A parallel geographical mosaic of morphological and behavioural aposematic traits of the newt, Cynops pyrrhogaster (Urodela: Salamandridae)

Abstract

Aposematic animals advertise their unprofitability to potential predators with conspicuous coloration, occasionally in combination with other life-history traits. Theory posits that selection on functionally interrelated aposematic characters promotes the unidirectional evolution of these characters, resulting in an increase or decrease in the effectiveness of the signal. To test whether this prediction applies on a microevolutionary scale, the intra- and interpopulational variations in aposematic coloration, behaviour (which enhances the effectiveness of the coloration) and body size of newts, Cynops pyrrhogaster (Urodela: Salamandridae), were investigated. A parallel geographical mosaic of variation in aposematic coloration and behaviour among populations, independent of body size, was found. Newts on islands displayed more conspicuous aposematic traits than those on the mainland, both morphologically and behaviourally. There was no significant relationship between variation in coloration and behaviour within populations. Male newts displayed more conspicuous coloration than females. Surveys of potential predators suggest that variable natural selection at a local scale, such as predation pressure, may primarily be responsible for the microevolution of variable aposematic traits in newts.

INTRODUCTION

Aposematism has been a focus of attention for biologists since the 1800s (Darwin, 1887; Wallace, 1889), but theories regarding its origin and maintenance remain controversial (Mappes, Marples & Endler, 2005). Aposematic animals are most effectively protected when they are common because of purifying frequency-dependent selection (Greenwood, Cotton & Wilson, 1989; Lindström et al., 2001a; Ihalainen et al., 2008). Rare or new variants may suffer because they are not recognized as unprofitable prey by predators (Langham, 2004; Ihalainen et al., 2008). Therefore, the benefit of close resemblance, through the sharing of predator education, should drive unpalatable animals to display monomorphic aposematic signals both within and between species (Speed, 1999; Rowland et al., 2007). However, in nature, this is not the case: there is not only interspecific but also intraspecific variation in aposematic signals (Mallet & Joron, 1999; Ojala et al., 2005; Ojala, Lindström & Mappes, 2007).

Natural selection is an important driver of microevolution, such as intraspecific variation in aposematic coloration. Ojala et al. (2007) stated that, in order to achieve a realistic view of aposematism, we need to assess the life history of animals and consider that various selection pressures act in concert on such coloration. Despite considerable discussion and experimental study, we still have a poor understanding of how selection operates on variation in nature. The few empirical studies that have investigated natural variation in aposematic coloration suggest that intraspecific variation may occur thorough Müllerian mimicry rings, local differences in mate choice, thermal selection, differences in diet or phenotypic plasticity dependent on population density (Brown & Benson, 1974; Brakefield, 1985; Owen et al., 1994; Sword, 1999; Siddiqi et al., 2004; Ojala et al., 2007). There are inter- and intraspecific variations in a predator's ability to recognize and handle aposematic prey (Exnerovàet al., 2003, 2007), and variation in this ability is expected to be an important driver of microevolution of aposematic coloration (Mallet, 1999; Mallet & Joron, 1999; Endler & Mappes, 2004). However, no studies have considered how predation pressure operates on aposematic variation in nature, perhaps because predation on aposematic animals is rarely observed (Siddiqi et al., 2004).

One theory posits that selection on functionally interrelated characters will cause their correlated evolution (Summers & Clough, 2001). If one trait enhances a predator's ability to discriminate and learn an aposematic signal, that trait should provide an important clue to the evolution and maintenance of diversity in aposematic coloration. Unprofitable animals that have more effective aposematic coloration are expected to have been exposed to greater selective pressures, resulting in more effective enhancer traits of their coloration. Variation in a few life-history traits can affect the efficacy of aposematic coloration. These traits include toxicity (Darst & Cummings, 2006; Darst, Cummings & Cannatella, 2006), odour (Roper & Marples, 1997; Lindström, Rowe & Guilford, 2001b), aggregation (Gagliardo & Guilford, 1993; Gamberale & Tullberg, 1996a), body size (Gamberale & Tullberg, 1996b, 1998), flight behaviour (Chai & Srygley, 1990), and defensive behaviour (Johnson & Brodie, 1974). Previous studies have found cases in which conspicuousness of coloration is correlated with unprofitability (Summers & Clough, 2001; Santos, Coloma & Cannatella, 2003) and body size (Hagman & Forsman, 2003), and there is high concordance between origins of aposematic coloration and aggregation (Tullberg, 1988, 1993). These findings suggest that both aposematic coloration and enhancer traits are exposed to identical directional selection.

