Modeling the costs and benefits associated with the evolution of endothermy using a robotic python

The maintenance of high and relatively constant temperature by metabolic means (homeothermic endothermy) constitutes a major innovation that appeared independently in two vertebrate lineages: birds and mammals (Bennett and Ruben, 1979). While daily and seasonal heterothermic endothermy are widespread among birds and mammals and mitigate energetic costs (Grigg et al., 2004), the extreme complexity and energetic demands of homeothermic endothermy suggest that there must have been substantial selective advantages driving its evolution (Bennett and Ruben, 1979; Koteja, 2004). While one can postulate advantages to the ability to maintain elevated temperature, it is unclear which selective forces may have initiated the evolution of endothermy (Hayes and Garland, 1995; Koteja, 2000; Ruben, 1995).

The evolution of homeothermic endothermy has generated considerable scientific controversy, and several hypotheses have been advanced (Hayes and Garland, 1995; Ruben, 1995). These include thermoregulation per se (Crompton et al., 1978; Heinrich, 1977), increased aerobic capacity (Bennett and Ruben, 1979) and intense parental care (Farmer, 2000, 2003; Koteja, 2000). The reproductive model was formulated by Farmer (2000), and it divides the evolution of homeothermic endothermy into two steps. As a first step, increased thermogenesis was selected for (at low associated costs) to provide embryos with a developmental environment that had a relatively high mean temperature and low variation. During the second step, additional benefits, such as the capacity for prolonged physical activity, would have then permitted the emergence of extensive postnatal parental care including food provisioning to progeny, a common feature of birds and mammals (Koteja, 2000). This second step would thereby result in the fixation of high performance (and thus high-energy demand systems) and homeothermic endothermy as displayed by birds and mammals. The reproductive model is distinct from previous theories in that endothermy is primarily beneficial to the progeny rather than the adult (Farmer, 2000).

The reproductive model has been extensively discussed and debated (Angilletta and Sears, 2003; Farmer, 2003), with the debate centering on whether the thermal benefits accrued by the offspring during development (e.g. increased survivability, decreased incubation period) outweigh the energetic costs exacted on the parents. Opponents not only suggest that these benefits are unlikely to exceed the costs but also point to the mixed results attained by studies that have attempted to quantify these benefits (Andrews et al., 2000; Shine et al., 1997). Accurate assessment of the problem, however, requires that the energetic costs to the parent must be assessed relative to these benefits, including factors that may affect such costs (e.g. nest site insulation).

While studies on heterothermic organisms have provided valuable comparative data (Grigg et al., 2004), empirical tests of any model for the evolution of homeothermic endothermy are rare. The unique characteristics of pythons, in which females of some species are facultatively endothermic during egg brooding (Vinegar et al., 1970), have been used to support arguments for and against the reproductive model (Angilletta and Sears, 2003; Farmer, 2000). More recently, Tattersall et al. (2016) provided evidence that the relatively large tegu lizard, Salvator merianae, also employs facultative endothermy to support reproductive efforts. However, little work has been conducted to empirically examine the relationship between the energetic costs to the female and the thermal benefits to the developmental environment, or how biotic and abiotic factors related to the reproductive event may influence these costs and benefits. In fact, while python brooding is widely mentioned in discussions of facultative endothermy, only a few species are known to have significant thermogenic capability (Brashears and DeNardo, 2015; Stahlschmidt and DeNardo, 2010).

As female snakes cannot be induced to produce heat or refrain from producing heat, we designed a pseudoserpent (an artificial snake) that mimicked the insulating properties of a Children's python (Antaresia childreni), a species that has become an effective model for examining parental care tradeoffs (Lorioux et al., 2012; Stahlschmidt and DeNardo, 2009a,b). While A. childreni does not use facultative thermogenesis when brooding, we were able to regulate the thermogenic capability of the pseudoserpent. We used this pseudoserpent to (1) calculate the power required to create a constant preferred developmental temperature under various conditions with respect to environment and clutch, and (2) determine the thermal consequences of the various conditions when thermogenic capability was absent or limited to 50% of the power required for developmental homeothermy. Specifically, we conducted thermometric sets of trials that included three thermogenic potentials (homeothermy, ectothermy and an intermediate condition). We imposed these three sets of condition under multiple diel cycles of ambient temperature (2, 4, 6°C amplitude with the zenith at the preferred developmental temperature) and multiple clutch sizes (5, 10 and 15 eggs). Additionally, we conducted trials in both non-insulated and insulated nest conditions. As heat production depended on the power supplied to the pseudoserpent, for each trial condition we were able to directly estimate the energetic costs and thermal consequences for the developmental environment.

