Diet, development and the optimization of warning signals in post-metamorphic green and black poison frogs

Authors


Correspondence author. E-mail: j.d.blount@exeter.ac.uk

Summary

  1. Many prey species are chemically defended and have conspicuous appearance to deter predators (i.e. aposematism). Such warning signals work because predators pay attention to the colour and size of signals, which they associate with unprofitability.
  2. Paradoxically, in early life stages, aposematic species are often warningly coloured, but their chemical defences are lacking because they have yet to be acquired through the diet or synthesized endogenously. This state of being conspicuous yet poorly defended must place individuals at increased risk of predation, but how they minimize this risk during development is unclear.
  3. We reared larval green and black poison frogs (Dendrobates auratus) on a relatively low or a higher food supply and tested the hypothesis that individuals with more resources should grow larger while reducing their investment in warning signals at metamorphic completion. We also assayed markers of oxidative balance (malondialdehyde, superoxide dismutase and total antioxidant capacity) to ascertain whether there were resource-allocation trade-offs that differed with diet treatments.
  4. Low-food froglets were relatively small, and their body size and signal luminance (perceived brightness) were positively correlated. In contrast, in high-food froglets body size and warning signal luminance were negatively correlated, suggesting either a resource-allocation trade-off or alternatively a facultative reduction in luminance exhibited by larger froglets.
  5. The reduction in luminance in relatively large, high-food froglets did not appear to arise because of oxidative stress: signal luminance and markers of oxidative stress were positively correlated in high-food froglets, but were negatively correlated in low-food froglets suggesting a trade-off.
  6. Our results highlight developmental plasticity in body size and coloration as affected by resource (i.e. food) supply. Such plasticity seems likely to minimize predation risk during the vulnerable period early in life when individuals are warningly coloured and must make the transition from an undefended phenotype to a mature aposematic state.

Introduction

Many prey species are defended and have conspicuous colour, pattern, acoustic or olfactory signals, which advertise unprofitability to predators (i.e. aposematism; (Poulton 1890; Eisner & Grant 1981; Ratcliffe & Fullard 2005). Until recently, aposematic signals were assumed to not vary within species to facilitate predator learning and avoidance (reviewed in Ruxton, Sherratt & Speed 2004). However, contrary to this expectation, it is becoming clear that intraspecific variation in aposematic colour and pattern is widespread (e.g. Summers, Cronin & Kennedy 2003; Bezzerides et al. 2007; Tullberg et al. 2008; Mochida 2011; Wang 2011). While many mechanisms have been suggested for the initial evolution of aposematism (receiver psychology: Speed 2001; adaptive predator behaviour: Sherratt 2002; dietary conservatism: Thomas et al. 2003; physical defences: Speed & Ruxton 2005a; defensive secretions: Gohli & Högstedt 2009), the developmental factors that give rise to phenotypic variation in aposematic signals remain little studied (Mappes, Marples & Endler 2005; Stevens & Ruxton 2012). Predators vary in hunting strategies, perceptual sensitivities or experience, which may contribute to polymorphism in aposematic signals (reviewed by Endler & Mappes 2004) and differ according to the local predator assemblage (Wang & Shaffer 2008; Mochida 2011; Valkonen et al. 2012). In addition, variation in aposematic signals may result from genetic drift (Reynolds & Fitzpatrick 2007), sexual selection (Rudh, Rogell & Höglund 2007; Maan & Cummings 2008) or environmental factors such as developmental diet (Grill & Moore 1998; Blount et al. 2012), prey density (Sword et al. 2000), parasitism (Losey et al. 1997; Lindsey & Altizer 2009) or thermoregulation (Lindstedt, Lindström & Mappes 2009).

The environment in which early development occurs is fundamental for shaping the form of the mature phenotype (Cheverud & Moore 1994; Rossiter 1996; Monaghan 2008) and may elicit the expression of plastic phenotypes adapted to one or a set of environments (Nijhout 2003; West-Eberhard 2003). Yet, in aposematic species, only a handful of studies have considered the effects of early development on adult phenotypes. Early nutrition in particular has been shown to affect the size or colour of aposematic signals at adulthood in ladybird beetles (Grill & Moore 1998; Grill 1999; Blount et al. 2012) and arctiid moths (Ojala, Lindström & Mappes 2007; Lindstedt et al. 2010). Furthermore, variability in the quantity of defensive compounds sequestered from the diet or metabolically transformed from dietary precursors during early development may directly affect the amounts of such chemicals in the body at adulthood (e.g. Daly et al. 1994; Sime, Feeny & Haribal 2000). The development of aposematic signals may also be constrained if integument colour has additional functions, such as thermoregulation (Lindstedt, Lindström & Mappes 2009). Variation in aposematic coloration has been also found in response to prey population density (e.g. Sword et al. 2000) and across seasons (e.g. Tullberg et al. 2008). In amphibians, integument coloration is largely the result of light-absorbing (e.g. xanthophores, melanophores) or light-reflecting(e.g. leucophores, iridophores) chromatic cells and their arrangement and dispersion in the skin (Grether, Kolluru & Nersissian 2004). The expression of aposematic coloration is known to change with age in some amphibians (Hoffman & Blouin 2000) and potentially can be affected by the amount of particular pigments inside the chromatic cells (e.g. carotenoids, melanins and pterines) (Fox 1976; Grether, Kolluru & Nersissian 2004).

While it is clear that there is plasticity in aposematic traits, what is less clear is how variation in resource availability during early development may drive relative investment in warning signals versus growth. Early development may be a risky stage of life in aposematic species, because chemical or other forms of secondary defences must either be produced de novo or sequestered through the diet, and therefore may be absent or incompletely developed in juveniles (Sime, Feeny & Haribal 2000; Nylin, Gamberale-Stille & Tullberg 2001; Nishida 2002). This state of being conspicuous yet poorly defended must place individuals at increased risk of predation, but how they minimize this risk during development is unclear. Hence, if predators can discern poorly defended froglets from well-defended adults, the optimal strategy during development could be to invest maximally in growth, but to reduce investment in aposematic signals and thereby minimize detection.