Correlated microevolution among aposematic coloration and enhancer traits has not, however, been described in natural populations. Thompson (1994, 2005) stated that we can only predict the overall evolutionary dynamics of species' interactions, and the organismal characters important to these interactions, when interactions at the population level are known. Ecological variables are thought to change the strength and nature of species' interactions across a wide array of spatial scales (‘geographical mosaic of selection’; Thompson, 2005). The geographical view of species' interactions predicts that organismal characters that are shaped by species' interactions will vary among populations. We can therefore see a geographical mosaic in variation of morphological or behavioural traits through species' interactions. Moreover, the mosaic of selection might form a parallel geographical mosaic of variation among functionally interrelated characters, such as aposematic coloration and enhancer traits, when both traits are exposed to the same directional selection.

Correlated microevolution between aposematic coloration and enhancer traits is found in the Japanese fire-bellied newt, Cynops pyrrhogaster. This species releases tetrodotoxin (TTX) from its skin glands (Tsuruda et al., 2002) and has a wide distribution in Japan. Its ventral coloration consists of red and black areas, and this colour is employed as an aposematic signal (Brodie, 1977). To show its ventral coloration to potential predators, the newt performs a specific display, the Unken Reflex (Brodie, 1977; Fig. S2, see Supporting Information), which enhances the effectiveness of the aposematic ventral coloration (Johnson & Brodie, 1974). There is distinctive local variation in the ventral colour patterns (Sawada, 1963). This red colour derives from carotenoid pigments that must be obtained from prey animals, as the newts cannot synthesize carotenoids themselves (Goodwin, 1986; Matsui, Marunouchi & Nakamura, 2002). As a result, the aposematic coloration of this newt might have been influenced by selection from both predation and carotenoid–prey resource availability. Furthermore, the ventral colour pattern of these newts does not seem to be under sexual or thermal selection, because this species is primarily nocturnal and does not use the ventral coloration in mating (Tsutsui, 1931). There is no significant difference in the ventral colour pattern between newts from high- and low-density populations (K. Mochida, unpubl. data), and no animals are known to mimic the ventral coloration of this newt.

In this study, the aposematic coloration of 700 newts from 21 populations in western Kyushu, Japan was investigated, and a geographical mosaic of variation in coloration was described. It was predicted that a mosaic in aposematic behaviour (an enhancer trait) that parallelled the variation in ventral coloration would be found. Furthermore, it was hypothesized that these two traits might be exposed to identical directional selection, which might either increase or decrease the effectiveness of the signal. To test this hypothesis, the inter- and intrapopulational variations in the tendency of newts to perform the Unken Reflex were analysed. Finally, it was discussed how natural selection might operate on a parallel mosaic of geographical variation in aposematic coloration and behaviour. Although there are many nonexclusive explanations, a combination of field censuses and a literature survey of potential predators indicates that geographical variability in predation pressure may have influenced the evolution of these aposematic traits. This study provides the first example of a parallel geographical mosaic in the variation of an aposematic trait and an enhancer trait within a species.

MATERIAL AND METHODS

ANIMAL SAMPLING

Cynops pyrrhogaster is distributed on mainland Japan and several adjacent islands. To analyse the geographical variation in aposematic coloration, behaviour and body size (snout-to-vent length, SVL), 700 animals from 21 populations (Fig. 1) in western Kyushu were collected between 2004 and 2007. All animals were adults found in rice paddy fields. Geographical variation in the aposematic coloration, behaviour and SVL were examined in animals from 19 of the 21 populations. Animals from two populations (sites 9 and 13) were examined only for coloration and SVL. Eighty animals from eight populations (sites 1, 2, 4, 6, 8, 10, 11 and 12) were selected for detailed ventral colour pattern analysis. These eight populations were selected to reflect the range of geographical variation in the extent of red coloration.

