weight °C−1 is plotted for each species through ontogeny in Fig. 6. This revealed a general pattern of declining slopes with increasing stage. These patterns were very similar in all 11 crustacean species. Early larval stages show an inverse or no TSR, whereas later stages show a strong TSR, with weight changes varying between −1 and −4·5% per °C (Fig. 6). Data for 7 of the 8 species where very early larval or egg sizes were available (Stages 0–2, Fig. 6) show no significant change in progeny size with temperature. In many cases, the adult stage showed the strongest temperature-size response, but this was not always the case, and some showed a reduction in the response into the final stage(s).

Figure 6.

 The temperature-size response through ontogeny in a variety of crustacean species; slopes are expressed as % change in weight per °C for a given stage. Species-specific life stages were assigned arbitrary values from egg (0) to first instar (1) onwards. Therefore, there is no relation, for example, between stage x of a crab species and stage x of a copepod species, but these are comparable within a group (copepod–copepod, crab–crab etc.). Arrows indicate the adult stage. Daphnia pulex data are embryonic stages (development within mother); therefore, adult stage is not indicated. Error bars represent 95% confidence intervals.


There were clear differences between the temperature dependence of growth and development rates in A. franciscana. While the temperature dependence of growth rates decrease with size/life stage (see slopes in Fig. 2), development rates had similar temperature dependence throughout ontogeny (see slopes in Fig. 3). Other development rate data from the crustacean literature support this outcome more widely. Copepods have been shown to maintain the temperature dependence of development during ontogeny; this is in effect the widely observed equiproportional development rule, where specific life stages occupy a fixed proportion of the total life cycle across different temperature regimes (Hart 1990). As earlier larval stages of A. franciscana show a greater temperature dependence of growth rate than later larval stages, this suggests these rates are stage/size dependent (see Fig. 2). Evidence for this in other crustaceans is scarce. Data for the crab Carcinus maenas follow a similar pattern, with growth rates of early zoeal stages being more temperature dependent than later larval stages (Dawirs, Puschel & Schorn 1986). There is also weak support from data for growth rates in marine copepod species (Forster, Hirst & Woodward 2011b); specifically, the slopes of early nauplius logged growth rates vs. temperature were found to be steeper than those of later copepodite stages, although this result was not significant (2 sample t-test, t = 1·94, P = 0·057; Forster, Hirst & Woodward 2011b). A’decrease in the temperature dependence of growth through ontogeny (with increasing stage/size) in A. franciscana has important implications for the TSR. Growth rates are more temperature dependent than development rates in the early larval stages (Figs 2 and 3), resulting in a reverse TSR: body size in early larval stages increases with increasing temperature (Fig. 4). Only during later stages, when the temperature dependence of growth is less than that of development, is the TSR established. Therefore, the appearance of the TSR is not determined solely by the temperature dependence of growth rates changing through ontogeny, but on growth having a lower temperature dependence than development rate in later larval stages. This highlights the importance of following changes in both growth and development rates throughout ontogeny.

Does a mechanism exist to explain why the temperature dependence of growth (the slopes in Fig. 2) decreases with increasing size in Artemia? Previous mechanistic models to explain changes in growth rate with size and temperature have been based on the Von Bertalanffy (1957) growth equation:

image(eqn 4)

where = body weight, k is the coefficient of anabolism, l is the coefficient of catabolism and m and n are exponent parameters. Increasing temperature can alter maximal body size by changing either the coefficients or the exponents. Previous work by Perrin (1995) and Strong & Daborn (1980) has produced two mutually exclusive mechanisms based on the Von Bertalanffy (1957) growth rate model to explain changes in growth rates associated with temperature. Perrin (1995) showed optimal life history to follow the TSR when the temperature dependence of the catabolism coefficient l is greater than that of anabolism k, assuming exponents m and n are constants (0·75 and 1, respectively). Conversely, Strong & Daborn (1980) used data for the isopod Idotea baltica to argue that smaller size is driven by a decrease in m (from approximately 1·0 to 0·7) and increase in n (from approximately 0·7 to 1·0) with increasing temperature, resulting in different allometries of anabolism and catabolism. Our results suggest that neither of these proximate mechanisms are sufficient to explain the change in growth rates in A. franciscana. Perrin’s (1995) model assumes a decelerating rate of weight-specific growth through ontogeny for small size to be optimal at higher temperatures. This is not the case in A. franciscana, where growth rates are faster during the later stages in which abdominal segments are added (Fig. 1), nor is it the case in a large number of species including amphibians, cnidarians, crustaceans, fish, insects, molluscs and reptiles [see review by Angilletta, Steury & Sears (2004)]. Strong & Daborn’s (1980) model implies that the temperature for maximal growth rate decreases with increasing size. We do not find support for this, as growth rate was always at its maximum at the highest temperature (32·5 °C) in A. franciscana. Both interpretations of the Von Bertalanffy (1957) growth equation are therefore inadequate at describing changes in growth rates in A. franciscana. Both coefficient terms and exponents would have to change to accommodate differences in growth rates across different phases and at different temperatures (Kozłowski, Czarnoleski & Danko 2004). The lack of mechanistic explanation provided by the Von Bertalanffy (1957) highlights the problems associated with this model type; indeed, a mechanistic explanation for why the temperature dependence of growth decreases with increasing size remains elusive.

