Rate of egg maturation in marine turtles exhibits ‘universal temperature dependence’



1. The metabolic theory of ecology (MTE) predicts that, after correcting for body mass variation among organisms, the rates of most biological processes will vary as a universal function of temperature. However, empirical support for ‘universal temperature dependence’ (UTD) is currently equivocal and based on studies of a limited number of traits.

2. In many ectothermic animals, the rate at which females produce mature eggs is temperature dependent and may be an important factor in determining the costs of reproduction.

3. We tested whether the rate of egg maturation in marine turtles varies with environmental temperature as predicted by MTE, using the time separating successive clutches of individual females to estimate the rate at which eggs are formed. We also assessed the phenotypic contribution to this rate, by using radio telemetry to make repeated measurements of interclutch intervals for individual green turtles (Chelonia mydas).

4. Rates of egg maturation increased with seasonally increasing water temperatures in radio-tracked green turtles, but were not repeatable for individual females, and did not vary according to maternal body size or reproductive investment (number and size of eggs produced).

5. Using a collated data set from several different populations and species of marine turtles, we then show that a single relationship with water temperature explains most of the variation in egg maturation rates, with a slope that is statistically indistinguishable from the UTD predicted by MTE. However, several alternative statistical models also described the relationship between temperature and egg maturation rates equally parsimoniously.

6. Our results offer novel support for the MTE’s predicted UTD of biological rates, although the underlying mechanisms require further study. The strong temperature dependence of egg maturation combined with the apparently weak phenotypic contribution to this rate has interesting behavioural implications in ectothermic animals. We suggest that maternal thermoregulatory behaviour in marine turtles, and many other reptiles, is consistent with a strategy of adaptively increasing body temperatures to accelerate egg maturation.


Virtually all biological rates are affected by temperature. This is particularly apparent in ectotherms where body temperatures are more sensitive to fluctuations in ambient conditions. Development times (Gillooly et al. 2002), growth rates (Lopez-Urrutia 2008), fecundity (Berger, Walters & Gotthard 2008), and longevity (Munch & Salinas 2009) of ectothermic animals have been shown to be temperature dependent, which may in turn favour thermoregulatory behaviour that optimises such rates (Lillywhite, Licht & Chelgren 1973; Autumn & De Nardo 1995). Although ubiquitous in nature, the mechanisms underpinning biological rate–temperature relationships remain controversial. According to the metabolic theory of ecology (MTE), biological rates at all levels of organisation are rooted in the thermodynamics of enzyme-catalysed reactions and will therefore share a common dependence on temperature (Brown et al. 2004). MTE is founded on the premise that by limiting the flux of energy and resources in organisms, metabolic rate sets the pace of most higher-order ecological processes, ranging from whole-organism physiology to ecosystem dynamics (Brown et al. 2004). Proponents of MTE further suggest that whole-organism metabolism is governed primarily by body size and the effects of temperature on biochemical kinetics (Gillooly et al. 2001). Thus, after correcting for body mass variation, the majority of biological rates (R) are predicted to vary as a function of temperature according to a Boltzmann-Arrhenius relationship of the following form (Gillooly et al. 2001; Brown et al. 2004):

image(eqn 1)

where b0 is a normalisation constant, T is the absolute temperature (in °K), k is Boltzmann’s constant (8·62 × 10−5 eV/K), and E is the activation energy of rate-limiting metabolic reactions (in electron-volts, or eV). Moreover, MTE makes the more specific prediction that E will consistently fall in the range 0·6–0·7 eV for aerobic animals, reflecting the average activation energy of the respiratory complex (Gillooly et al. 2001; Brown et al. 2004). This has been termed ‘universal temperature dependence’ (UTD; Gillooly et al. 2001) by proponents of MTE and supported with empirical data for a number of different biological rates (e.g. Gillooly et al. 2001, 2002; Brown et al. 2004; Savage et al. 2004; Allen et al. 2006).