APOSEMATIC COLORATION AND DEFENSIVE BEHAVIOUR

The aposematic ventral coloration of newts consists of a complex of black and red regions (Sawada, 1963). This colour pattern is heritable and varies very little with growth after sexual maturity, although it is not completely invariable (K. Mochida, unpubl. data). To analyse the inter- and intrapopulational variation in this colour pattern, photographs were taken in the laboratory of the ventral surface of 700 newts using a Nikon digital camera (model E2100). The relative extent of the red area in the ventral colour pattern, ranging from snout to vent and excluding the limbs, was measured using NIH Image 1.63 (available online from the US National Institutes of Health; http://www.nih.gov/).

The colour patterns of 80 newts from eight populations were analysed in greater detail by recording the size and number of black spots. In four of the eight populations, the distribution of black spots was also analysed. To compare the overlapping area of colour patterns among individuals within a population, photographs were imported into Adobe Photoshop CS 8.0 (Adobe Systems Inc.). The photographs were fitted into a generalized body outline that was drawn on the computer, and were resized on the basis of several points on the generalized body outline (tip of snout and base of skull, four limbs, and vent). All black spots of the adjusted photograph were arranged on the entirely red body outline (see Fig. S1). Forty-five pairwise comparisons were then made among individuals from each of the four populations. Pairs of the shape-adjusted figures were overlain digitally, and the overlapping red and black area of the two newts was measured with NIH Image 1.63. The average overlap for each of the four populations was then calculated. The average was used as an index of resemblance of the colour patterns within each population.

Laboratory behavioural trials were started 1 week after each newt was captured. Two trials were conducted for each newt, with an interval of 1 week between trials. Before a trial, the newt was kept at a stable temperature (4 °C) for 48 h to ensure that it would elicit a behavioural response efficiently (T. Hayashi, pers. comm.). In each trial, a predator stimulus was administered by grasping a newt around the midtrunk with forceps, lifting it to 15 cm, and then dropping it onto wet paper. This process was repeated three times at 15 s intervals. Both C. pyrrhogaster and Notophthalmus viridescens have been shown to display the Unken Reflex frequently in response to this stimulus (Ducey & Dulkiewicz, 1994; K. Mochida, unpubl. data). The reaction displayed was considered to be the Unken Reflex if the newt maintained a rigid posture, exposing the ventral surface between the chin and tail for more than 5 s (Brodie, 1977; Fig. S2). Newts that showed the Unken Reflex to an earlier stimulus also displayed to a later stimulus within the same trial. The tendency to perform the Unken Reflex was recorded as a score: if a newt showed the Unken Reflex for the first time against the first, second or third stimulus, the score was ranked three, two or one, respectively. If a newt did not respond to the three stimuli, the score was zero. In the statistical analyses, the total score of the two trials for each animal was used.

LOCAL POTENTIAL PREDATOR FAUNA

Because predation pressure affects the maintenance of aposematic signals, field censuses of potential predator faunas were conducted from 2006 to 2007 at the eight sites examined in the detailed coloration study above. Predation on other amphibian prey was occasionally observed during this study, but not on newts. However, some studies have described animals attacking adult newts (Table S2, see Supporting Information). These animals are also known to feed on other amphibians. Therefore, potential predators were defined as not only those that have been observed to attack adult newts, but also those that feed on amphibians of a similar size. Potential predators were divided into three groups. The first are potential avian predators, against which aposematic coloration is likely to be most effective because they are visually oriented hunters. For this group, route censuses of the animals were conducted along a stream adjoining the newts' paddy fields to examine the species' richness and biomass of avian predators at each site. The census routes were 50 m wide and 2500 m long. Censuses were conducted in June and October, during which the sites were surveyed three times, at daybreak, noon and evening; and also in February, when the censuses were conducted twice, at daybreak and noon. The second group are potential mammalian predators. Colour signals are less effective or completely ineffective against them, as they are colour blind. Snakes are the third group of potential predators and generally are not colour-oriented hunters, although some snakes have colour vision (Repérant et al., 1992). In the last two groups, only the species richness was surveyed at each site, using the literature and, for mammals, field censuses searching for excrement and tracks. The census methods were the same as for the avian censuses.