The pseudoserpent was designed to mimic the size, shape and insulation properties of a brooding Children's python (A. childreni). The body of the pseudoserpent was constructed by inserting a 130 cm spring into a 135 cm length of latex tubing (outside diameter 24 mm, inside diameter 18 mm; McMaster-Carr, Santa Fe Springs, CA, USA). To provide the pseudoserpent with the ability to generate heat, we soldered 18 resistors (0.33 Ω, All Electronics Corporation, Van Nuys, CA, USA) in series with sufficient distance between them that the total length approximated the length of the pseudoserpent's body. A piece of 18 G wire was soldered to each end of the resistor string, and then the resistors were enclosed in a piece of waterproof heat shrinkable tubing. The sealed resistor array was then inserted into the body of the pseudoserpent. A no. 3 rubber stopper was inserted into each end of the tubing, with the circuit wire exiting through a small (∼3 mm) hole in each of the stoppers. Additionally, a syringe needle (18 G, 2.5 cm) fitted with a stopcock penetrated one of the stoppers. The stopcock and needle were used for filling the pseudoserpent with water prior to each set of trials.

The complete pseudoserpent was then coiled into a beehive posture typical of a brooding female and secured in that position using fine wire (Fig. 1). To mimic the seal typically provided by the snake's head, we secured a piece of latex over the opening at the top of the coil. Antaresia childreni females coil so effectively around their clutch that the eggs are completely confined when the female is in a tightly coiled position (Stahlschmidt and DeNardo, 2008). Therefore, we placed the brooding female and her clutch on a thin bed of synthetic insulation and filled any small gaps between coils with small pieces of the same insulation. Pseudo-eggs were made by removing the fingers of large-sized latex gloves and filling them with 15±0.5 ml of water before knotting the end. The desired number of eggs (depending on trial set, see below) was then placed within the female's coils. The insulating properties of the resulting brooding pseudoserpent closely mimicked those of a live A. childreni female brooding her eggs (Fig. 2).

To regulate the thermogenic activity of the pseudoserpent, we supplied power to the pseudoserpent through an amplifying circuit that was controlled by a datalogger (21X micrologger, Campbell Scientific Instruments, Logan, UT, USA). The program increased power to the pseudoserpent, and thus current and heat production, in proportion to the difference between the pseudoserpent–pseudo-egg interface temperature (T_Int) and the target incubation temperature (T_Set, 30.5°C, which approximates the preferred incubation temperature of A. childreni; Lourdais et al., 2007).

Trials were conducted in a darkened, temperature-controlled test chamber that had a sinusoidal diel thermal cycle tightly controlled by the same datalogger used to power the pseudoserpent. Chamber temperature was controlled using feedback from a 3-thermocouple array that surrounded the pseudoserpent at a distance of 10 cm, thus averaging its immediate environmental temperature (T_Env). The environmental chamber was equipped with a small fan to maximize thermal homogeneity within the chamber.

In addition to chamber temperature, the datalogger also recorded input from a thermocouple at the center of the clutch (T_Clutch) and from a thermocouple array averaging T_Int over three points of contact. During sets of trials with nest insulation, T_Env was measured just outside the insulating box, and we additionally measured nest temperature using a thermocouple placed within the insulating box but not in contact with the pseudoserpent (T_Nest).

Trial sets were designed to test the relationship between power consumption, T_Clutch and T_Int at a specified set point (T_Set) in the presence of a diel cycle. Each trial set lasted 80 h and consisted of constant environmental conditions and clutch characteristics (see below), but variable thermogenic capability as the trial set progressed. On day 1, the pseudoserpent was supplied with unlimited power (up to 4 W, which was ∼300 mW above the power needed to maintain T_Clutch even in the most demanding conditions). On day 2, the power limit was set to half of the maximum power consumed on the previous day. At the end of day 2, power to the pseudoserpent was discontinued (i.e. an ectothermic condition was imposed). The first 8 h of day 3 were used as an equilibration period to assure that starting temperatures were the same for all three power conditions. After the 8 h equilibration period, we recorded all temperatures for a continuous 24 h period.