Body size has been considered to evolve in concert with aposematic signals in poison frogs (Hagman & Forsman 2003), particularly in highly toxic species (Santos & Cannatella 2011). It seems unlikely that restricting resource allocation to growth would be favoured by selection, because larger froglets will be better able to meet the metabolic demands of foraging for toxic dietary items during the first few, critical days following metamorphosis (Taigen & Pough 1983). Chemical defences in post-metamorphic poison frogs are in the form of neurotoxic alkaloids acquired from dietary arthropods (Daly et al. 1994; Saporito et al. 2009). However, during early development, the mechanisms of alkaloid acquisition may vary amongst species and environmental contexts. For example, maternal provision of alkaloids to embryos via the egg has been reported in the aposematic bufonid Atelopus chiriquiensis (Pavelka, Kim & Mosher 1977) and apparently to tadpoles via trophic eggs in Oophaga [Dendrobates] pumilio (Stynoski 2012). Nevertheless, in many poison frog species, the larval diet comprises alkaloid-free foods (Caldwell 1993; Caldwell & de Araújo 1998). This is the case with the green and black poison frog (Dendrobates auratus) whose tadpoles are palatable to odonate and mosquito larvae (Fincke 1994, 1999) and which lacks maternal provision of trophic eggs (Summers 1990); as such sequestration of alkaloids during larval stages seems unlikely, although this remains to be conclusively demonstrated. Acquisition and storage of toxins could be costly (reviewed in Alonso-Alvarez et al. 2004; Ruxton, Sherratt & Speed 2004) especially during immature life stages when the anatomical organization of poison glands is incomplete (Angel, Delfino & Parra 2003; Saporito et al. 2010), and resources should be mainly devoted to meet growth demands. In addition, other anatomical (e.g. mouth size), behavioural (e.g. foraging capacity) and physiological (e.g. metabolic rate) factors may constrain the acquisition of toxins in juveniles (Donnelly 1991; Saporito et al. 2010).

Relative investment in growth and aposematic signals is likely to be influenced by environmental conditions, such as the quality of the developmental diet. High levels of investment in growth or signals may be physiologically costly. In particular, rapid growth may incur production of reactive oxygen species (ROS) (Alonso-Alvarez et al. 2007; Menon & Rozman 2007). ROS are atoms or molecules that are important for intracellular signalling (e.g. Hurd & Murphy 2009), but also can cause serious damage to DNA, proteins and lipids (reviewed in Selman et al. 2012). Where there is an imbalance between ROS production and the capacity of the antioxidant defence system to inactivate ROS, a state of oxidative stress results (reviewed in Selman et al. 2012). The antioxidant system is complex and includes both endogenous (e.g. glutathione, catalase, superoxide dismutase) and exogenous, diet-derived components (e.g. vitamin E, carotenoids) (reviewed in Selman et al. 2012). However, all antioxidants are potentially limiting resources for wild animals, because antioxidants or the resources (amino acids, energy) required for their biosynthesis must be obtained in the diet, and/or because antioxidants are traded amongst competing physiological demands (reviewed in Selman et al. 2012). Indeed, all types of pigments responsible for integument coloration in animals, including poison frogs (e.g. carotenoids, melanins, pterines) (Fox 1976; Hoffman & Blouin 2000; Grether, Kolluru & Nersissian 2004), may function as antioxidants in vivo (McGraw 2005). It has recently become clear that oxidative stress can explain variation in the expression of sexual signals (Alonso-Alvarez et al. 2004; Mougeot et al. 2010). It therefore seems possible that developmental trade-offs in aposematic animals may be modulated by oxidative stress, as affected by diet. Previous studies have shown that investment in life-history traits such as growth rate, development time and size can trade against development of aposematic signals (e.g. Grill & Moore 1998; Ojala, Lindström & Mappes 2007; Lindstedt et al. 2010). However, whether biomarkers of oxidative stress correlate with warning signal production has not been studied before.

Here, we assessed the effects of variation in food supply during early development on post-metamorphic body size and aposematic coloration in the green and black poison frog. Dendrobatids undergo a complex early development, from an aquatic-cryptic phenotype to a conspicuous terrestrial one; yet, only after reaching this latter stage does dietary sequestration of toxins begin (Daly et al. 1994; Saporito et al. 2009). These characteristics make dendrobatids a good model to investigate phenotypic plasticity and resource allocation to aposematic signals versus growth during early development. We reared D. auratus larvae on either a relatively low or a higher food supply until metamorphosis was complete, whereupon we measured morphology and the spectral reflectance of skin to assess coloration and conspicuousness to predators. We hypothesized that D. auratus froglets would show developmental plasticity, and resource allocation to growth and aposematic signals would be modulated by oxidative stress. We predicted that under high-food provision froglets would grow relatively large, and they would reduce investment in warning signalling compared with low-food froglets. We also predicted that there may be a negative correlation between warning signal expression and levels of oxidative stress in low-food froglets that were resource constrained, whereas no such apparent trade-off would be evident in high-food froglets.

Materials and methods

Capture and Breeding of Adults

Adult D. auratus were collected during April and May 2010, at a shade organic coffee plantation in Santa Fe, Veraguas province, central Panama (8°31′ N, 81°03′ W). Near the plantation, individuals were randomly paired in glass terrariums (26 × 50 × 35 cm), using a simple shelter open on all sides to permit natural daylight and temperatures. All terrariums were similarly furnished with Sphagnum sp. moss and bromeliads (Catopsis wangerini Mez and Wercklé, 1904) collected from the study site. Water and food were provided ad libitum in Petri dishes (diameter: 10 cm), replenished twice daily (07:30 h and 17:00 h). Moisture levels inside each terrarium were checked at that time to ensure c. 90% relative humidity controlled by misting with filtered reverse osmosis water. Food consisted of soldierless live termites (Nasutitermes nigriceps Haldeman, 1853) collected in the field. We used termites because they are part of the natural diet of D. auratus (Taigen & Pough 1983; Caldwell 1996), could be collected in sufficient quantities at our study site, and their nutritional value is comparable to an ant/mite diet (Huey & Pianka 1981; Redford & Dorea 1984). Termites are not a source of alkaloids for dendrobatids (Daly et al. 1992), but it seems unlikely that a diet of exclusively termites would have affected skin levels of alkaloids of the breeding frogs because skin alkaloids are known to persist for years in captive D. auratus (Daly et al. 1992, 1994). Temperature and humidity in the terrariums, based on daily readings throughout the study, were 24·60 ± 0·22 °C (mean ± SE) and 91·04 ± 0·99%, respectively. Another Petri dish containing a small volume of reverse osmosis water, and covered by an upturned black plastic flowerpot, served as a site for egg deposition.

Figure 1.