While appealing in its generality, UTD has provoked a great deal of criticism (reviewed in O’Connor et al. 2007a). Points of contention include the validity of extrapolating Boltzmann kinetics to describe complex biological pathways (Clarke 2006; O’Connor et al. 2007a); the failure of UTD to account for acclimation, phenotypic variation or evolutionary adaptation in the temperature responses of biological rates (Clarke 2004; O’Connor et al. 2007a); and the difficulty with constructing a test of the theory itself, given that its quantitative predictions do not necessarily rely on the proposed mechanisms (O’Connor et al. 2007a). On these grounds, critics have argued that UTD should be considered a statistical approximation rather than a mechanistic model based on first principles (Clarke 2004, 2006; O’Connor et al. 2007a; but see Gillooly et al. 2006; Allen & Gillooly 2007 for a rebuttal). Nonetheless, UTD is at the very least an intriguing phenomenological relationship that warrants further investigation. For example, mass-corrected metabolic rate (Gillooly et al. 2001; Meehan 2006), development times (Gillooly et al. 2002; O’Connor et al. 2007b), individual and population growth rates (Savage et al. 2004; Lopez-Urrutia 2008), longevity (Munch & Salinas 2009) and rates of DNA evolution and speciation (Allen et al. 2006) have all been shown to exhibit Arrhenius temperature dependence that approximates the predicted range of = 0·6–0·7 eV. However, other studies have reported significant deviations from UTD for many of the same rates (Frazier, Huey & Berrigan 2006; Algar, Kerr & Currie 2007; de Castro & Gaedke 2008). Empirical tests drawing on a wider range of biological rate processes are therefore an important step in evaluating the predictions of UTD.

In many ectothermic animals, the rate at which females produce mature eggs is temperature dependent (Carroll & Quiring 1993; Calbet & Agusti 1999; Hays et al. 2002a; Berger, Walters & Gotthard 2008; Lourdais, Heulin & DeNardo 2008). This rate is likely to affect important fitness outcomes, such as fecundity (Carroll & Quiring 1993; Berger, Walters & Gotthard 2008), or the time invested in reproduction and the associated physiological and/or survival costs for females (see Shine 1980 and Williams 2005 for reviews). However, studying the factors that influence egg maturation rates is difficult in practice. In this study, we use marine turtles as models to test whether rates of egg maturation vary with temperature in accordance with UTD. Female marine turtles lay multiple clutches within a season (Miller 1997), and their reproductive physiology is such that the interval separating successive clutches (‘internesting intervals’) can be used to estimate the time required to produce mature eggs. In particular, ovulation immediately follows the laying of a previous clutch (Owens 1980), and embryonic development arrests at a very early stage prior to oviposition (mid gastrulation; Miller 1985), meaning ontogenetic growth is expected to contribute relatively little to overall rates of egg maturation. As yolk formation is complete prior to the arrival at the nesting grounds in marine turtles (Rostal et al. 1997), we define ‘egg maturation’ as the post-ovulatory phase of egg production, including fertilisation, albumen deposition, shell membrane development and calcification (Owens 1980). Although Ridley turtles (Lepidocelys spp.) often retain mature eggs in the oviducts to synchronise with mass nesting events (Owens & Morris 1985), egg retention has not been reported in other species of marine turtles when suitable nesting conditions are available. We therefore expect internesting intervals to be a reliable (inverse) estimate of the rate at which eggs mature in non-Ridley species.

Previous studies of green (Chelonia mydas) and loggerhead turtles (Carretta carretta) have shown that the lengths of internesting intervals decrease as a function of increasing water temperature (Sato et al. 1998; Hays et al. 2002a), in a manner that is consistent among different species and populations (Hays et al. 2002a). However, these studies have also revealed considerable residual variation not explained by temperature, which might indicate an additional phenotypic component to the rate of egg maturation (Sato et al. 1998; Hays et al. 2002a). The objectives of this study were therefore twofold. First, we assess the relative contributions of temperature and maternal phenotypes to rates of egg maturation in green turtles, using radio-telemetry to make repeated measurements of internesting interval lengths for individual females across a seasonal water temperature gradient. We also relate internesting intervals to specific maternal phenotypes which might be expected to influence egg maturation rates, such as body size (as per the MTE; Brown et al. 2004) and reproductive investment (size and number of eggs produced). Second, we examine whether the temperature-dependence of egg maturation supports the quantitative predictions of MTE, i.e. that variation in rate with temperature is described by an Arrhenius relationship with a slope in the range of = 0·6–0·7 eV.