The temperature dependence of growth and development rates for any particular phase of A. franciscana did not change between the first and second generation, with similar ontogenetic patterns in the decoupling of growth and development rates in both (see Fig. 4). Further, generation number did not have a significant effect on body size through ontogeny (i.e. there were no size differences between first and second generation of organisms). We therefore suggest that acclimatory compensation of growth and development rates to novel thermal environments may be extremely rapid. Is this supported by data for other species in the wider literature? Although there is a lack of growth and development rates measurements over multiple generations, we can infer the acclimatory responses of these rates by examining available data for body size. Data for Drosophila melanogaster size vs. temperature showed the effect of generation (1st vs. 2nd generation) to have significant effects on organism size; however, these size changes were extremely small and explained only 0·23% of the variation in body weight found (compared to 82% of variation explained by temperature; Karan et al. 1998). Similarly, small but significant changes have been shown to occur in egg and adult size in D. melanogaster, driven by differences in the parental thermal environment (Crill, Huey & Gilchrist 1996). Fischer et al. (2003) showed that the butterfly Bicyclus anynana lays larger eggs at cooler temperatures, but that the effect of oviposition temperature does not significantly alter size at later larval stages when reared at a common temperature. Data for the yellow dung fly Scatophaga stercoraria showed maternal temperature did not have a significant effect on offspring growth rates (Blanckenhorn 2000). Similarly, data for the hawkmoth M. sexta, where eggs were hatched at different temperatures then reared at a common temperature, showed the hatch temperature to affect initial larval size, but that this disappeared by the fourth instar (Potter, Davidowitz & Woods 2011). These studies, and our own, suggest rapid acclimation of growth and development rates in ectothermic species.

We found size to decline in later stages at the lowest temperature (Fig. 5A). This concave thermal response of adult has previously been found in other ectotherm species, including Drosophila (data for both wing length and mass, David et al. 1997; Karan et al. 1998; Petavy et al. 2001; Ray 1960), aphids (Lamb et al. 1987), aquatic insects (Vannote and Sweeney 1980), leeches (Young and Ironmonger 1982), frogs (Smith-Gill & Berven 1979), copepods (Kimoto, Uye & Onbe 1986; Hansen et al. 2011) and a moth (Davidowitz & Nijhout 2004). This suggests there may be common temperature-size patterns in adult ectotherms, but that these are not simply linear or exponential terms. Indeed, applying empirical relationships between growth and development rates data have previously also resulted in the prediction of this concave shape (Davidowitz, D’amico & Nijhout 2004; Forster, Hirst & Woodward 2011b). We found A. franciscana did not attain adult stage at the lowest temperature in the 2nd generation, thus this lower temperature may be harmful over multiple generations. Low survivability, coupled with long generation times, make rearing ectotherms and obtaining data at lower thermal limits more difficult, which may explain why the majority of studies do not show a concave shape (Kingsolver & Huey 2008). Further, the low survivability associated with cold stress suggests that this aspect of the TSR may not be relevant in the field, as maintaining populations at these lower temperatures over multiple generations was not possible.

Examining the temperature-size response through ontogeny in A. franciscana, we found no relationship in eggs, an inverse TSR in early larval stages and a significant TSR established at stage 12 (Figs 4 and 6). Although the establishment of a significant temperature-size response occurs at the same point as the shifts from slower to more rapid growth (and from thoracic segments being added to abdominal segments, Fig. 1), this appears largely circumstantial: the change from a negative temperature-size response is cumulative, with stage 1 showing the most negative temperature-size response, and this getting less negative with increasing stage, until a significant temperature-size response is established at stage 12. Other crustacean species follow similar patterns. Early larval stages show little or no temperature dependence of their size (and sometimes a reverse TSR), whereas later stages show the more typical TSR, with size declining with increasing temperature. This suggests that the temperature-size relationship is commonly ‘reset’ at the beginning of each generation (Forster, Hirst & Atkinson 2011a). Indeed, the data of Leandro, Tiselius & Queiroga (2006) for Acartia tonsa (see our Fig. 6) show that individuals acclimated for at least two generations to their thermal environment show this same pattern. This lends further support to the theory that crustaceans follow a common pattern of size responses to temperature, with initial size being relatively temperature independent even when organisms are maintained at temperatures for multiple generations.

Although we have restricted our analysis to crustaceans, data from other ectothermic groups have shown egg size to be less temperature dependent than adult size (e.g. see the synthesis of Forster, Hirst & Atkinson 2011a). Further support for the TSR emerging only in later larval stages comes from the insect M. sexta (Davidowitz, D’amico & Nijhout 2004, Davidowitz & Nijhout 2004; Diamond & Kingsolver 2010). It should be noted, however, that some species have significant changes in egg and/or early larval size with temperature (Crill, Huey & Gilchrist 1996, Van Voorhies 1996; Ernsting & Isaaks 1997; Blanckenhorn 2000; Fischer, Brakefield & Zwaan 2003; Hassall et al. 2006; Steigenga & Fischer 2007), with size changes following the TSR. The fact that egg and early larval stages are temperature dependent in some ectotherms, but typically not in crustaceans, shows that different groups with different life history patterns respond to temperature in different ways. This suggests different proximate mechanisms bring about temperature-size changes in different taxa, which in turn gives weight to the idea that the TSR is an adaptive response (Atkinson 1994; Atkinson, Ciotti & Montagnes 2003), i.e. there is a fitness benefit to smaller size at warmer temperature and organisms achieve this through a variety of mechanisms.


J. Forster was supported by a Natural Environment Research Council studentship (NE/G523655/1). We thank Wolf Blanckenhorn, Joel Kingsolver and two anonymous reviewers for their insightful comments.