Relationship between dorsal luminance (modelled based on bird vision) and snout-vent length (SVL) in froglets of the two food supply groups. Filled circles and solid line: high-food individuals; open circles and the dashed line: low-food individuals. Lines are predicted from GLMM analyses.

Larvae Rearing and Diet Manipulation

In total, 19 breeding pairs of adults produced fertile clutches, with an average clutch size of 5 ± 0·57 eggs and a latency to lay of 12 ± 2·36 days (mean ± SE). A total of 120 eggs were laid, and clutches were transferred individually to a similar empty glass terrarium where they were monitored daily. Some 30 of 120 eggs (25%) showed signs of mould infection and were carefully removed using a sterile plastic pipette and discarded; all remaining eggs hatched (90 eggs). Immediately after hatching, larvae were carefully transferred to a 700-mL plastic tub containing 100 mL reverse osmosis water covered with mosquito net, and using a split-brood design, they were randomly assigned to a food supply (treatment) group. Thus, at the start of the rearing period, the sample sizes in each group were n = 34 individuals from n = 13 families in the low-food group and n = 28 individuals from n = 9 families in the high-food group. Larvae were fed daily with King British cichlid fish flakes (Fish and Fins Ltd., East Sussex, UK). To standardize presentation, only red flakes were used, because laboratory analysis showed that total concentrations of carotenoids differed between red and brown flakes (our unpublished data). The quantity of food provided was recalculated weekly using the average body mass of low-food larvae as a reference [low-food, 8% body mass (w/w); high-food, 15% body mass (w/w)]. The same or similar levels of food have previously been employed to yield differences in growth rates of frog larvae without causing starvation (Alford & Harris 1988; LaFiandra & Babbitt 2004). Before providing fresh food, any uneaten food was removed. Once each week, 50% of the water in each plastic tub was replaced. At the onset of metamorphic climax (developmental stage 43–46; Gosner 1960), feeding ceases and nutritional needs are instead met by tail resorption and catabolism of other body tissues (Lötters et al. 2007). Therefore, we ceased food provisioning at this point. The two food supply groups had a similar duration of the tail resorption period (general linear mixed model (GLMM); with food as a fixed factor and family as a random factor; high-food: 6·32 ± 0·21 days; low-food: 6·01 ± 0·28 days; food, F1,8 = 1·22, = 0·30).

Morphometric Measurements

At weekly intervals, larvae were carefully removed from their tub and blotted dry with filter paper before being placed in a Petri dish (diameter: 5 cm) containing a known volume of reverse osmosis water and weighed to the nearest 0·001 g using an Ohaus Scout Pro balance (Ohaus Europe GmbH, Switzerland). The dorsum of each individual was digitally photographed under standardized conditions with a Canon Power Shot G6 (7.1 megapixel) camera (Canon Inc., Japan). A metal ruler was included to provide a scale for the image. Snout-vent length (SVL) was measured in each individual. SVL is considered an anatomical character that shows plasticity under different environmental conditions in anuran larvae (Vences et al. 2002; Vonesh & Warkentin 2006). Larvae were photographed daily from the first day they climbed out of the water to monitor tail length. When tail length stopped decreasing, metamorphic climax was determined to have been reached (see Caldwell & de Araújo 2004). All measurements (0·001 mm precision) were carried out using ImageJ 1.43q (Rasband 1997).

Analysis of Aposematic Signals and Background Spectra

To be discernible to a predator's eye, prey must contrast against the background where they are normally perceived in terms of colour and luminance (perceived brightness based on photoreceptor outputs) (Stevens & Ruxton 2012). We therefore measured the spectral reflectance of the skin of metamorphosed froglets on the day of metamorphic climax using a USB2000 spectrometer (Ocean Optics Inc., Dunedin, FL, USA) with a bifurcated 400-μm UV/VIS optic fibre probe connected to a pulse xenon lamp (PX-2) at an angle of 45° and corrected for lamp drift using a white diffuse spectral standard (WS-1) (Maan & Cummings 2008). Measurements were taken from four body regions in duplicate (head, dorsum, left and right flanks) and then averaged for subsequent analyses (Fig. S1 in Supporting Information). We did not measure the reflectance of the ventral skin that is unlikely to function as an aposematic signal in poison frogs (Savage 2002; Maan & Cummings 2008; Wang & Shaffer 2008). The spectral reflectance of 136 samples of background leaves collected from the forest floor was also measured in triplicate and averaged following the methodology described above (Fig. S2). The light that reaches the eye is a product of the reflectance spectra of the object observed and the irradiance of ambient light (i.e. radiance spectra; Endler 1990); therefore, measures of ambient light irradiance I(λ) were obtained at several locations in the field where adult frogs were captured for use as breeders (see above); n = 90 measurements on a sunny day and n = 85 measurements on a cloudy day (Fig. S3), using a cosine-corrected irradiance probe (CC-3-UV-T) with 180° field view, connected to an USB2000 spectrometer by means of a 400-μm UV/VIS optic fibre following the method described in Endler (1993). In all cases, spectral reflectance data were collected between 300 and 750 nm and averaged to 1-nm interval for analyses.

Modelling Predator Vision and Froglet Conspicuousness

Birds are visual-oriented predators of many aposematic species (Collins & Watson 1983; Lindström, Alatalo & Mappes 1999; Exnerová et al. 2008) including poison frogs (Master 1999). Birds seem able to perceive conspicuousness in poison frogs (Maan & Cummings 2012), and domestic chickens (Gallus domesticus) have attacked poison frogs in experiments (Darst, Cummings & Cannatella 2006). Furthermore, birds account for an appreciable amount of attacks on artificial aposematic prey in field experiments (e.g. Saporito et al. 2007; Noonan & Comeault 2009; Chouteau & Angers 2011). Colubrid snakes and spiders have also been reported to predate poison frogs (Summers 1999; Gray, Kaiser & Green 2010; Santos & Cannatella 2011), and Gray & Christy (2000) reported that the grapsid crab Armases angustum preys on D. auratus tadpoles. Therefore, our main results are based on psychophysical models of bird vision, but we replicated all analysis based on snake and crab vision following methods described in Maan & Cummings (2012). We used the Vorobyev–Osorio visual model of colour discrimination (Vorobyev & Osorio 1998), which assumes that noise in the photoreceptors limits discrimination. Discrimination (conspicuousness) values are JNDs (just noticeable differences), with a value of 1 being the threshold for discrimination, and values of between 1 and 3 generally considered to mean that two objects could only be discriminated under ideal viewing conditions (rarely the case in the field) (Stevens et al. 2013). This model has been employed to calculate discrimination values in intraspecific (Maan & Cummings 2009; Ostrowski & Pröhl 2011; Wang 2011) and interspecific studies of poison frogs (Siddiqi et al. 2004; Darst, Cummings & Cannatella 2006).