Materials and methods

Study site and field procedures

Data were collected for green turtles (Chelonia mydas) nesting at Ascension Island, South Atlantic Ocean (14°20′W, 7°55′S) during the 2007 breeding season. The study was conducted on Long Beach, which supports the highest density and numbers of nesting turtles on the island (Godley, Broderick & Hays 2001). To assess the contribution of maternal phenotype to the rate of egg production, radio-telemetry was used to monitor the nesting activity of individual females and make repeated measures of nesting interval length (days from laying until subsequent nesting activity; Sato et al. 1998; Alvarado & Murphy 1999; Hays et al. 2002a) as they returned to lay successive clutches. VHF radio transmitters were attached to the carapaces of a randomly selected sample of females (= 20) nesting between the 2nd and 12th of January, using a two-part epoxy resin (Power-Fast+; Powers Fasteners Inc., Brewster, NY, USA). Females were also fitted with a Passive Integrated Transponder tag (PIT, Identichip; Animalcare Ltd., York, UK) implanted into the triceps muscle of the right fore-flipper to assist identification (Godley, Broderick & Moraghan 1999). Over 95% of nesting activity at Ascension Island occurs between January and May (Godley, Broderick & Hays 2001), so there is a high probability that tagged females were encountered while depositing their first clutch. Nesting females were subsequently re-located using a scanning AR-8200 VHF receiver (AOR, Tokyo, Japan) and YAGI antenna (Biotrack, Dorset, UK) during nightly patrols of the nesting beach (20.00–05.00) conducted at 1-h intervals until 1st April when tracking was terminated. Females were observed from a distance to allow nest excavation and then approached to confirm whether eggs were present. Nesting interval length was then calculated as the number of days elapsed between laying and the subsequent nesting attempt (following Sato et al. 1998; Alvarado & Murphy 1999; Hays et al. 2002a).

Body mass is a significant determinant of metabolic rate and the rates of other biological processes (Gillooly et al. 2001), but is logistically difficult to measure in adult marine turtles (e.g. Hays et al. 2002b). Curved carapace length (CCL; ±1 cm) was therefore used as an estimator of body mass for study females, as CCL and mass are highly correlated in the study population (r2 = 0·69; Hays et al. 2002b). As the size or number of eggs produced may also contribute to the time necessary to produce a clutch, mean egg mass (±0·2 g) for each clutch was calculated from a sample of 3 eggs collected at approximately the 10th, 50th and ca. 10th from last positions in the laying sequence (to capture the full range of intra-clutch egg size variation; Hays, Adams & Speakman 1993). The locations of all clutches were marked and clutch size estimated post-hatching from the number of hatched and unhatched eggs (see Broderick et al. 2003). Total clutch mass was calculated as clutch size × mean egg mass.

Water temperature measurements

Mean water temperature for the duration of each observed nesting interval was estimated from night-time MODIS-Aqua sea surface temperature (SST) images (4 km resolution; Goddard Space Flight Centre). Daily mean SST for a box of 1° latitude × 1° longitude centred on Ascension Island (14°20′W, 7°55′S) was extracted using MATLAB v7 (MathWorks, Natick, MA, USA), and these data were used to calculate mean SST between the start and end date of each nesting interval.

The reliability of SST as a proxy for water temperatures experienced by internesting green turtles was validated using data from animal mounted time-depth recorders (TDRs) collected as part of a previous study at Ascension Island (= 6 females; Hays, Metcalfe & Walne 2004). Mean SST during the period for which each TDR was deployed was calculated as described earlier. Mean TDR temperatures were slightly higher than mean SST measured across the same inter-nesting period (TDR range = 27·60–27·75 °C; SST range = 27·53–27·59 °C; Generalised linear mixed models (GLMM) with measurement method as a fixed factor and internesting period as a random effect, inline image = 5·1, = 0·004), but the difference between methods was marginal (estimate ± SE = 0·12 ± 0·02 °C) when compared with the 2·5 °C variation in SST over the course of the present study (range = 25·2–27·7 °C; Fig. 1). A close correspondence between SST and water temperatures experienced by turtles is expected at an oceanic nesting site such as Ascension Island, where substantial surface mixing dispels any thermal gradients over the normal range of dive depths (e.g. Hays et al. 2000).

