Developmental costs of yellow colouration in fire salamanders and experiments to test the efficiency of yellow as a warning colouration


Warning colouration reduces predation risk by signalling or mimicking the unpleasantness of prey and therefore increases survival. We tested in two experiments the evolutionary costs and benefits of the yellow colour pattern in fire salamanders (Salamandra salamandra), which display a yellow/black colour pattern usually associated with toxic alkaloids. Our first experiment aimed to test whether the development of colouration is condition dependent and thus related to developmental costs, i.e. influenced by resource availability during the developmental process. Therefore, we reared fire salamander larvae under different nutritional conditions and compared the relative amount of yellow they developed after metamorphosis. Fire salamander larvae reared under limited food conditions had a lower proportion of yellow following metamorphosis than control larvae reared under superior food conditions. In a second experiment we tested whether the proportion of yellow has an impact on the risk of being attacked using artificial models. We tested, in salamander-free and salamander-occupied natural habitats, whether artificial clay models with different proportions of yellow and black receive different attack rates from potential predators (birds, mammals, insects). In clay models the proportion of yellow and the site had a significant effect on predation risk. Models with larger amounts of yellow had fewer bite marks from predators such as carabid beetles and birds, but only in sympatry with salamanders. In conclusion, the early expression of conspicuous colouration seems to be condition dependent and therefore potentially costly. Furthermore, the yellow colouration of fire salamanders act as a signal that potentially reduces their risk of being attacked by predators. Thus, the yellow colouration of fire salamanders seems to represent an adaptive trait that reduces the risk of predation, which can be expressed in higher quantity by individuals of a certain condition.


Background
Predation is one of the strongest evolutionary forces driven by natural selection resulting in many adaptations displayed by predator and prey species. One example of such an adaptation is aposematic or warning colouration (Poulton, 1890;Cuthill et al., 2017) by which individuals signal secondary defences such as toxicity or unpalatability (Summers and Clough, 2001;Saporito et al., 2007;Stevens and Ruxton, 2012). Various examples have shown that warning colouration deters potential predators and therefore increases survival of potential prey species (Arenas, Walter and Stevens, 2015). Assuming that conspicuousness correlates with the likelihood of being detected by a predator, conspicuous colouration might act as a "handicap" being costly to produce (Zahavi, 1977). Empirical evidence has shown that colour saturation is an honest signal and correlates with toxicity (Blount et al., 2009(Blount et al., , 2012Arenas et al., 2015; but see Crothers et al., 2016 for a negative correlation of colour reflectance, i.e., brightness and toxicity). Apart from colour saturation, the conspicuousness of a warning signal -expressed by the amount of conspicuous colour (Endler, 1978) -could also function as an honest signal. However, empirical evidence for this hypothesis is still lacking.
Warning colouration in animals often involves the colours red, yellow and black (Stevens and Ruxton, 2012). In combination with being poisonous, fire salamanders (Salamandra salamandra) have conspicuous colouration, which is essentially composed of yellow and black and is therefore considered a classic example of an aposematic species (Schuler and Hesse, 1985). All four species of fire salamanders are poisonous and produce steroid alkaloid toxins in granular glands of their skin (reviewed in Lüddecke et al., 2018) that can be secreted or even sprayed when being attacked by predators (Brodie and Smatresk, 1990;Mebs and Pogoda, 2005). Patterns and shapes of the yellow-black colouration differ between the species and subpopulations of fire salamanders and across their distribution range in Europe (Salamandra salamandra), the Near East (S. infraimmaculata), North Africa (S. algira) and Corsica (S. corsica) (Eiselt, 1958;Thiesmeier, 2004).