As the only bird documented to prey upon poison frogs (Baryphthengus martii) is a close relative to passerines (Livezey & Zusi 2007), we used cone sensitivity data for the blue tit (Cyanistes caeruleus), as a tetrachromatic visual model to calculate predicted photon catches for the different cone types (absorbance spectrum templates and oil droplets data from Hart et al. 2000). To model snake vision, we used a trichromatic visual model from the coachwhip colubrid snake Masticophis flagellum, with calculation of absorptance curves using a rhodopsin vitamin-A1 template according to Govardoskii et al. (2000). For crabs, we used a dichromat visual model based on absorptance curves for the fiddler crab Uca tangeri with LW sensitivity curves manually digitized from Jordão, Cronin & Oliveira (2007) and electrophysiological measures for the crab Uca thayeri with sensitivity to SW according to Horch, Salmon & Forward (2002). Details of parameters used for the vision models are provided in Table 1, and details of calculations are provided in Appendix S1.

Table 1. Vision system parameters used in bird, snake and crab psychophysical models
 BirdSnakeCrab
  1. Further details of modelling and formulas are provided in Appendix S1.

Predator vision modelledBlue tit (Cyanistes caeruleus)Coachwhip snake (Masticophis flagellum)Fiddler crab (composite of Uca tangeri and Uca thayeri)
Tetrachromat version of the Vorobyev–Osorio model (Vorobyev & Osorio 1998)Trichromat version of the Vorobyev–Osorio model (see Siddiqi et al. 2004)Dichromat version of the Vorobyev–Osorio model (see Cummings et al. 2008)
Colour perceptionBased on relative stimulation of visual cone photoreceptors with sensitivity to UV (λmax = 371 nm), short (λmax = 448 nm), medium (λmax = 503 nm) and long (λmax = 563 nm) wavelengths (Hart et al. 2000)Based on relative stimulation of visual cone photoreceptors with sensitivity to UV (λmax = 362 nm), short (λmax = 458 nm) and long (λmax = 561 nm) wavelengths (Macedonia et al. 2009)Based on relative stimulation of visual cone photoreceptors with sensitivity to long (λmax = 593 nm) and short (λmax = 430 nm) wavelengths (Horch, Salmon & Forward 2002; Jordão, Cronin & Oliveira 2007)
Luminance perceptionBased on double-cone photoreceptors specialized in achromatic sensitivity (Kelber, Vorobyev & Osorio 2003; Osorio & Vorobyev 2005).Achromatic sensitivity based on the absorbance of long-wavelength cone pigments (Macedonia et al. 2009)Achromatic sensitivity based on the absorbance of long-wavelength cone pigments (Jordão, Cronin & Oliveira 2007)
Weber fraction0·05 (Siddiqi et al. 2004)0·05 (Siddiqi et al. 2004)0·12 (Hempel De Ibarra et al. 2000)
Relative number of photoreceptorsnL = 1·00, nM = 0·99, nS = 0·71, nUV = 0·37, nD = 1·00 (Hart et al. 2000)nL = 0·85, nS = 0·10, nUV = 0·05 (Sillman et al. 1997)nL = 1·0, nS = 1·0 (Cummings et al. 2008)
Prey detection systemMostly based on vision (Martin 1999; Fernández-Juricic & Tran 2007)Mostly olfactory, but visual cues may be involved (Brodie & Tumbarello 1978; Stuart-Fox, Moussalli & Whiting 2008; Macedonia et al. 2009)The use of visual cues for prey discrimination is unknown.
Predation events on poison frogs reportedOne published report of predation in the wild (Master 1999); predation experiments using captive domestic hens (e.g. Darst & Cummings 2006; Darst, Cummings & Cannatella 2006), predation on artificial prey in the wild (e.g. Saporito et al. 2007; Noonan & Comeault 2009; Chouteau & Angers 2011) and psychophysical models of bird vision (e.g. Siddiqi et al. 2004; Wang 2011; Maan & Cummings 2012) suggest that birds are predators of poison frogs.Several accounts of predation in the wild (Santos & Cannatella 2011 supporting information); predation experiments using captive snakes (Brodie & Tumbarello 1978) suggest their likely importance as predators.One published report of predation of tadpoles in the wild (Gray & Christy 2000), apparent use of colour discrimination mostly for intraspecific communication (Detto 2007; Cummings et al. 2008; Baldwin & Johnsen 2012).

Antioxidant Activity and Oxidative Damage

On the day of metamorphic climax, 25 froglets from the high-food group and 25 from the low-food group were sampled at random and euthanized by stepped hypothermia and then stored at −80 °C until biochemical analyses. Remaining individuals were released into the wild. Samples were thawed, finely chopped using scissors and then weighed to the nearest 0·0001 g and 50 mm HEPES (N-2 hydroxyethylpiperazine-N′-2-ethanesulfonic acid) buffer solution was added on a 20% w/v basis. Samples were then homogenized ready for the following assays. We used whole frog homogenates to provide an organismal-level overview of antioxidants and oxidative damage, because individual tissues are likely to differ in antioxidant levels and oxidative damage (López-Torres et al. 1993). Assessment of oxidative stress requires a range of assays that include measures of enzymatic and non-enzymatic antioxidants because components of antioxidant defences do not act in isolation, but react and compensate for each another. In addition, it is imperative to assay oxidative damage because this reflects the outcome of oxidative stress (reviewed in Selman et al. 2012). Therefore, we assayed oxidative damage in terms of levels of malondialdehyde (MDA), which is formed by the β-scission of peroxidized polyunsaturated fatty acids (Agarwal & Chase 2002), and has been reported as a physiological marker of oxidative stress during larvae metamorphosis (Mahapatra, Mohanty-Hejmadi & Chainy 2001; Menon & Rozman 2007). In addition, we assayed superoxide dismutase (SOD), a metalloenzyme that catalyses the dismutation of superoxide into oxygen and hydrogen peroxide and forms a crucial part of intracellular antioxidant defence, for example, during early stages of development in amphibians (Montesano et al. 1989; Menon & Rozman 2007). Finally, we assayed total antioxidant capacity (TAC), which measures low-molecular weight non-enzymatic antioxidants such as vitamin E, carotenoids and flavonoids (reviewed in Selman et al. 2012).