Figure 1.

 Sea surface temperature at Ascension Island during the 2007 green turtle nesting season (from night-time MODIS-Aqua satellite data). The shaded area shows the period during which the present study was conducted.

Statistical analysis

Generalised linear mixed models with female identity included as a random factor were used to evaluate the effects of SST, clutch size, egg size and total clutch mass (clutch size multiplied by mean egg mass) on internesting interval length. Clutch characteristics could not be obtained for all nesting intervals as females were occasionally encountered during the post-laying, cover-up phase of nesting and a number of clutches were destroyed by tidal inundation or excavation by other turtles before clutch size could be determined. Analyses including measures of maternal reproductive investment were therefore performed using a subset of intervals for which clutch data were available (see Results for sample sizes). The significance of fixed effects in GLMMs (SST and clutch characteristics) was assessed using likelihood ratio tests, and models were simplified by stepwise deletion of non-significant predictors (α = 0·05), starting with the effect with the highest P-value (Crawley 2007).

To estimate the overall proportion of variation in nesting interval length explained by phenotypic differences among females, we calculated the intra-class correlation coefficient, or repeatability score (R) using the equation inline image, where inline image is the among-female variance component and inline image is the within-female (or residual) variance component from a GLMM fit using restricted maximum likelihood (Nakagawa & Cuthill 2007; Zuur et al. 2009). We limited the analysis to only those females for which two or more interval length measurements were made during the study. Significance of the among-female variance component was assessed by comparing a GLMM containing the random effect to a linear model fit by generalised least squares using a likelihood ratio test (as recommended by Zuur et al. 2009).

Details of further statistical tests are presented in Results. All analyses were performed using R v. 2.5.1 (R Development Core Team 2007).


Determinants of internesting interval length in green turtles

A total of 46 internesting intervals from 18 individual green turtles was recorded over the course of the study (two tagged females were never relocated). Females that were relocated laid a minimum average ± SD of 3·9 ± 1·1 clutches (range = 2–5) over the study period, separated by an average interval of 13·3 ± 1·6 days (range = 11–17 days).

As a result of climatic trends at Ascension Island, SST increased across the study period (Fig. 1), and this was associated with a significant reduction in the length of internesting intervals of study females (likelihood ratio test, inline image = 33·5, < 0·001, = 46 intervals from 18 females; Fig. 2a). To account for the possibility that seasonally changing factors other than temperature contributed to this trend (e.g. increased efficiency of egg production, reduced harassment by males), we constructed a model containing both SST and the midpoint date of each internesting interval (days since 1st January 2007) as explanatory variables. Date did not explain significant variation in interval length after controlling for SST (likelihood ratio test, inline image = 1·2, = 0·27), suggesting that temperature drove the observed seasonal decrease in interval lengths.

Figure 2.

 Effects of temperature and maternal phenotype on nesting interval length for green turtles nesting at Ascension Island. (a) Relationship between mean sea surface temperature (SST) during the nesting interval and interval length. (b) Repeatability of nesting interval lengths for individual turtles measured two or more times during the season (= 14) after controlling for the effects of SST. Positive values indicate longer nesting interval lengths than predicted by the relationship with water temperature.

In contrast to the effects of temperature, internesting interval length was largely independent of maternal phenotype. For females where two or more measurements were made (n = 14 females and 42 intervals), individual repeatability of nesting interval length was low and not statistically significant (= 0·03, inline image = 0·013, = 0·91; Fig. 2b). Similarly, after controlling for SST there were no significant relationships between the length of the interval preceding a clutch and the number, size or total mass of eggs produced (likelihood ratio test, clutch size: inline image = 0·13, = 0·71; egg mass: inline image = 0·017, = 0·90; total clutch mass: inline image = 0·07, =0·79; = 36 clutches from 15 females; Fig. 3). Female size, measured as CCL, was also unrelated to interval length (multiple regression with SST as a covariate; F1,44 = 0·04, = 0·85). Thus, of the variables measured, SST was the only significant predictor of inter-nesting interval lengths in green turtles.