Most species with warning colouration develop their colouration during early ontogeny (Booth, 1990;Grant, 2007). Aposematically coloured butterflies, such as the European swallowtail Papilio machaon, have conspicuously coloured larvae and adults, but cryptic pupae (Wiklund and Sillén-Tullberg, 1985). The same is true for several aposematically coloured amphibians, such as poison frogs (Saporito et al., 2007) or fire salamanders, in which individuals in the larval stage do not show any conspicuous colouration until the start of the metamorphosis process. In some populations of fire salamanders females use temporary ponds as well as first order streams to deposit their larvae (Weitere et al., 2004;Caspers et al., 2014). During larval development, the yellow-black colouration begins to be expressed once a larva enters metamorphosis, which also marks the transition to the terrestrial phase of the life cycle. After metamorphosis, the juveniles begin to completely adopt a terrestrial lifestyle, and the yellow-black colouration is fully expressed. It remains unknown whether fire salamanders become toxic once they start to express their aposematic colouration or even before. During the terrestrial life cycle, the risk of predation is highest shortly after metamorphosis, with birds and carabid beetles being the main known predators (Thiesmeier, 2004). Once the fire salamanders have grown to adult size, predation is almost negligible (Thiesmeier, 2004), which may be a result of increasing toxicity with increasing size/age. Accordingly, natural selection may act most strongly on the phenotypic expression of colouration shortly after metamorphosis.
Our first experiment (experiment 1) aimed to understand the selective forces that shape different phenotypes after metamorphosis. Larval development in fire salamanders is highly influenced by several abiotic and biotic factors, such as water temperature, oxygen levels, predation pressure or food availability (Zakrzewski, 1987;Reinhardt et al., 2013;Reinhardt, 2014;Reinhardt, Steinfartz and Weitere, 2015). Food availability appears to be a major difference between salamanders from different larval habitat types (Weitere et al., 2004;Steinfartz, Weitere and Tautz, 2007). Whereas food availability in first order streams, for example, is sufficient, potential food items in temporal ponds decrease over time (Reinhardt, 2014). Larvae reared under high nutritional conditions, i.e., with high quantities of food, showed behavioural and body condition differences compared to those reared under low nutritional conditions (Krause, Steinfartz and Caspers, 2011), and these differences also remained in individuals after metamorphosis (Krause and Caspers, 2016) despite identical post-metamorphic diets. Here, we analysed whether nutritional treatment during larval development also persists and influences the colour phenotype of fire salamanders shortly after metamorphosis. Specifically, we tested whether juvenile terrestrial salamanders that were reared as larvae under superior nutritional conditions were more efficient in allocating resources to exhibit a more conspicuous phenotype, i.e., a higher proportion of yellow, than salamanders that were exposed to low nutritional conditions during the larval stage.
The conspicuousness of colour patterns has been demonstrated to influence predator-prey interactions (Stevens and Ruxton, 2012;Adams, 2015, 2016;Rößler et al., 2019). Conspicuousness increases avoidance learning of predators (Darst and Cummings, 2006) and decreases the predation risk (Forsman and Merilaita, 1999;Riipi et al., 2001;Rößler et al., 2019). These findings led us to the hypothesis that individuals with a larger amount of yellow coloration might have a higher survival by decreasing their predation risk. To test this prediction, we performed a second experiment (experiment 2) using artificial clay models of black and yellow, differing in the relative proportion of yellow and black. By documenting the attacks of different predators, we investigated whether an increase in the amount of yellow decreases predation risk. With the combination of these two experiments, we aimed to better understand i) potential developmental constraints in the expression of yellow colouration in fire salamanders and ii) potential functions of yellow for the survival chances of young terrestrial fire salamanders.

Study organism
Similar to most amphibians, European fire salamanders (Salamandra salamandra) are aquatic during their larval stage. Once metamorphosed, they are completely terrestrial and females only return to first order streams or small ponds for larval deposition (Thiesmeier, 2004). After metamorphosis, S. salamandra typically exhibits a colour phenotype with yellow stripes and dots on a uniform black background. The age at metamorphosis can range from several weeks to several months. The age and the weight/size at metamorphosis is strongly dependent on the larval habitat, i.e., pond or stream (Reinhardt, 2014). The typical yellow-black colour pattern begins to be expressed when the larvae show clear signs of metamorphosis (e.g., reduction of gills).