Trial sets were conducted with T_Env following a sinusoidal diel cycle with a zenith temperature of 30.5°C and a nadir temperature of 28.5, 26.5 or 24.5°C (a 2, 4 or 6°C diel cycle, respectively). This range of temperature cycles reflects the range reported for free-ranging nesting females of a python species that is sympatric to A. childreni (Shine et al., 1997). Three sets of trials were conducted at each of the three diel cycles, and clutch size for these trials was 10 eggs, which approximates a typical clutch size for A. childreni (Stahlschmidt and DeNardo, 2008).

Trial sets were conducted in triplicate for clutch sizes of 5, 10 and 15 eggs, which approximates the clutch size range of captive A. childreni (D.F.D., unpublished data).

Data were analyzed by fitting multiple models to the data and comparing Akaike information criterion (AIC) values. Independent factors were the type of diel cycle (2, 4, 6°C), the clutch size (5, 10, 15 eggs) and insulation (no insulation, one layer). Supplied power (zero, half, full) fitted as an independent factor produced the lowest AIC values, but, as these power levels were programmed into the trials, and we were interested in the relationships within the levels, the final model was a general linear model (GLM) with supplied power as a covariate. All data were tested for parametric assumptions. The percentage of time the clutch (T_Clutch) and interface (T_Int) spent at suboptimal temperatures (Δ­T≥1.0°C) was calculated for each trial, then arcsine transformed and tested across each treatment (diel, clutch size, insulation) with supplied power as a covariate. All data are presented as means±s.e.m. Analyses were performed using R 2.14.0 (R Development Core Team, 2011) and models were run using the lme4 package (Bates et al., 2011).

The power required by the pseudoserpent during the trials reflected the energetic costs of maintaining T_Clutch above T_Env (Table 1). Power metrics modeled a female's metabolic rate during brooding. The maximum recorded power during a 3 day trial increased with the amplitude of the diel cycle (F_2,71=31.431, P<0.001) but was not significantly decreased with the addition of insulation (F_1,71=3.190, P=0.078). The total power supplied during a single 24 h period reflected the average power over that period. Total power was increased by the diel cycle amplitude (F_2,71=33.734, P<0.001) as well as clutch size (F_2,71=5.928, P=0.004). The addition of insulation strongly decreased total power (F_1,71=18.866, P<0.001).

The mean interface temperature difference (ΔT­_Int=T_Set−T_Int) increased with diel cycle amplitude (F_2,71=34.417, P<0.001) and was weakly decreased by the addition of insulation (F_2,71=3.806, P=0.055; Table 2). Both the maximum recorded ΔT­_Int and the percentage of time during which ΔT­_Int was at a suboptimal temperature (Δ­T ≥1.0°C) increased with diel cycle amplitude (maximum ΔT­_Int: F_2,71=57.623, P<0.001; percentage time: F_2,22=25.672, P<0.001; Table 2) but decreased with the addition of insulation (maximum ΔT­_Int: F_1,71=6.790, P=0.011; percentage time: F_1,65=14.045, P<0.001; Table 2). The variance of T_Int increased only in response to the amplitude of the diel cycle (F_2,71=20.107, P<0.001; Table 3). Clutch size had no effect on any interface metrics.

The amplitude of the diel cycle significantly increased the amount of time T_Clutch experienced a suboptimal environmental temperature by all metrics of T_Clutch: the calculated mean ΔT_Clutch (ΔT_Clutch=T_Set−T_Clutch; F_2,71=72.722, P<0.001), the maximum recorded ΔT_Clutch (F_2,71=87.118, P<0.001), the percentage of time the clutch spent at a suboptimal temperature (F_2,22=15.411, P<0.001) and the variance in temperature the clutch experienced (F_2,71=23.053, P<0.001). Interestingly, the addition of insulation also increased the mean ΔT_Clutch (F_1,71=6.921, P=0.010), the maximum recorded ΔT_Clutch (F_1,71=8.406, P=0.005), the thermal variance that the clutch experienced (F_1,71=3.610, P=0.062) and the percentage of time the clutch spent at a suboptimal temperature (F_1,65=26.398, P<0.001). Clutch size did not affect most metrics, but larger clutches increased the percentage of time a clutch experienced suboptimal temperatures (F_2,43=8.357, P<0.001).

We designed an artificial snake model (pseudoserpent) to estimate the cost and benefits associated with reproductive endothermy in pythons. The pseudoserpent mimicked a brooding Children's python (A. childreni), which, although not facultatively endothermic, has become a useful model for examining brooding tradeoffs (Lorioux et al., 2012; Stahlschmidt and DeNardo, 2009a,b).