Figure 3.

 Effect of maternal reproductive investment on nesting interval length for green turtles at Ascension Island. Relationships between (a) egg size and (b) clutch size and nesting interval length after controlling for the effects of sea surface temperature.

Does egg maturation exhibit UTD?

To test whether the temperature dependence of egg maturation supports the quantitative predictions of UTD, we compiled a data set of 82 internesting interval lengths and water temperatures for green and loggerhead turtles nesting in Japan (= 24; Sato et al. 1998), Cyprus (= 10; Hays et al. 2002a) and Ascension Island (= 46, this study; = 2, Hays et al. 2002a). Water temperature was measured directly using animal-borne data loggers in previous studies (Sato et al. 1998; Hays et al. 2002a) and estimated from nocturnal SST in the present study. Egg maturation rate (R) was calculated as the reciprocal of internesting interval length (in days−1) and modelled as a function of water temperature according to an Arrhenius relationship of the form proposed by MTE (eqn 1). Maternal body size is not a significant determinant of interval length in either species represented in the combined data set (this study; Sato et al. 1998), so R was not mass-corrected. Rearranging eqn (1) generates a linear model of the following form:

image(eqn 2)

Thus, according to MTE, an Arrhenius plot of log-transformed rate against inverse absolute temperature (1/kT) should generate a linear relationship with a slope of E that is consistently in the range of 0·6–0·7 eV (i.e. UTD). This prediction was supported by our data set. A linear regression of R on 1/kT explained a majority of variation in egg maturation rates (F1,80 = 226·7, < 0·001, r2 = 0·74; Fig. 4), with an estimated slope of = 0·78 eV (95% CI: 0·68–0·89) that was not significantly different from = 0·7 eV at the upper bound of the MTE predicted range (F-test, F1,80 = 2·486, = 0·12). Residual deviances from the fitted model were not significantly different for data taken from different sources (one-way anova, F2,79 = 0·035, = 0·97) or species (F1,80 = 0·09, = 0·76), indicating that the single model was sufficient to describe the temperature dependence of egg maturation in all cases.

Figure 4.

 Arrhenius plot of the relationship between inverse water temperature (1/kT) and rate of egg maturation (R, days−1) for green turtles (solid symbols) and loggerhead turtles (open symbols) nesting in Cyprus (triangles), Ascension Island (circles) and Japan (diamonds). Data are plotted according to the Boltzmann-Arrhenius factor (eqn 2), where k is Boltzmann’s constant (8·62 × 10−5 eV/K) and T is absolute water temperature (in °K).

Others have noted that the Arrhenius thermal scaling relationship favoured by MTE is only one of several statistical models that can be used to describe the temperature dependence of biological rates (Clarke 2004, 2006; O’Connor et al. 2007a). We therefore compared the relationship between egg maturation rates and water temperature using several different models including exponential, double logarithmic, and quadratic. All of the models fit the data equally parsimoniously when evaluated using the sample size–corrected Akaike Information Criterion (i.e. all ΔAICc < 2; Burnham & Anderson 2002).


This study has shown that egg maturation rates in marine turtles vary with environmental temperature in accordance with the predictions of the MTE, but are largely independent of maternal phenotype. An Arrhenius thermal scaling relationship of the form proposed by MTE (eqn 1) explained 75% of the variation in egg maturation rates, with a slope that was statistically indistinguishable from the predicted range of = 0·6–0·7 eV. This range has been described as the UTD of biological rates (Gillooly et al. 2001). Interestingly, while the estimated value of = 0·78 eV for egg maturation rates in marine turtles is marginally outside the MTE predicted range (albeit not significantly so), it is remarkably consistent with the temperature dependence of basal metabolic rate estimated for reptiles in general (= 0·76 eV, extracted from figure 1F in Gillooly et al. 2001).