Experiment 1 -effects of early developmental conditions on colouration
Nutritional treatment during larval phase. We collected potentially pregnant female fire salamanders on their way to their deposition habitats in the Kottenforst southwest of Bonn in western Germany (coordinates: 50°39 45.07 N, 7°4 17.85 E) on rainy nights in the spring of 2010 (Krause et al., 2013;Caspers et al., 2014), with the permission of the nature reserve authority of the city of Bonn. The females were kept individually in plastic fauna boxes to enable larviposition as described in Caspers et al. (2014). After larviposition, two sibling larvae from each of 12 females were randomly chosen and assigned to one of the two early nutritional treatment groups (Krause, Steinfartz and Caspers, 2011). Each larva was kept individually in a small plastic cup. The two nutritional treatment groups differed only in the amount of food (Chironimus larvae) offered, i.e., the superior group was fed six days a week, whereas the individuals of the other group received food only twice a week. Thus, the latter group had to cope with limited food conditions (for a detailed description of the rearing conditions, see Krause, Steinfartz and Caspers, 2011). Independent of the treatment groups, fire salamander larvae ate all Chironimus larvae within 24 hours. After metamorphosis, all of the individuals received intermediate dietary conditions for the rest of their lives, all individuals received a diet based on crickets and earthworms (Krause, Steinfartz and Caspers, 2011). Three larvae died before metamorphosis, two from the superior treatment and one from the limited food conditions, resulting in a total of 9 complete sibling pairs reaching metamorphosis.

Analysis of yellow and black proportion in colouration.
To quantify the amount of yellow on the skin surface, a picture of each individual was taken 50 and 100 days after the day we have seen it the first time on land (i.e., after completing metamorphosis) using a digital camera (Panasonic Lumix DMC-FZ50). Each individual was photographed on a white background from above (supplementary fig. S1). The pictures were analysed using computer-aided software (Sanchez et al., 2018), an add-on for the software Neurocheck (NeuroCheck GmbH, Remseck, Germany). In simple terms, for each individual, the proportion of yellow-toblack colouration was analysed. Therefore, the number of yellow pixels and black pixels were measured. The resulting proportion was calculated as the number of yellow pixels divided by the number of yellow plus black pixels (see supplementary table S1).

Experiment 2 -artificial model preparation and experimental design
This experiment aimed to understand potential consequences of higher amounts of yellow colouration, as found in the first experiment, on the attack rate by different potential natural predators (birds, rodents and beetles), as this may be an indicator of survival chances of young terrestrial S. salamandra. To examine this, we used artificial clay models consisting of yellow and black modelling clay (Modello, Lyra, Germany) as it is not possible for ethical reasons to perform predator attack trials with living salamanders. To test whether the clay models are similar in color characteristics to the fire salamanders, we made pictures (652 px × 490 px) of three different sets of clay models used in the field and two different salamanders using a color (mvBlueCOUGAR-S120aC) and a monochrom (In-Sigth 5000) camera with a FUJINON 16 mm lens and a Balluff 100 mm ringlight. The RGB color characteristics of pictures were afterwards analyzed with the program NeuroCheck 6.0.1. Clay models and fire salamanders had similar reflectance patterns and thus were considered to be similar in colour. However, it is important to mention that our system measured reflectance patterns not in the UV range. Thus, as carabid beetles and birds have UV vision, the clay models and salamanders might be perceived very differently by the potential predators. For future studies it is important to measure reflectance patterns also into UV light. The artificial models had a cylindrical shape with a 1 cm (±0.1 cm) diameter and a length of 5 cm (±0.2 cm) ( fig. 1), roughly mimicking the cylindrical body shape and size of young salamanders. We used models that were relatively simple and efficient to construct because we were interested primarily in testing the effect of colour. Even if we could have made models with a more realistic shape, we could not have added movement, which is probably the more relevant predator stimulus. Thus, we tested the effect of yellow colouration on predation risk per se, and not explicitly the effect of yellow colouration on predation risk of young fire salamanders. To avoid displacement of the models by the wind, black wooden sticks with a length of 4.5 cm were used as legs. One set contained four clay models, each with a different proportion of black and yellow clay. Each clay model consisted of five variable segments. Two of the four models were plain-coloured in black or yellow. The remaining two models of a set consisted of either two yellow and three black segments or two black and three yellow segments ( fig. 1A, B). The sequence of the two mixed models were Y-B-Y-Y-B and B-Y-B-B-Y, respectively (Y = yellow, B = black). This way we kept the contrast of the two mixed artificial models the same, while varying the amount of yellow.