STATISTICS

R 2.5 was used for generalized linear models (GLMs) and generalized linear mixed models (GLMMs), and JMP 5.1 (SAS Institute Inc.) for other statistical analyses. A pattern artefact was predicted in which the extent of the red area was negatively correlated with the size or number of black spots on the ventral surface, and this prediction was tested for eight sites. Interpopulational variation in SVL was analysed separately for each sex because of sexual differences in body size. The correlation between the interpopulational variation in SVL of males and females was then examined. Because there was a significant correlation between geographical variation in male and female body size (see Results), the standardized SVL (measured SVL/mean SVL of all male or female newts) was used for statistical analyses. The mean SVLs of male and female newts are 47.76 and 55.32 mm, respectively.

To examine the relationship of coloration among populations with the behaviour and standardized SVL, the relative extent of the red area on the ventral surface was fitted as a response term, and the behavioural score and standardized SVL as explanatory terms, in a GLM. In this model, the effect of the sex ratio of newts in each sample was also examined, as there was a slight sexual difference in coloration (see Results). The population averages of each value were used as response and explanatory terms.

In the analyses of intrapopulational variation, the relationship between coloration and behaviour, standardized SVL, and sex was also examined using GLMMs. In this model, the relative extent of the red area was fitted as a response term, the behavioural score, standardized SVL and sex as explanatory terms, and the identity of the population as a random term. This analysis uses the restricted maximum likelihood model (REML) to decompose the variances and to derive parameter estimates.

In the eight sites in which censuses of the local predator fauna were conducted, the factors correlating with geographical variation in aposematic coloration were tested. These factors included species' richness and biomass of avian predators in each season, species' richness of mammalian and snake predators, and temperature. The biomass of avian predators was calculated as the sum of the biomass of each bird species. Each species' biomass was calculated by multiplying the weight of one individual by the largest number of individuals counted within two or three censuses in one season. The relationship between topographical characters, such as altitude, latitude or distance from the mainland to the sampling island, and the geographical variation in coloration was also analysed. Distance was converted to a logarithm. If a site was located on the mainland, it was given a distance of zero. Because a significant correlation was detected between variation in coloration and the distance from the mainland to the island (see Results), GLMs were used to analyse the effect of distance on coloration, behaviour and standardized SVL among all samples from 21 sites.

RESULTS

There was significant interpopulational variation in the relative extent of the red area in aposematic coloration (one-way Welch: F_20,198 = 53.97, P < 0.0001), and it seems to form a geographical mosaic (Fig. 1). There were significant differences in the size and number of black spots among the eight populations analysed (size: one-way Welch: F_7,54 = 30.07, P < 0.0001; number: one-way Welch: F_7,30 = 3.53, P = 0.0019). Newts from some populations, such as Fukue (site 1), displayed an almost entirely red ventral surface without black spots or with very small black spots. The size of the spots on the newts from Fukue was 0.06 ± 0.04% [mean ± standard error (SE)] of the ventral surface (N = 10; spot number, 43.9 ± 14.05). In contrast, populations, such as Isahaya (site 10), had a comparatively low percentage of red on the ventral surface. The size of black spots in this population was 0.82 ± 0.20% of the ventral surface (N = 10; total spot number, 20.6 ± 5.11). Interpopulational comparisons of the extent of red area, black spot size and spot number revealed a significant correlation only between the extent of red area and spot size (Pearson's partial correlation coefficient: r = -0.90, P = 0.0021) and none between the extent of red area and spot number (Pearson's partial correlation coefficient: r = 0.45, P = 0.2680). The relative extent of red area and the resemblance index of the colour pattern were highly correlated, and this approached significance (Spearman's coefficient of concordance: r = 1.00, P = 0.05).