The unlimited thermogenic pseudoserpent was able to maintain the temperature of the clutch within an optimal range (Table 2; Fig. 3), but doing so required considerable amounts of energy (Table 1; Fig. 3). At the 6°C amplitude with a 10-egg clutch, the pseudoserpent used 148±4 kJ over the 24 h trial (Table 1). In comparison, a much larger (16 kg) brooding and facultatively endothermic Burmese python (Python molurus) maintaining a 6°C gradient averages approximately 772 kJ day⁻¹ (21 g fat day⁻¹; Brashears and DeNardo, 2013). Despite heating a much larger clutch mass (50-fold), energy requirements to maintain a similar developmental homeothermic state are only 5-fold higher for the Burmese python. Such large differences in energy requirements per gram of clutch are a result of the high thermal conductivity of smaller females. Using the maximum recorded power and the temperature differential the pseudoserpent maintained at that power, the thermal conductivity of our pseudoserpent is 0.63±0.1 W kg⁻¹ °C⁻¹, closely matching the thermal conductivity of 0.65 W kg⁻¹ °C⁻¹ reported for the slightly larger ball pythons (Python regius) (Ellis and Chappell, 1987). In contrast, the two facultatively endothermic species, the Burmese python (P. molurus) and the diamond python (Morelia spilota), are considerably larger and have reduced thermal conductivity (0.27 and 0.114 W kg⁻¹ °C⁻¹, respectively) (Ellis and Chappell, 1987).

In addition to energy savings resulting from size-based differences in conductivity, larger snakes also have greater energy reserves. At 148 kJ day⁻¹, a female A. childreni would use 188 g of fat over a species-typical 48 day brooding period (Lorioux, et al., 2012), or 63% of body mass assuming a 300 g post-parturient body mass (Stahlschmidt and DeNardo, 2008). If she were to maintain the maximum power we recorded during the entirety of the same trial, she would need to use, over the 48 day brooding period, 375 g fat, which is more than her entire body mass. However, over a species-typical 60 day incubation period (Ramesh and Bhupathy, 2010), the female Burmese python, at 772 kJ day⁻¹, would require a total of 1.2 kg of fat, which represents only 7.5% of her body mass, a very feasible investment.

While endothermy buffers the developmental thermal environment from ambient conditions, the costs of endothermy are still highly sensitive to ambient conditions. Thus, even for facultatively endothermic brooders, nest site selection can be critically important for balancing the costs and benefits of endothermy. For the pseudoserpent, the cost of heating a 10-egg clutch for 48 days went from 188 g of fat (63% of body mass) for the 6°C diel cycle to 67 g of fat (22% of body mass) for the 2°C cycle.

In addition to the thermal variation of a nest, heat loss through conductivity also greatly influences the cost of endothermy (Tattersall, 2016). While reptiles do not possess an insulating coat (e.g. fur or feathers), it is plausible that an endothermic python could reduce the energy costs of regulating the egg developmental environment by altering blood perfusion and directing greater peripheral flow to her body wall that faces her eggs, while reducing flow to the side away from the eggs. Alteration in blood flow is a well-documented component of thermoregulation in squamates (Porter and Witmer, 2015; Tattersall et al., 2016; Weathers, 1971), and peripheral flow adjustments associated with brooding remain unexplored and deserve attention.

While we could not evaluate changes in conductivity associated with altered perfusion, we could evaluate the effect of nest insulation on conductive heat loss, which assuredly would influence the cost of regulating the developmental environment (Farmer, 2001; Tattersall, 2016). The addition of a single layer of insulation during the 6°C diel cycle required 132 g of fat (44% of body mass), which is 30% less than the 188 g of fat calculated for our non-insulated trials. The thermal conductivity of the insulated nest environment we created for our experimental trials was 2.0±0.3 W °C⁻¹ m⁻¹. In comparison, the thermal conductivity of soil can range widely, from 0.3 to 4 W °C⁻¹ m⁻¹ depending on the type of soil and the amount of moisture. For dry soil, the thermal conductivity is approximately 1 W °C⁻¹ m⁻¹ (Somerton, 1992), and a burrow with this conductivity would reduce the energetic costs of our model even further, requiring 98 g of fat (31% body mass) during a 48 day brooding period.

It is pertinent to note that convection within our test chamber, created by a small fan used to avoid thermal stratification, may have led to an overestimation of the energetic costs associated with brooding if the female nested in a location with minimal air flow. However, the natural nesting environments of A. childreni are currently unknown, and both facultatively endothermic species (P. molurus and M. spilota) have been reported brooding on the surface where convection could easily equal or exceed that present in our chamber (Ramesh and Bhupathy, 2010; Slip and Shine, 1988).