Different runs of this experimental setup were performed by students in a practical student's course on proximate and ultimate aspects of behaviour in the summer terms of 2012 and 2013 under our intense supervision (B.A.C, I.H. & S.S). The artificial clay models were set up at two different sites, which differed in the presence/absence of salamanders. One site (Krebsbachtal, Teutoburger Wald, Bielefeld, Germany N52°01.200 , E008°26.000 ) was inhabited by fire salamanders, whereas the other site (the garden of the Animal Behaviour Department of the University of Bielefeld; N52°2.125 , E008°29.780 ) is in 3500 m distance from the first site and not inhabited by fire salamanders (hereafter the "salamander-free habitat"). We are aware that the two sites might also differ in other aspects and a samples size of two is far from being ideal to investigate one factor that differs among the two sites (Hurlbert, 1984). In June 2012, a total of 320 models were set up for four days in a cross formation ( fig. 1A) with the plain-coloured models placed on opposite sides. The mixed artificial models were also placed opposite of one another with a 90°-shift so the outer segments were never the same colour. The distance between two oppositely placed models ranged from 30 to 80 cm, and the distance between two sets was irregular and ranged from 5 to 20 m. One hundred and ninety-six models were set up in the salamander habitat, and 124 models were set up in the salamander-free habitat.
In June 2013, another 180 clay models were set up again for four days as described above, except that they were not set up in a cross formation, but instead individually on a transect with a minimum distance of two metres between two adjacent artificial clay models. The sequences of the artificial clay model type were randomized within each set to guarantee that our results from 2012 were not driven by a higher visibility of the clay models because of the setup design (Rowland et al., 2008). In 2013, we set up 120 clay models in the salamander site and 60 clay models in the salamander-free site.

Identification and analysis of bite marks
To quantify whether the amount of yellow colouration influences the risk of being attacked, we collected each artificial clay model after exposure and classified and counted all bite marks. To improve the identification quality of the encountered bite marks on the models, we compared them with those obtained from potential prey species under captive conditions. In cooperation with the local zoo (Tierpark Olderdissen in Bielefeld, Germany), we laid out a set of clay models in Rattus rattus and Mus musculus cages as representative rodents and Corvus corax aviaries as the representative group of birds. Captured Carabus problematicus and Pterostichus burmeisteri kept in terraria were used as representatives for the carabid beetles. In all cases at least some individuals attacked the clay models, and bite marks could be recorded as a reference ( fig. 1C-H). Those reference bite marks were used to characterize the predator type that attacked the artificial clay models in our experimental setup.
After we collected the models from the field sites, we analysed whether they were attacked or not. Lost models were removed from the analyses. In total, 16 clay models (3.2%) were lost. The models that had been attacked were visually inspected in detail and the bite marks were documented according to different predator types, i.e., rodents, birds, carabid beetles and unknown. Each clay model was counted once for each predator type present on the model. Hence, we did not count the number of bite marks of a specific predator type, as the number of bite marks is not informative for independent attacks (i.e., the same individual could have attacked each model repeatedly). If one model was attacked by more than one predator type, it was classified according to all of the identified predator categories (data available in supplementary table S2).