Newts from different populations were significantly different in body size (males: one-way Welch: F_20,102 = 8.21, P < 0.0001; females: one-way Welch: F_20,68 = 7.13, P < 0.0001). Male and female geographical variation in body size correlated significantly (Pearson's correlation coefficient: r = 0.76, P = 0.0001). Using 520 animals from 19 populations, significant interpopulational variation was also found in the tendency to perform the Unken Reflex (Kruskal–Wallis: ?² = 83.00, P < 0.0001).

The interpopulational variation in the relative extent of red area on the ventral colour pattern was adequately explained by three terms: behavioural score, standardized SVL and sampled sex ratio (R² = 0.38). In GLM, only the behavioural score was significant as an explanatory term (behaviour: Coef = 5.51, t = 3.48, P = 0.0034; standardized SVL: Coef = -0.11, t = -0.30, P > 0.05; sex ratio: Coef = -12.67, t = -1.30, P > 0.05). There was a significant positive correlation between the relative extent of red area and the behavioural score among populations (Pearson's correlation coefficient: r = 0.64, P = 0.0029) (Fig. 2).

GLMM was used to examine the relationship between variation in coloration, and the behaviour score, standardized SVL and sex within populations. In this model, only sex was significant as an explanatory term (behaviour: Coef = 0.24, t = 1.38, P > 0.05; standardized SVL: Coef = -0.09, t = -1.28, P > 0.05; sex: Coef = 3.99, t = 5.24, P < 0.0001). Male newts displayed a larger extent of red area on their ventral colour pattern than did female newts.

In eight sites, a significant correlation was found between the variation in coloration and the species' richness of mammalian potential predators (Spearman's coefficient of concordance: r = -0.89, P < 0.005). The other factors examined with regard to the potential predator fauna and temperature were not significantly related to the variation in coloration (Tables S3, S4, see Supporting Information). Only distance from the census site to the mainland was correlated with both a geographical variation in coloration (Spearman's coefficient of concordance: r = 0.94, P < 0.0025) and the species' richness of mammals (Spearman's coefficient of concordance: r = -0.95, P < 0.001) (Table S4). Using GLM, the correlated trend between coloration and distance to the mainland was confirmed employing data from 21 populations (R² = 0.48, Coef = 2.42, t = 4.41, P = 0.0003). In another GLM, it was adequately fitted when the behavioural score was adopted as a response term and the distance from the mainland (Coef = 0.21, t = 2.65, P = 0.017) as an explanatory term (R² = 0.29). However, the distance from the mainland did not explain the variation in standardized SVL (Coef = -0.18, t = -0.47, P > 0.05). In summary, it was found that newts on islands displayed a larger extent of red area on their ventral surface and showed a higher tendency to perform the Unken Reflex than did those on the mainland. These tendencies were more distinctive the more distant the populations from the mainland.

DISCUSSION

The aposematic ventral coloration of the newt C. pyrrhogaster consists of three features: the size, number and distribution of black spots against the red ventral colour. Results from eight populations showed that the relative extent of the red area on the ventral surface increased with decreasing size of the black spots and was independent of the number of spots. Although not specifically examined, it appeared that almost all 21 populations examined here conformed to this tendency. A positive correlation was found between the average relative extent of the red area and the overlapped red and black area of the colour pattern. This suggests that a decrease in the size of the black spots led to an increase in the extent of red area on the venter and formed a more uniform red colour pattern.

Newts usually hide their aposematic ventral coloration and display it with the Unken Reflex when a potential predator attacks. Aposematic signals are expected to work best if they have high contrast within the animal (Endler & Mappes, 2004). When the newt performs the aposematic behaviour, its dorsal black coloration will highlight its ventral red coloration, especially if the newt has almost entirely red ventral coloration. However, if the ventral surface of a newt is almost entirely black, the efficacy of the signal will be limited, at least as measured for human vision (Fig. S2).