The tight brooding posture of A. childreni (Fig. 1) reduces clutch evaporative water loss 15-fold (Lourdais et al., 2007), but water loss still occurs from the female and her clutch, and the entropy of vaporization associated with this water loss contributes to heat loss from the brooding female. In a previous study, water loss from brooding females and their eggs was calculated to be, on average, 90.47 mg h⁻¹ (Lourdais et al., 2007), which would result in 7.24 g of water lost over an 80 h period. Over the 80 h experimental trials, the pseudoserpent and her artificial eggs lost an average of 8.1 g, all attributable to water loss. These similar values suggest an insignificant difference in heat loss due to vaporization between the natural and artificial systems.

Clearly, microhabitat selection, altered blood perfusion and behaviors such as intermittent basking (Slip and Shine, 1988) can reduce the cost of homeothermic endothermy, and these influences are especially important to small snakes, which have limited energy reserves to support endothermy. Even using such mechanisms to reduce costs, homeothermic endothermy might still be beyond the capability of some female pythons as they invest enormous amounts of fat and muscle reserves into the production of the clutch (Lourdais et al., 2013), and brooding occurs immediately following this investment.

The effect of the thermal developmental environment on offspring fitness is still being explored in pythons. Early studies showed that P. molurus eggs are sensitive to increases in ΔT_Clutch, with survival rates quickly approaching zero when ΔT_Clutch>3°C (Vinegar, 1973). In the ectothermic water python (Liasis fuscus), however, egg survival seems robust to daily temperature fluctuations (maximum ΔT_Clutch=8°C), although higher developmental temperatures decrease the length of incubation (Shine et al., 1997), which can result in an increase in hatchling survival rate (Madsen and Shine, 1999).

The pseudoserpent was able to maintain an optimal environment for the artificial clutch with unlimited power, mimicking a thermogenic python capable of homeothermic endothermy. The fraction of time the clutch spent at suboptimal temperatures was greatly reduced, as were the mean ΔT_Clutch (Table 2) and the variance of T_Clutch (Table 3). When the pseudoserpent was supplied with zero power, mimicking the ectothermic condition, the mean ΔT_Clutch increased by approximately 3°C, and the maximum ΔT_Clutch increased by approximately 5°C during the 6°C diel cycle (10-egg clutch; Table 2). Additionally, the clutch was below optimal developmental temperature 61–83% of the time (Table 2) and the variance of T_Clutch increased by approximately 3.5°C (Table 3).

As there is considerable overlap in the ranges of A. childreni and L. fuscus, it is likely that the egg development in A. childreni is similarly robust to the effects of temperature. Although pre-oviposition body temperature in a female has been shown to affect offspring fitness (Lorioux et al., 2012), no data currently exist on the relationship between clutch temperature and offspring fitness in A. childreni. Even if the A. childreni eggs are as thermally sensitive as those of P. molurus, it seems highly improbable that the benefits to offspring fitness provided by a fully thermogenic female would outweigh the energetic costs to the female, even if these energetic costs were reduced environmentally. Thus, we conclude that the high thermal conductance of small females, combined with their inability to store the large fat reserves that thermogenesis would require, prohibits them from being facultatively endothermic.

The reproductive model for the evolution of endothermy posits that thermogenesis was initially selected for the benefits it provided to the developing offspring (Farmer, 2000). In pythons, this would have entailed small increases in thermogenesis that provided sufficient offspring benefits to drive full facultative endothermy. As much of the debate regarding the reproductive model centers on how the costs of such limited thermogenesis would be offset by increases in offspring benefits (Angilletta and Sears, 2003; Farmer, 2003), we examined these costs when the maximum amount of power supplied to the pseudoserpent was reduced by half, mimicking a limited thermogenic state.

In the limited thermogenic state, the pseudoserpent spent more time drawing its maximum power. Thus, mean and total energy did not decrease by 50%, but decreased by around 30% across treatments. This reduced the total energy cost that an unlimited thermogenic, 600 g female would have to expend during 60 days under a 6°C diel cycle from 235 g of fat (39% body mass) to 165 g of fat (28% body mass). These costs were offset by substantial benefits to the clutch over the fully ectothermic state. Under a 6°C diel cycle, the maximum ΔT_Clutch and the percentage of time the clutch spent below optimal developmental temperature decreased by 52% and 46%, respectively. The mean ΔT_Clutch decreased by 58%, and the variance of T_Clutch decreased by 73%. This trend was repeated under the 2 and 4°C diel cycles. Thus, a limited thermogenic female under a 2°C diel cycle would lose only 62 g of fat (10% initial body mass) but still gain a 50–75% reduction in thermal costs.