Statistical analysis
Experiment 1. To test whether nutritional conditions during the larval stage affected the colour phenotype after metamorphosis, we compared the proportion of yellow (yellow / (yellow + black)) of the fire salamander sibling pairs that were reared under the superior or limited nutritional conditions from day 50 and day 100 post metamorphosis in a single linear mixed model (LME). In the LME we used nutritional treatment (2 levels) and age post metamorphosis (2 levels: 50 and 100 days) as fixed factors and individual ID nested within maternal ID (18 individuals from 9 mothers) as random factors. The model was run using R. 3.3.1 (R Core Team, 2016) and nlme (Pinheiro et al., 2017). The residuals of the model were controlled for normal distribution using the qq-plot and Shapiro-wilk tests. The significance levels were set to α = 0.05 and all tests were twotailed.
Experiment 2. We analysed whether the proportion of yellow in the clay models is a good predictor for the risk of being attacked by various potential predators. Therefore, we conducted three generalized linear models (GLM) with a binomial distribution for each predator type (beetles, rodents and birds). We analysed whether the probability that clay models showed attack signs of a certain predator type was influenced by explanatory factors (year, habitat, colour composition). Colour composition of the clay model was a four-level factor, i.e., i) plain yellow = YYY; ii) more yellow than black = YYB; iii) more black than yellow = BBY and iv) plain black = BBB. The factor site had two levels i.e., whether the clay models had been deposited in the salamander-free site or in the site occupied by salamanders. The interaction of these two factors was also taken into account. Additionally, we included the year (2012 or 2013) when the experiment was performed as a factor in the model. The statistical analysis was done in R 3.3.1. P-values were obtained with the sequential analysis of deviance table by using the Chi-square tests.

Experiment 1 -effect of early nutritional conditions on colouration
Fire salamanders raised under limited nutritional conditions during the larval stage expressed significantly lower amounts of yellow colouration 50 to 100 days after metamorphosis compared to their siblings that were raised under ad libitum nutritional conditions table 1; fig. 2). Furthermore, the proportion of yellow decreased in age in both nutritional treatments (table 1; fig. 2).

Experiment 2 -colouration and predation risk
In 2012 and 2013, we placed a total of 500 clay models (125 sets of the four different model types) into the different sites (see table 2). On average, 83% of the clay models that were collected showed bite marks and were inspected in detail to determine the type of bite mark. In both years, the proportion of the clay models showing bite marks was higher in the salamanderfree site compared to the salamander site (in 2012: 14% more models with bite marks in the salamander-free habitat, in 2013: 8.4%). Moreover, the proportion of models with bite marks was higher in 2012 compared to 2013 (in 2012 about 11% higher in the salamander site and 16.6% higher in the salamander-free site compared to 2013). In the same habitat, the distribution of predator types was similar between the two years, with carabid beetles responsible for the majority of the attacked clay models in the salamander habitat. In the salamander-free habitat, birds were found to be the main predator type ( fig. 3).