Xanthophores, which contain concentrations of carotenoid granules, are located throughout the ventral surface of the newts, but are rare in black areas (Matsui et al., 2002). Larger numbers of xanthophores are present when the relative extent of the red area is greater. The hue of the red coloration on the ventral surface is closely related to the number of carotenoid granules and the content of carotenoids in each granule (Matsui et al., 2002). If the amount of carotenoids possessed by a newt is low, its ventral coloration will consist of black and off-white or dull yellow regions (Matsui et al., 2002). To display the conspicuous red coloration, the newt must obtain high concentrations of carotenoids by consuming carotenoid-containing prey, as newts cannot synthesize carotenoids themselves (Goodwin, 1986). Apart from their use as signals, carotenoids have been demonstrated to function in general immunity in various organisms (Olson & Owens, 1998). The immunological benefit of carotenoids might cause a scarcity of carotenoids in the newts as a signal resource (Olson & Owens, 1998). Thus, carotenoid-based signals are implicitly costly; therefore, the greater the extent of the red area, the higher the costs incurred by the individual. There may be a trade-off between the benefits of conspicuousness against predators and the costs of obtaining carotenoid sources for aposematic coloration.

This study unexpectedly found a sexual difference in aposematic coloration. The relative extent of the red area of the aposematic colour pattern of males was larger than that of females. In the analyses, relative values were compared as a measure of signal size, but Ojala et al. (2007) stated that absolute signal size is important for aposematism. Even in terms of absolute signal size, males displayed larger signals than females (there was an extra model in GLMM that included body size as a covariate: t = 5.501, P < 0.0001). In addition to displaying carotenoid signals, females divert carotenoids into eggs and for other sex-specific reproductive functions (Hill, 1999). Because females invest carotenoids in their offspring, they will have lower levels of carotenoids available for aposematic signalling, which may explain the sexual differences in aposematic coloration found here.

Apart from balancing the costs and benefits of carotenoid use at each locality, other factors may be important in determining the interpopulational variation in aposematic coloration. Newts do not display their ventral coloration in mating, and the courtship sequence takes place on the bottom of water bodies (Tsutsui, 1931). Therefore, the ventral coloration of newts would not be exposed to sexual selection. As newts are ectothermic organisms, their body temperature and activity may depend on their coloration (e.g. Brakefield, 1985). However, newts are primarily nocturnal, and their ventral coloration is not exposed to daylight. In spring, newts can be seen during the day, but they are still predominantly active at night. The present study also found no significant relationship between the ventral coloration and ambient temperature. Therefore, it seems that thermal selective pressure does not act on the ventral colour pattern, although there are no direct data comparing activity among differently coloured newts.

Experimental studies have revealed that body size may affect variation in aposematic coloration (Gamberale & Tullberg, 1996b, 1998). However, no significant relationship was found between the relative extent of the red area of ventral coloration and the standardized SVL. This result suggests that the body size of C. pyrrhogaster does not affect the differences in the coloration detected in the present study.

In poison frogs (Dendrobatidae), toxicity is linked to variation in aposematic coloration (Darst & Cummings, 2006; Darst et al., 2006). Tsuruda (2001) investigated interpopulational variation in toxicity of C. pyrrhogaster in northwestern Kyushu. Although many of her study areas overlapped with those of this study, the variation in toxicity found did not match the geographical variation in aposematic coloration (cf. Fig. 1 with Tsuruda, 2001). Although the ultimate source of the toxin of newts remains unknown (Hanifin, Brodie & Brodie, 2002; Cardall et al., 2004; Lehman, Brodie & Brodie, 2004), it is possible that TTX is sequestered from dietary sources in C. pyrrhogaster (Tsuruda et al., 2002). If so, the toxicity of C. pyrrhogaster would be dependent on the local prey base (Tsuruda, 2001; Tsuruda et al., 2002).