Such close correspondence in the temperature dependence of metabolic rate and higher-order physiological processes is intriguing; however, the mechanistic implications are far from clear (Clarke 2006; O’Connor et al. 2007a). Proponents of MTE argue that because UTD is derived from first principles, empirical support for its predictions helps to validate the underlying mechanism, i.e. that the thermodynamics of enzyme-catalysed metabolic reactions underpins many other biological rates (Brown et al. 2004). However, critics have questioned the mechanistic provenance of UTD and suggested it should (at best) be considered a statistical approximation with as yet unknown causes (Clarke 2004, 2006). We should note that our purpose here was to evaluate the quantitative predictions of UTD, not the underlying mechanisms. However, several points are worth making. First, the time taken for females to produce mature eggs is likely to reflect a complex integration of physiological rates (e.g. albumen biosynthesis, active transport of Ca2+ into the oviducts), many of which are dependent on cellular energy production (e.g. Thompson et al. 2007). It therefore seems intuitive that maternal metabolism might limit egg maturation rates, and that we might expect similar temperature dependences for these rates. However, it is also clear that statistical similarities in the fit of empirical data by an Arrhenius relationship do not alone support a particular causal mechanism (O’Connor et al. 2007a). Indeed, consistent with other recent studies (Clarke 2004; O’Connor et al. 2007a; Munch & Salinas 2009), we found that alternative statistical models described the relationship between temperature and egg maturation rates equally parsimoniously. While this does not invalidate our main finding that the temperature dependence of egg maturation in marine turtles is quantitatively similar to that of many other ecological rates (e.g. Gillooly et al. 2001, 2002; Brown et al. 2004; Allen et al. 2006; O’Connor et al. 2007b), it highlights a clear need for robust tests of underlying mechanisms.

Despite agreement with UTD, there was nonetheless significant residual variation in egg maturation rates that could not be explained by either temperature or phenotypic differences among females. Contrary to the allometric scaling predictions of MTE, egg maturation rates in marine turtles were independent of maternal body size (a strong correlate of mass; see also Sato et al. 1998). This is consistent with several recent studies suggesting that body size is a weak predictor of biological rates at the intra-specific level (Tilman et al. 2004; Munch & Salinas 2009), while temperature continues to explain substantial variation (Munch & Salinas 2009). Egg maturation rates were also independent of either the size or number of eggs produced in a clutch. This might be expected in marine turtles, given that all of the eggs in a clutch mature simultaneously (Owens 1980) and albumen deposition, which accounts for much of the variation in egg mass (Wallace et al. 2006), is completed soon after ovulation and unlikely to be rate limiting (Owens 1980). Sato et al. (1998) hypothesised that physiological differences among females (e.g. hormonal state or fat reserves) might account for much of the residual variation in internesting interval lengths not explained by temperature. However, the low intra-individual repeatability of interval length in the present study suggests that such endogenous factors may be of minor importance.

Given the small amount of variance explained by maternal phenotypes, it seems possible that much of the residual variation in internesting interval lengths may arise from stochastic events that interfere with nesting patterns, e.g. rough sea states, human disturbance, level of female harassment by males, and timing of arrival at beaches relative to tidal cycles. The integer scale on which nesting interval length is typically measured may also introduce error, as the nocturnal nesting habits of most marine turtle species mean that interval lengths cannot be measured on a truly continuous axis. However, while many unknown factors could plausibly influence nesting periodicity (and therefore our estimates of egg maturation rates), it is worth noting that the mean (±SD) residual deviation in internesting-interval length from the fitted relationship with water temperature was just 1·0 ± 0·1 days (back transformed from Fig. 4). Thus, females typically nested within 24 h of the exact time predicted by temperature.

The strong temperature dependence of egg maturation, combined with the relatively weak phenotypic contribution to this rate, has potentially important behavioural implications in ectothermic animals. Reproduction is costly for females: in addition to the direct energetic investment in eggs, mothers also incur costs through the physiological and behavioural changes that accompany vitellogenesis, e.g. altered hormonal states and reduced foraging opportunities (Shine 1980; Williams 2005). Many of these indirect costs are likely to be proportional to the time invested in reproduction, and may therefore favour thermoregulatory strategies that increase maternal body temperature in order to accelerate egg maturation.