The influence of predator type and artificial clay model type on predation risk
The risk of being attacked by carabid beetles was significantly affected by the colour composition of the artificial clay models and by the site (table 3a, fig. 4a). The risk of being attacked by carabid beetles was overall significantly higher in the salamander site compared to the salamander-free habitat. Independent of the site, the risk of being attacked by carabid beetles was significantly higher with an increasing amount of black ( fig. 4a). The likelihood of clay models being attacked by birds was significantly affected by the site and the colour composition of the artificial clay model, as well as the interaction of both (table 3b, fig. 4b). In the salamander-free habitat, more artificial clay models were attacked by birds, and all colour variants were attacked similarly often. In contrast, the likelihood of being attacked was significantly reduced with an increasing amount of yellow in the salamander habitat. We further found variation between the observation years (table 3b, fig. 3). In 2012, Table 3. Results of the attacks on the clay models from a) carabids, b) birds and c) rodents. The R 2 of the models were for carabids (0.07), for birds (0.17), and for rodents (0.035).  4. Absolute number of clay models found with bite marks from a) carabid beetles and b) birds. a) The artificial clay models with a higher proportion of yellow showed a significantly reduced risk of being attacked by beetles, which was independent of whether the site where the models had been deposited was inhabited by salamanders or was salamander-free. b) In birds, the proportion of yellow had an effect on the predation risk only in the site with salamanders. The artificial clay models with a higher proportion of yellow showed a significantly reduced risk of bird bite marks in contrast to the salamander-free habitat. The colours of the bars encode the proportion of yellow to black of the artificial clay models, with four different colour compositions, yellow = plain yellow; YYB = more yellow than black; BBY = more black than yellow, black = plain black. more artificial models were attacked by birds compared to the observations made in 2013. One potential reason for this finding might be the differences in the artificial clay model setup between 2012 and 2013 and a potential higher visibility of clay models in 2012 (in 2012 the models were setup in a cross-like shape and in 2013 in a line). The likelihood of being attacked by rodents was only marginally influenced by the site and not by the colour or the year (table 3c).

Discussion
We found that (i) a higher amount of yellow colouration after metamorphosis correlates with a higher amount of food availability during the larval stage and is likely a costly trait. Further, we showed that (ii) a larger amount of yellow in artificial models reduced the probability of being attacked. Combined, these two findings suggest that fire salamanders exhibit a condition dependent conspicuous yellow / black colour pattern, acting as an efficient warning signal.

The effect of early diet on colouration
The results of our first experiment demonstrated that resource limitation during early development, i.e., the larval stage, limits the development of yellow colouration after metamorphosis. Because all of the individuals experienced the same nutritional conditions after metamorphosis, and we controlled for genetic and maternal effects, the observed differences in colouration can only be a result of the different nutritional conditions during the larval stage.
The influence of the larval diet on adult colouration has been investigated in a variety of species (Grill and Moore, 1998;Blount et al., 2012;Davis, 2014). In ladybird beetles (Coccinella septempunctata), larvae that were reared under optimal nutritional conditions possessed higher elytra carotenoid concentrations (Blount et al., 2012), similar to the findings of our experiment. Fire salamander larvae raised under limited food conditions showed a lower proportion of yellow in their colour patterns compared to their siblings that were reared under superior food conditions. One possible explanation is that yellow colour is costly to produce and that individuals reared under better conditions can afford to produce more. That resource availability is influencing the production of defence chemicals has been shown in female wood tiger moth Arctica plantaginis (Burdfield-Steel et al., 2019). Other studies, however, found no effect of diet on warning signals (Lindstedt, Suisto and Mappes, 2019). Our larvae reared under limited nutritional conditions were significantly smaller at metamorphosis (Krause, Steinfartz and Caspers, 2011) and less yellow after metamorphosis and therefore did neither invest in size nor in coloration as much as those larvae that were reared under ad libitum nutritional conditions. Whether toxicity was different, or whether larvae reared under limited conditions invested more in toxicity to compensate for the deficits is unclear yet. Nevertheless if toxicity is correlated with size and age in fire salamanders, as has been found in the Brazilian red-belly toad (Jeckel, Saporito and Grant, 2015), those individuals able to invest in size and yellow pigments should benefit.
A second possible explanation is that individuals experiencing limited nutritional conditions are less toxic and therefore do or cannot invest in conspicuous colouration. A link between toxicity and colour saturation of aposematically ladybird beetles has recently be found (Arenas, Walter and Stevens, 2015). As we did not quantify the toxin content of the young terrestrial salamanders in the context of our experimental setup, we cannot currently disentangle the two options. A recent study investigating the link between yellow colouration and toxicity in fire salamanders, found no evidence that more yellow coloration correlated with more toxicity (Preißler et al., 2019), instead the authors found a sex specific influence on yellow coloration, with males being more yellow (Preißler et al., 2019), which indicates the potential role of colour pattern during mate choice. Regardless whether toxicity correlates with the amount of yellow coloration or not, the amount of diet, (i.e., the availability of resources) has a clear impact on the colour pattern in fire salamanders and probably the survival of recently metamorphosed fire salamanders in nature.