Analysis of MDA was performed in duplicate using polypropylene screw cap tubes as described previously (Nussey et al. 2009) with some modifications. To 10 μL frog homogenate, we added 10 μL butylated hydroxytoluene solution (BHT; 0·05% in ethanol), 80 μL phosphoric acid solution (0·44 m) and 20 μL thiobarbituric acid (TBA) solution (42 mm). Tubes were capped, vortexed for 5 s and then heated for 1 h at 100 °C on a dry bath incubator. Tubes were then cooled on ice for 5 min. To extract the MDA–TBA complex, 80 μL of n-butanol (HPLC grade) was added to each tube, vortexed for 20 s then centrifuged at 15 000 g for 3 min at 4 °C to separate the two phases. A 60 μL aliquot of the upper, butanol phase was carefully removed and transferred to an HPLC vial ready for HPLC analysis. Samples (40 μL) were injected into a Dionex HPLC system (Dionex Co., Camberley, Surrey, UK) fitted with a Hewlett-Packard Hypersil 5 μ ODS 100 × 4·6 mm column with a 5 μ ODS guard column. The column temperature was set at 37 °C. The mobile phase (isocratic, flow rate: 1·0 mL min−1) consisted of a 40:60 (v/v) mixture of methanol (HPLC grade) and buffer, the buffer being 50 mm potassium monobasic phosphate (anhydrous) solution, pH adjusted to 6·8 using 5 m potassium hydroxide solution. Fluorescence detection utilized a RF2000 detector (Dionex Co.) set at 515 nm excitation and 553 nm emissions. Results are expressed as nmol MDA per g frog tissue. Forty-nine samples were assayed in duplicate, and repeatability calculated according to Lessells & Boag (1987) was high (F48,49 = 8·66, < 0·001; = 0·79).

Total SOD was assayed by measuring the dismutation of superoxide radicals generated by xanthine oxidase and hypoxanthine (Cat. No. 706002; Cayman Chemical Co., Ann Arbor, MI, USA). Following kit instructions, we mixed 100 μL of frog homogenate (see above) with 100 μL of Buffer Solution (containing 50 mm of HEPES, 2 mm EGTA, 420 mm D-mannitol and 140 mm sucrose, pH adjusted to 7·2 by adding 5 m potassium hydroxide). After vortexing for 10 s, samples were centrifuged at 1500 g for 5 min at 4 °C, and the supernatant was collected and diluted 1:20 v/v in kit sample buffer, which had itself been previously diluted 1:10 v/v in HPLC grade water. From this step onwards, kit instructions were followed exactly. Absorbances were read at 440 nm using a Spectramax M2 plate reader (Molecular Devices Corp., Sunnyvale, CA, USA). Results are expressed as units of SOD activity per mg frog tissue. 50 samples were assayed in duplicate, and repeatability calculated according to Lessells & Boag (1987) was high (F49,50 = 5·42, < 0·001; = 0·69).

TAC was assayed in terms of the capacity of the antioxidants in the sample to inhibit oxidation of ABTS® (2,2′-azino-di-[3-ethylbenzthiazoline sulphonate]) (Cat No. 70901; Cayman Chemical Co.). Frog homogenates were diluted 1:1 (v/v) in 50 mm HEPES buffer solution, vortexed for 10 s and then centrifuged at 1500 g for 5 min at 4 °C. From this step onwards, kit instructions were followed. Absorbances were read at 750 nm using a Spectramax M2 plate reader (Molecular Devices Corp.). Results are expressed as nmol of Trolox equivalents per g of frog tissue. Fifty samples were assayed in duplicate, and repeatability calculated according to Lessells & Boag (1987) was high (F49,50 = 25·15, < 0·001; = 0·92).

Data Analyses

Maximum SVL growth rate (SVLGR) was calculated by fitting a logistic curve to the data based on weekly changes in SVL and then calculating the first derivative at the point of inflection as described in Cavallini (1993). Principal component analysis (PCA) was conducted on biomarkers of oxidative balance using a varimax rotation. This approach is well suited when there is an a priori expectation of distinct groups of variables, for example, as in markers of oxidative balance (Hõrak & Cohen 2010). The first principal component explained 41% of the total variance and was associated with variation in MDA and TAC (hereafter, PCMDA&TAC), while the second principal component explained 32% of the total variance and was associated with variation in SOD (hereafter, PCSOD). Factor loadings for the first principal component were −0·683, −0·549 and −0·481, whereas for the second principal component, they were 0·024, 0·642 and −0·767 for MDA, TAC and SOD, respectively. As MDA is a marker of oxidative damage, higher values of PCMDA&TAC indicate higher levels of oxidative stress. Similar PCAs were conducted on the relative photon catches of single cones derived from psychophysical models for bird, snake and crab, respectively; this removes absolute variation that would otherwise result in the first principal component corresponding to overall variation in photon catch values (i.e. brightness variation) (Endler & Mielke 2005). For the bird vision model, the first principal component captured 72% of the variance of single cones and was associated with variation in MW and LW versus SW and UV wavelengths, so this was used to represent variation in the type of colour (hereafter PCCOL). Factor loadings were 0·534 and 0·533, for MW and LW, and −0·444 and −0·484, for UV and SW, respectively. For the snake vision model, the first principal component captured 67% of the variance of single cones and was associated with variation in UV and SW versus LW, with the following factor loadings: 0·581, 0·409 and −0·704, for UV, SW and LW, respectively. For the crab vision model, the first principal component captured 100% of the variance and had the factor loadings: 0·707 and −0·707, for SW and LW, respectively.

GLMMs were fitted including food supply (treatment) as a fixed factor and family as a random factor unless otherwise stated. In certain analyses, SVL, SVLGR, PCMDA&TAC or PCSOD were included as covariates, in which case the interaction with food supply was also tested. To meet parametric assumptions, JND colour based on bird vision was transformed as (1/JND colour)2, and signal luminance and JND colour were log transformed when based on the snake and crab vision models. Sample sizes differed slightly between analyses because a small number of reflectance spectra showed erratic readings (i.e. consistent negative values) or bad calibration, and these values were therefore excluded from analyses. < 0·05 was considered statistically significant, and models were simplified by backward elimination starting with the interaction term(s) where appropriate. There were no significant effects of latency to lay or laying date on any of the above response terms (including all possible interactions) (all < 3·43 and > 0·13). Therefore, latency to lay and laying date are not considered further. Analyses were conducted using R v.2.12.1 (R Development Core Team 2010). All values reported in the Results are predicted means (±SE) from the statistical models, unless otherwise indicated.