Indeed, descriptions of habitat use by breeding marine turtles are broadly consistent with this hypothesis. For example, studies of loggerhead turtles suggest that females actively seek out warm water microhabitats during the internesting period (Naito et al. 1990; Sakamoto et al. 1993; Schofield et al. 2009). Female leatherback turtles (Dermocelys coriacea) have similarly been shown to restrict their movements to the warmest habitat patches at some nesting sites (Fossette et al. 2009; but not others, Wallace et al. 2005). Thermal stratification of the water column means that thermoregulatory opportunities also exist through selection of diving depths in marine turtles. For example, the contrasting dive profiles of gravid green turtles in the Mediterranean (shallow dives, stratified water column; Hays et al. 2002a,c; Schofield et al. 2009; Fuller et al. 2009) and Ascension Island (deep dives, mixed water column; Hays et al. 2000, 2002c) are consistent with a strategy of remaining in the warm surface mixed layer while maximising more energy-efficient deep dives (see Hays et al. 2000). Dive profiles also suggest that female sea turtles may alter their depth selection diurnally and seasonally to track variations in water column stratification, e.g. diving shallower in summer or at midday when warm surface layers form (Hays et al. 2000, 2002c; Yasuda et al. 2008). Maternal thermoregulatory strategies that maximise body temperature are perhaps not surprising in sea turtles, given the prevalence of such behaviour in other reptiles. In many species of lizards and snakes, gravid females maintain higher body temperatures than conspecifics through behavioural thermoregulation (reviewed in Shine 2006), which reduces the time for which eggs must be carried (Lourdais, Heulin & DeNardo 2008). However, the adaptive significance of ‘maternal thermophily’ is ambiguous in these taxa. Lizards and snakes are unusual among reptiles in having significant embryonic development prior to oviposition (Andrews 2004), meaning that thermoregulatory behaviour may serve to optimise the developmental conditions for offspring as well as accelerating egg maturation (Wapstra 2000; Shine 2006). In contrast, marine turtles (like other chelonids) lay their eggs at a very early stage in development (Miller 1985), suggesting that maternal thermophily is more likely to reflect a strategy of increasing egg maturation rates in order to reduce the time invested in reproduction.

In conclusion, we have shown that egg maturation rates in marine turtles vary with temperature in accordance with the UTD predicted by MTE, but are largely independent of maternal phenotypes. This may have significant behavioural implications for in ectothermic animals, favouring a strategy of ‘maternal thermophily’ to accelerate egg maturation and reduce the time invested in reproduction. Empirical tests of UTD have so far yielded mixed results, with some studies supporting its quantitative predictions (this study, Gillooly et al. 2001, 2002; Allen et al. 2006; Meehan 2006; O’Connor et al. 2007b; Lopez-Urrutia 2008), and others rejecting them (Frazier, Huey & Berrigan 2006; Algar, Kerr & Currie 2007; de Castro & Gaedke 2008; Irlich et al. 2009). This has prompted suggestions that UTD at best describes a central tendency in ecology, with much of the interesting biology lying in deviations from the general model (Clarke 2006; Irlich et al. 2009). Rigorous tests of the mechanisms underlying UTD are clearly needed to resolve these issues. However, we agree with Clarke (2006) that, at the very least, UTD represents an intriguing phenomenological description of ecological rate–temperature relationships. It is striking that processes as apparently unrelated as egg maturation in sea turtles, larval development in fish (O’Connor et al. 2007b) and speciation rates in foraminifera (Allen et al. 2006) converge on a similar dependence on temperature. Explaining why this is so (and indeed why some rate–temperature relationships differ) remains an important challenge for the future.


The authors thank G. Hays and three anonymous reviewers for comments that significantly improved an earlier version of this manuscript. We are also grateful to G. Hays for providing data from animal-mounted temperature loggers, the Ascension Island Government’s Conservation Department for the use of their facilities, N. Reed for helpful discussions and J. Ellick for help with fieldwork. The study was funded by NERC (SW, JDB, ACB), the Royal Society (JDB), the Darwin Initiative and the European Social Fund (ACB and BJG). MJW is funded by the Peninsula Research Institute for Marine Renewable Energy (PRIMaRE).