From previous studies, we know that limited food conditions influence metamorphic timing and body conditions at metamorphosis (Weitere et al., 2004), with larvae reared under limited food conditions metamorphosed later and at a smaller size (Krause, Steinfartz and Caspers, 2011). Four different chromatophore types (melanophores, xanthophores and erythrophores, and iridophores (reivewed in Pederzoli and Trevisan, 1990;Klewen, 1991;Lüddeke et al., 2018) are present in the yellow skin of the alpine salamander, Salamandra atra aurorae, which is a sister species of the fire salamander. In contrast, the black skin contains melanophores only (Pederzoli and Trevisan, 1990). This higher diversity of cell types in the yellow skin may be the reason why yellow colouration is more costly to produce. In addition, yellow colouration in vertebrates is often carotenoid-based and carotenoids need to be acquired from the diet (Olson and Owens, 1998), thus making carotenoid-based colouration condition dependent. Interestingly, the proportion of yellow decreases with age in post metamorphic fire salamanders, being highest shortly after metamorphosis and decreasing until adulthood, which is at about three to four years of age (see fig. 2, Krause, Steinfartz and Caspers, pers. obs.). As we cannot say anything about the colour intensity or hue, it still might be possible that fire salamanders with lower amounts of yellow invest in other important aspects of colouration. This needs to be investigated in future experiments.
In combination with the findings of the second experiment, one might speculate that the amount of yellow colouration decreases predation risk for young fire salamanders, as in experiment two the more conspicuous artificial models, i.e., models with a higher amount of yellow, were less likely to be attacked by birds.

The effect of colouration on the predation risk by different types of predators
For our study, we used artificial clay models with varying amounts of black and yellow to resemble the relative colouration variation (yellow: black) possessed by fire salamanders. By tracing back bite marks to different types of predators (carabid beetles, birds and rodents), we were able to determine whether artificial conspicuousness, i.e., the relative size of the yellow pattern, affected the probability of being attacked by a predator. The relative amount of yellow in the artificial models was associated with a decreased probability of being attacked by birds and carabid beetles, both known to be predators of young fire salamanders (García-París et al., 2003;Thiesmeier, 2004). We speculate that this could also indicate that fire salamander individuals with a higher amount of yellow in nature might have a higher survival rate. Future studies in the wild are needed to confirm our idea.
Our idea is in line with similar experiments on conspicuousness of colouration across species in poison frogs (Maan and Cummings, 2012) and ladybird beetles showing that conspicuousness signals unpalatability and/or toxicity more efficiently (Blount et al., 2012). However, it needs to be acknowledged that, in addition to colour, movement also increases visibility in aposematic species (Paluh, Hantak and Saporito, 2014;Blanchette, Becza and Saporito, 2017), a point that we consciously did not address in this study. Thus, it might be possible that less colourful individuals compensate their disadvantageous appearance by moving more when confronted with visually-oriented predators.
Artificial clay was most abundantly attacked by carabid beetles, birds and rodents. The first two groups are known to be common predators of young fire salamanders (Thiesmeier, 2004). For attacks by rodents, neither the amount of yellow colouration nor the site showed a measurable effect on the potential predation of salamanders. In contrast, we found a strong effect of the amount of yellow colouration on the risk of being attacked by birds. The artificial clay models with a high proportion of yellow were attacked significantly less frequently than the black models in the site where birds co-existed with salamanders. In contrast, we did not see any pattern based on differences in colouration in bird bite marks in the salamander-free site. Our finding is consistent with a study on poison frogs, in which birds avoided aposematic coloured local poison frogs whereas they attacked aposematic, non-local models (Noonan and Comeault, 2009), suggesting an important role for predator learning or adaptation to local environmental conditions. Differences in visual contrast (Endler, 1978) may only partially explain our artificial model results, as the two mixed artificial models had the same contrasts, i.e. two edges between black and yellow clay each. Nevertheless, birds significantly attacked the mixed artificial models with the larger amount of yellow less often. It is well known that visually-oriented predators, such as birds, are good models for testing the effects of colouration on the selection pressure of predators (Brodie, 1993).When attacked, fire salamanders exhibit a defensive posture during which the individual flexes its head downwards, bringing the yellow-coloured paratoid glands in prominent position, from which poison can also be sprayed in the direction of the threat (Brodie and Smatresk, 1990).