Results

Survival, Development Time and Growth Rate

Some 28 of 90 larvae died before they reached metamorphosis, but the proportion of individuals that survived was similar in both food supply groups (GLMM with binomial errors: χ2 = 0·007, d.f. = 1, = 0·93). Compared with low-food individuals, high-food individuals reached metamorphosis sooner and were larger in terms of SVL and body mass (Table 2; Table S1); while relative body mass (i.e. mass controlling for size) was similar in both treatments (food, F1,8 = 7·16, = 0·028; SVL, F1,38 = 173·30, < 0·001; food x SVL, F1,37 = 0·51, = 0·48). Hereafter, we report results based on body size variation, although conclusions were qualitatively the same based on body mass variation. High-food individuals grew faster in terms of body size and body mass (Table 2; Table S1).

Table 2. Effects of food supply on development time, body size and mass, and growth rates in juveniles of the two food supply groups
 d.f. F P
Development time1,826·79<0·001
Body size (SVL)1,829·340·004
Body mass1,848·72<0·001
SVL growth rate1,823·210·001
Body mass growth rate1,829·25<0·001

Luminance, Colour and Conspicuousness

We found no statistically significant effect of food supply on overall luminance, JND luminance, colour (PCcol) or JND colour modelled based on bird vision (Table 3A). Similar results were found for models based on snake and crab vision (Table 3A).

Table 3. Effects of food supply, body size and growth rates on signal luminance and conspicuousness based on models of bird, snake and crab vision
Source of variationBird visionSnake visionCrab vision
d.f. F P d.f. F P d.f. F P
  1. General linear mixed models with food as a fixed factor and family as a random factor; boldface indicates significant values. For the bird model, colour contrast was transformed as (1/JND colour)2, and for the snake and crab, visual models luminance and JND colour were log transformed to meet parametric assumptions.

(A) Warning signalling and food supply
Luminance
Food1,80·430·531,80·350·571,80·040·84
JND luminance
Food1,81·110·321,80·850·381,80·300·60
PCCOL
Food1,80·400·541,80·070·801,80·360·56
JND colour
Food1,81·070·331,80·200·671,80·620·45
(B) Warning signalling and body size
Luminance
Food1,80·320·581,80·350·571,80·040·84
SVL1,380·850·361,380·040·831,390·070·79
Food × SVL1,385·20 0·028 1,373·480·071,383·600·06
JND luminance
Food1,81·110·321,80·850·381,80·300·60
SVL1,381·020·321,380·000·931,390·010·93
Food × SVL1,384·36 0·043 1,372·930·091,383·080·08
Colour (PCCOL)
Food1,80·400·541,80·070·801,80·360·56
SVL1,390·150·701,380·100·751,390·240·63
Food × SVL1,380·860·361,371·430·241,380·050·81
JND colour
Food1,81·070·331,80·200·671,80·630·45
SVL1,390·480·491,380·190·661,390·010·93
Food × SVL1,380·090·761,370·070·791,381·160·29
(C) Warning signalling and growth rate
Luminance
Food1,80·430·531,80·350·571,80·040·84
SVLGR1,390·330·571,380·010·921,390·000·97
Food × SVLGR1,380·470·491,370·740·401,380·770·39
JND luminance
Food1,81·110·321,80·850·381,80·300·60
SVLGR1,390·120·731,380·000·991,390·000·94
Food × SVLGR1,380·540·471,370·990·331,380·900·35

Body Size, Growth and Aposematic Signals

In the high-food group, relatively large individuals at metamorphosis had lower luminance based on the bird visual model, while the inverse (small individuals, greater luminance) occurred in the low-food group (Table 3B, Fig. 1). A similar statistical interaction was found for JND luminance (i.e. luminance conspicuousness) (Table 3B). A marginally non-significant interaction between food supply and body size was found for models based on snake and crab vision (Table 3B). In contrast, colour (PCCOL) and JND colour conspicuousness were not significantly related to body size (Table 3B). Similar results were found for snake and crab vision (Table 3B).Thus, while low-food individuals simultaneously maximized their investment in body size and signal luminance, high-food individuals that were larger had reduced luminance and conspicuousness (Fig. 1). However, signal luminance and JND luminance contrast as perceived by birds, snakes and crabs was not predicted by growth rate in terms of SVLGR (Table 3C).

Oxidative Stress and Aposematic Signals

Although growth rates were higher in the high-food group, markers of oxidative balance were similar at the treatment level in both food supply groups (Table 4). We also examined whether oxidative balance was significantly associated with variation in growth rates and whether this differed between food supply groups; however, we found no such relationships (Table S2). Yet, when oxidative stress levels were greater (high PCMDA&TAC) in the high-food treatment, signal luminance as perceived by bird, snake and crab predators was also higher, while conversely, high oxidative stress levels in the low-food group were associated with lower levels of signal luminance (Table S3A, Fig. 2). A similar result was found in terms of JND luminance conspicuousness modelled based on bird, snake and crab vision (Table S3A). There were no significant associations between signal colour (PCCOL) or JND colour conspicuousness modelled based on bird, snake and crab vision and levels of oxidative stress (Table S3A). Levels of enzymatic antioxidant activity (PCSOD) were independent of signal luminance, JND luminance, colour (PCCOL) or JND colour conspicuousness in both food groups (Table S3B).

Figure 2.

Relationship between dorsal luminance (modelled based on bird vision) and levels of oxidative stress (PCMDA&TAC) in body homogenates of juveniles of the two food supply groups. Filled circles and solid line: high-food individuals; open circles and dashed line: low-food individuals. Values on the x-axis were multiplied by −1 to facilitate interpretation, and therefore positive values denote higher values of PCMDA&TAC. Lines are predicted from GLMM analyses.