These results suggest that the evolution of limited thermogenesis might confer disproportionate benefits to the offspring, especially in the reduction of thermal variance, which can affect fitness (Bozinovic et al., 2011; DuRant et al., 2013; Shine et al., 1997). Although the high thermal conductivity of a small species would likely still make the energetic cost prohibitive, a full explanatory model must include the reduction in costs conferred by environmental variables (e.g. tropical environment, nest site selection). Our results also suggest that a female's maximum metabolic rate might be a constraint in the evolution of limited thermogenesis.

Our results highlight several novel aspects of python endothermy. Unexpectedly, clutch size had no significant effect on any energetic or thermal metric except the percentage of time the clutch spent at optimal temperature (Δ­T≥1.0°C; Tables 1–3), with a large clutch size slightly depreciating the thermal environment (Fig. 4). This result suggests there might not be an increased energetic cost associated with the thermal inertia of larger clutches, a result supported by data from P. regius showing that energy expenditure from coiling does not increase even with a 50% increase in clutch size (Aubret et al., 2005). The depreciation of the thermal environment, however, might act as a selective force in determining species clutch size.

Previous work has indicated that brooding P. molurus regulate their own body temperature during facultative endothermy and only indirectly regulate clutch temperature (Brashears and DeNardo, 2013). Similarly, the pseudoserpent increased heat production in proportion to the magnitude of ΔT_Int. As the thermocouples were in direct contact with the inner surface of the pseudoserpent, ΔT_Int was presumably largely driven by the pseudoserpent's temperature. This had the counter-intuitive effect of reducing the benefits to the clutch when insulation was added (Table 2). Under the insulated condition, the pseudoserpent consumed less power (Table 1) to regulate its own temperature (Table 2), which ultimately decreased heat transference to the clutch. Thus, the presence of a nest environment during the evolution of endothermy would only have been adaptive if the lowered energetic costs were not offset by a depreciated nest environment.

As pythons are capital breeders, energy allocated to thermogenesis would likely come at a cost in the form of reduced clutch size. Although the exact composition of A. childreni eggs is unknown, lipids are responsible for 14% of the wet mass of eggs in L. fuscus (Speake et al., 2003). For a 15 g pseudoserpent egg, this would translate into a minimal cost of 77.7 kJ per egg. At the maximum diel cycle and without insulation, providing developmental homeothermy would cost almost two eggs (Table 1). In addition to lipids, eggs contain protein and other nutrients, and their production requires metabolic energy (Thompson and Speake, 2003). Therefore, this approximation may overestimate the relative costs. It would not be surprising if facultatively endothermic pythons had reduced clutch sizes relative to female body mass, a line of inquiry that is becoming more tractable as comparative reproductive data emerge.

Our results provide some insight into the costs and benefits that may have driven the evolution of endothermy. The mean power consumed by the pseudoserpent increased dramatically in response to increased diel cycle amplitude in both the limited and full thermogenic states (Fig. 5A). However, increasing diel cycles increased only the thermal variance of the clutch for pseudoserpents mimicking the ectothermic and limited thermogenic states, but not the full thermogenic state (Fig. 5B). An intuitive interpretation is that environmental temperatures could have driven the evolution of facultative endothermy by initially increasing offspring fitness through a reduction of clutch variance, but afterwards environmental temperatures would mainly select for a reduction of energetic costs (e.g. increased insulation). The achievement of lower energetic costs could then allow new selective pressures to dominate. As heterothermy is likely the ancestral condition for birds and mammals (Grigg et al., 2004), a scenario in which new selection pressures extended endothermy to other life stages is plausible.

In summary, our results demonstrate that pseudoserpent thermometry is a useful tool for investigating the evolution of python endothermy and it provides insight into the evolution of endothermy in general. Our model was able to quantify many of the energetic costs and thermal benefits associated with python brooding. Future research will benefit from the expansion of the model to facultatively endothermic python species to gain greater understanding of the role played by body size in the evolution of endothermy.

We would like to thank C. Garcia for early work with developing the pseudoserpent and two anonymous reviewers whose comments improved this manuscript.