In addition to natural selection, stochastic events are important in determining geographical variation in a trait. Small population sizes and founder effects associated with island invasion may lead to genetic drift, leading to new traits that are selectively neutral (Mayr, 1954). However, it is unlikely that the observed geographical variation was established solely by stochastic events for three reasons. First, aposematic coloration of newts is not considered to be a neutral character because of its presumed significance in predator–prey interactions (discussed above). Second, the unidirectional variation in coloration among newts on islands, which have evolved to exhibit more conspicuous coloration than newts on the mainland, is inconsistent with stochastic events. Stochastic changes should lead to a random distribution of geographical variation among both island and mainland populations. Third, newts from islands 17 and 18 show polymorphism of loci and heterozygosity of allozymes at a level similar to those of mainland populations (Hayashi & Matsui, 1988), indicating that there is no founder effect in these populations.

This study supports the prediction that geographical variation in aposematic behaviour parallels aposematic coloration (Fig. 2). This prediction posits that both traits were exposed to identical directional selection. However, the ‘genetic coupling hypothesis’ (Brodie, 1989) could also explain the correlated geographical variation. If two traits are genetically coupled, the evolution of one trait may not be independent of the other (Falconer, 1981). This can explain nonadaptive differences in one trait through indirect selection on correlated characters (Brodie, 1989, 1992). A significant correlation between the two traits among populations, but no significant relationship within populations, suggests that the observed variation does not fit the genetic coupling hypothesis. Although there are other nonexclusive explanations, the results indicate that variable predation pressure or prey limitation may have played a leading role in the evolution of parallel geographical variation between aposematic coloration and behaviour, and that both traits may be exposed to identical directional selection, resulting in increasing or decreasing efficacy of the signal.

Although the surveys of the predator fauna at each location were incomplete, they suggest the type of geographical variation or ‘geographical mosaic in selection pressure’ (Thompson, 2005) that may exist in this system. In this study, there were no significant differences in species' richness and biomass of avian predators between islands and the mainland, but the species' richness of mammalian predators was lower on islands located more distant from the mainland (Table S3). This is expected on the basis of Island Biogeography Theory (MacAthur & Wilson 1967). A significant correlation was also found between interpopulational variation in aposematic traits and geographical variability in the species' richness of mammals. Most mammalian predators of amphibians are colour-blind, which renders colour signals less effective. However, avian predators are visually oriented, and such colour signals may be effective against them. These differences in visual ability among predators and the relative importance of different predators in each location would affect the evolution of weak aposematic signals at a local scale (Endler & Mappes, 2004). It may be advantageous for newts on islands rather than those on the mainland, which are under threat from both colour-blind and colour-oriented predators, to devote their carotenoid resources to producing more aposematic coloration. Because carotenoid coloration is costly to produce (Goodwin, 1986; Matsui et al., 2002), newts on the mainland might devote their carotenoid resources to other life-history traits.

However, the rigid immobile aposematic display that enhances the effectiveness of aposematic coloration is a reflex movement (Johnson & Brodie, 1974). Newts may perform such displays more often against potential avian predators to avoid misrecognition of the signal. If newts cannot distinguish between avian and mammalian predatory stimuli and respond inappropriately, the immobile display may result in death for the newts if they display to a colour-blind predator. This fatal mistake was often observed in laboratory trials; newts that reacted to a mammalian predator with the immobile display lost the opportunity to escape and were killed (K. Mochida and K. Matsui, unpubl. data). Therefore, it is predicted that newts on the mainland perform rigid, immobile displays at a lower frequency against potential predators than do those on islands, even if the predators are avian and the effectiveness of the aposematic signal is reduced against them. Thus, although mammalian predators are not the primary target receiver for aposematic signals, geographical variability in predation by mammals may indirectly alter the strength of local interactions between avian predators, which are the primary target of these signals, and newts.

ACKNOWLEDGEMENTS

I am grateful to A. Mori for his helpful comments on this study and for critical review of the manuscript, and to S. M. Barribeau, S. Cook and A. Savitzky for their comments on drafts of the manuscript. I thank O. Arakawa, T. Hayashi, M. Imafuku and K. Matsui for productive discussions and two anonymous reviewers for their constructive comments that improved the manuscript. T. Arioka, K. Ikeda, H. Kaneda, M. Kitada, D. Muramatsu and K. Tanaka helped with this work. This work was financially supported in part by the JSPS Research Fellowships for Young Scientists and the Global COE Program A06 to Kyoto University.

REFERENCES