Table 4. Markers of oxidative balance (mean ± SE) in body homogenates of juveniles of the two food supply groups
 Food supply
HighLow
MDA (nmol·g−1)54·64 ± 2·4152·13 ± 3·41
SOD (U SOD·mg−1)540·23 ± 66·20545·15 ± 93·62
TAC (μmol Trolox ·g−1)0·068 ± 0·0050·060 ± 0·007

Discussion

We found that the nutritional environment experienced during early development had important consequences for body size and aposematic signal expression in green and black poison frogs at metamorphic completion. Where food was abundant, frogs grew larger, and investment in signalling and hence conspicuousness diminished. Signal luminance and levels of oxidative stress were positively correlated in high-food froglets, but were negatively correlated in low-food froglets suggesting a resource-allocation trade-off when food availability is relatively low. Resource-limited froglets appeared to simultaneously maximize investment in growth and signalling within the limits of what they could attain, as constrained by oxidative stress.

Luminance and JND luminance (i.e. conspicuousness) of metamorphic D. auratus were affected by early nutrition and its interaction with body size. There was a positive correlation between body size and warning signal luminance in the low-food group, whereas in the high-food group where froglets were relatively large on average, body size and signal luminance were negatively correlated (Fig. 1, Table 2B). In general, models based on snake and crab vision generated qualitatively similar results to those based on bird vision, with the exception that the interaction between body size and food supply in the models that considered effects on signal luminance and JND luminance (i.e. conspicuousness) was marginally non-significant. This may be explained by the fact that, unlike birds, snakes and crabs do not possess cone cells specialized in luminance sensitivity (i.e. double cones; Osorio & Vorobyev 2008), and therefore snakes and crabs may be less sensitive to differences in brightness. Some snakes in the family Colubridae have evolved resistance to amphibian chemical defences (Brodie & Brodie 1999), in particular poison frogs (Brodie & Tumbarello 1978; Santos & Cannatella 2011). Therefore, such species would not be expected to discriminate amongst prey based on visual cues that advertise the level of chemical defence. Yet, while diurnal crabs seem to be able to discriminate conspicuousness in adult Oophaga [Dendrobates] pumilio (Maan & Cummings 2012), their sensitivity to detect changes in integument luminance contrast may be constrained by their limited visual sensitivity in comparison with birds. Crabs have been reported to prey on D. auratus tadpoles (Gray & Christy 2000), which are cryptically coloured, but whether they prey on frogs post-metamorphosis is not known. Furthermore, behavioural and experimental evidence suggest that in crabs, apparent colour discrimination is mainly devoted to intraspecific communication (Detto 2007; Cummings et al. 2008; Baldwin & Johnsen 2012). In general, birds seem to be better than snakes and crabs at decoding information about the levels of defences based on the expression of aposematic signals in Oophaga [Dendrobates] pumilio (Maan & Cummings 2012). Whether this finding can be generalized to other species of dendrobatids awaits verification. Birds and snakes are candidate predators which may impose selection for diversification of defensive strategies in this group (Toledo, Ribeiro & Haddad 2007). However, more work is needed to clarify the actual predators, their vision capabilities and their role in shaping colour variation in poison frogs.

As expected, when food was non-limiting (high-food group), larvae grew faster, and they were larger at metamorphosis. As in other dendrobatids, D. auratus undergoes its larval stage inside water-filled pools, where the amount of water varies depending on rainfall (Caldwell & de Araújo 2004). Rapid development may enable individuals to avoid desiccation and predation during this vulnerable stage of life when chemical defences are lacking (Caldwell 1993). Larger froglets are likely to have greater energy in the first few, critical days following metamorphic climax, when they must begin to forage to acquire chemical defences. In addition, if body size differences persist to sexual maturity, there could be important implications for reproductive success. In particular, larger males are better able to compete for mates (Summers 1989), and larger females may benefit in terms of higher fecundity and mate guarding (Wells 1977; Summers & Earn 1999). The likelihood of surviving until maturity is probably dependent upon rapidly obtaining toxic substances after metamorphosis (Daly et al. 1994; Saporito et al. 2010). A larger froglet may have a greater foraging capacity when searching for toxic prey on the forest floor (Santos & Cannatella 2011).

While selection appears to have favoured large body size in adult poison frogs (Hagman & Forsman 2003), it is unclear how selection might shape developmental strategies in terms of relative allocations of resources to growth versus signal expression in immature, palatable D. auratus. Indeed, we found no significant effects of food supply on signal luminance, colour or corresponding conspicuousness at the treatment level. In visual-oriented predators, texture and shape discrimination of small objects seems to be mediated by luminance contrast (Jones & Osorio 2004), and thus luminance is thought to be used by birds in motion detection of prey (Osorio & Vorobyev 2005). Distance may also influence how a signal is perceived by potential predators; for example, when viewed from afar, a small target may be camouflaged; yet, when viewed in near proximity, it may be readily discriminable because of its coloration (Tullberg, Merilaita & Wiklund 2005; Bohlin, Tullberg & Merilaita 2008). During close-up inspection, a highly conspicuous signal could deter predators and thus enable individuals to forage in the open and facilitate the acquisition of toxins from the diet (Speed, Brockhurst & Ruxton 2010).

Following detection, the colour contrast of a prey may influence predator wariness and ultimately guide its decision whether to attack (Guilford 1986; Lindström, Alatalo & Mappes 1999; Osorio & Vorobyev 2005; reviewed in Stevens & Ruxton 2012). Indeed, larger body size is likely to amplify the aversive effect of the colour component of warning signals (Forsman & Merilaita 1999). This could explain why high-food froglets did not reduce investment in the colour of their warning signal. Once detected, froglets may rely on automimicry (i.e. resemblance of adults in terms of coloration) to deter predators (Brower, Brower & Corvino 1967; Speed, Ruxton & Broom 2006). Toxic adults will have already-educated predators to avoid individuals with similar appearance (Speed et al. 2012), thus allowing automimics to coexist in a population (Darst, Cummings & Cannatella 2006). It is interesting that high-food froglets did not have greater coloration than low-food froglets. This could be explained by the fact that predators are wary of novel-coloured prey (Mappes, Marples & Endler 2005), and empirical evidence suggests there is strong selection against rare phenotypes in natural populations (Noonan & Comeault 2009; Wennersten & Forsman 2009; Chouteau & Angers 2011). Alternatively, there may be stabilizing selection on coloration because of its importance in intraspecific signalling at adulthood. In dendrobatids mate selection, intraspecific competition and territorial defence are known to be influenced by variation in skin coloration (Maan & Cummings 2008; Crothers, Gering & Cummings 2011; Ostrowski & Pröhl 2011). Indeed, uniformity in dorsal skin colour and pattern within populations of the strawberry poison frog (Oophaga [Dendrobates] pumilio) has been attributed in part to sexual selection (Summers et al. 1999; Siddiqi et al. 2004; Summers, Cronin & Kennedy 2004; Reynolds & Fitzpatrick 2007; Maan & Cummings 2008).