In the site where birds did not co-exist with salamanders, the amount of yellow of the clay models had no influence on the risk of being attacked by birds. A comparison of two sites is not sufficient evidence in support of the presence of fire salamanders is driving the differences in the results (see Hurlbert, 1984). However, it may indicate that conspicuous colouration did not deter birds in general, but may imply that birds have to experience poisonous and aposematically-coloured prey, such as fire salamanders, to learn and associate the "yellowblack" colour patterns with unpalatable prey. Indeed, avian predators can learn to avoid aposematic prey species (Mappes, Marples and Endler, 2005;Aronsson and Gamberale-Stille, 2008) and unpleasant experiences are crucial for this learning process (Skelhorn and Rowe, 2006). Accordingly, Ham et al., (2006) showed that great tits learned to avoid special/single colours of potential prey through harmful experiences; Marples, Roper and Harper, (1998) investigated the responses of wild birds to prey of novel, unfamiliar colours and found no consistent patterns and high inter-individual variation. Some birds preyed immediately on the unfamiliar coloured prey, whereas others never touched it. Unfortunately, our study design is not really suitable to test for the impact of learning or experience on aposematic coloured prey avoidance, and therefore these explanations are currently only speculation.
Carabid beetles showed a similar tendency of prey selection in both habitats. The artificial models that displayed more black colouration were attacked more frequently than those built with more yellow. Beetles very likely have a trichromatic colour vision system, similar to that of a honey bee (Kelber, 2006), which would allow them to differentiate between yellow and black or at least between light and dark. Carabid beetles also rely on olfactory cues during predation (Kielty et al., 1996;Oster et al., 2014). Although we cannot rule out the possibility that yellow and black clay differed in chemical cues being attractive to Carabid beetles, a very recent study found no evidence that Carabid beetles were attracted more to healthy prey items than to clay models (Ferrante, Barone and Lövei, 2017), thus we believe that our estimation of Carabid beetles attacks is correct. As Carabus problematicus is one of the main predators of young fire salamanders (Thiesmeier, 2004), it might therefore be possible that they are deterred by yellow coloration. However, this is only speculation at the moment and merits future research. Another known predator of fire salamanders are snakes such as Natrix natrix (Sauer and Weisbecker, 1994). Whether they respond to different colour patterns is currently unknown, however it is known from garter snakes that aposematic colour patterns enhance chemosensory recognition of noxius prey items (Terrick, Mumme and Burghardt, 1995).
Taken together, we found that high amounts of yellow in the skin of young metamorphosed fire salamanders are condition dependent and thus are likely costly to produce. Artificial clay models with a larger amount of yellow are less likely to be attacked. Combined, these findings suggest that young fire salamanders experiencing good nutritional conditions can produce more yellow colouration, which seems to come with a selective advantage, due to reduced predation risk. Future studies need to incorporate a proper measure of reflectance into the UV range to compare clay model properties with the real biological models, here the fire salamander. our manuscript. This study was financially supported by the German Research Foundation (DFG) from a grant to BAC (CA 889-1) and further inspired by the SFB TRR 212 (NC 3 ), also funded by the German Research Foundation (DFG). The collection of pregnant female fire salamanders was done with the permission of the nature reserve authority of the Stadt Bonn. The experimental procedures were conducted according to the German animal protection law.