Whether aposematic signals provide fine-scale, honest information about defensive capacity remains a matter of controversy (Stevens & Ruxton 2012). Theoretical models have predicted that more toxic prey should invest less in signalling, because they have better chances of surviving attacks and should therefore avoid the conspicuousness costs of signals (Leimar, Enquist & Sillen-Tullberg 1986; Speed & Ruxton 2005b, 2007). However, recent resource competition models have suggested that aposematic signals and defences should correlate positively under conditions where such traits utilize a shared resource that is in limited supply (Blount et al. 2009; Lee, Speed & Stephens 2011). One such resource could be antioxidants, which have been suggested to be necessary both to produce signals and to prevent oxidative stress caused by the production or storage of toxic chemicals (Blount et al. 2009). Nevertheless, these models do not take into account the fact that many aposematic species undergo a process of colour change during development (e.g. Grant 2007; Tullberg et al. 2008) and may have little or no secondary defence at this time (Nylin, Gamberale-Stille & Tullberg 2001; Nishida 2002; Saporito et al. 2010). Therefore, an interesting direction for future studies will be to investigate whether signal colour and luminance in froglets persists and how aposematic signals correlate with levels of defensive alkaloids at adulthood.

One intriguing possibility is that well-nourished larger froglets, which we found to have reduced signal luminance, are better foragers and thus become most toxic as adults. If so, this could result in a negative signal-defence correlation as some theoretical studies have predicted (Leimar, Enquist & Sillen-Tullberg 1986; Speed & Ruxton 2007; Blount et al. 2009) and as observed in some empirical studies (Darst, Cummings & Cannatella 2006; Wang 2011; Blount et al. 2012). In captivity at least, considerable sequestration of dietary alkaloids is readily achieved before adulthood in D. auratus (Daly et al. 1994). Rates of toxin accumulation may be lower in the wild, but this requires investigation. Moreover, in resource-limited environments if toxin sequestration causes oxidative stress and this trades against signal production or maintenance, as hypothesized (Blount et al. 2009), then positive signal-defence correlations may be expected at adulthood. Several recent studies have reported such correlations either within, or across, aposematic species (e.g. Summers & Clough 2001; Bezzerides et al. 2007; Cortesi & Cheney 2010; Blount et al. 2012). However, the importance of resource availability in determining the sign of signal-defence correlations has only recently begun to be studied (Blount et al. 2012), and further empirical data are needed.

Metamorphosis in amphibians is characterized by increased ROS production (Inoue et al. 2004). Indeed, significantly higher levels of lipid peroxidation and antioxidants have been found during this period in amphibians (Menon & Rozman 2007), and similarly, a positive correlation between levels of TAC and MDA has been reported previously in a study of European greenfinches Carduelis chloris (Hõrak et al. 2007). Therefore, high levels of PCMDA&TAC suggest that non-enzymatic antioxidants were accumulated and/or released in response to oxidation of cellular lipids during metamorphosis; in contrast, SOD functions largely within cells (Stead & Park 2000); therefore, its direct role in defence against lipid oxidation is unlikely. Growth is accompanied by formation of ROS as by-products of metabolism (Sies 1997), and rapid growth in particular has been linked to elevated oxidative damage or depleted antioxidant capacity in various taxa (e.g. Alonso-Alvarez et al. 2007; Nussey et al. 2009). However, we did not find any association between growth rate and levels of oxidative damage in tissue homogenates in either food supply group. Thus, individuals may have optimized their growth in relation to their antioxidant defence capability, and the dietary availability of antioxidants was sufficient in both food supply groups to enable all individuals to cope with ROS resulting from growth.

In high-food froglets, size at metamorphosis correlated negatively with luminance (Fig. 1). This reduction in luminance seems likely to have been facultative, rather than the consequence of a resource-allocation trade-off, because there was no correlation between growth rate and luminance in either food supply group. Moreover, any trade-off in the allocation of resources to warning signals versus body size would be expected to apply equally or to a greater extent in the low-food group. On the contrary, body size and luminance were positively correlated in low-food froglets. In the high-food group, higher levels of oxidative stress were associated with high luminance, whereas in the low-food group, relatively high levels of oxidative stress were associated with low luminance. We think a likely explanation for this finding is that ROS-induced oxidative stress constrained the ability to produce bright signals in the low-food group. Non-enzymatic antioxidant pigments are commonly responsible for skin pigmentation in poison frogs (Fox 1976), and oxidative stress may have depleted antioxidant pigments in low-food individuals, which in turn impaired signal production. Aposematic signal expression in particular can trade-off with growth, development time and body size (e.g. Grill & Moore 1998; Ojala, Lindström & Mappes 2007; Lindstedt et al. 2010). However, our results suggest that resource-limited individuals need not always trade-off investment in somatic versus aposematic traits; instead, it may be beneficial to become as big and bright as possible within the constraints of resource supply.

In conclusion, this study highlights the influence of developmental nutrition and oxidative stress on resource allocation to growth and aposematic signals. In particular, we found that when resources were abundant, individuals grew relatively large but reduced investment in signal luminance. These results generate predictions as to the likely importance of body size and aposematic signal expression as determinants of survival in wild froglets. Data to test these predictions are currently lacking, and this is an important topic for further work.

Acknowledgements

We are very grateful to John Endler for advice on field spectra measurements and comments on the manuscript; James Roper for comments on the manuscript; Joseph Macedonia for providing the absorptance sensitivity curves for the snake model; Tom Spande for comments on alkaloids in D. auratus; Carlos Seixas at University of Panama, Alberto Pinto, Katherine Rodriguez, Melva Olmos, Paul Budgen and Caroline Filmore and the Toribio family for assistance in the field, Rodolfo Contreras at ICGES in Panama and Christopher Mitchell for help with laboratory analyses. This work was supported by an IFARHU-SENACYT PhD scholarship awarded to EEF and by a Royal Society University Research Fellowship to JDB. MS was supported by a Biotechnology and Biological Sciences Research Council David Phillips Research Fellowship (BB/G022887/1). Scientific research permit SC/A-04-10 and CITES export permit SEX/A-63-10 were provided by The Panamanian National Authority for the Environment (ANAM), and CITES import permit 454151/01 was provided by DEFRA in the UK.

Ancillary