B.C. Sheldon, Department of Zoology, University of Oxford, South Parks Road, Oxford OX1 3PS, UK. Tel.: +44 1865 281069; fax: +44 1865 271168; e-mail: firstname.lastname@example.org
When breeding, male moor frogs Rana arvalis develop a bright blue dorsal coloration which varies in intensity between males. We tested whether this colour acts as a potential signal of a male's genetic quality to female moor frogs by artificially crossing pairs of males differing in the extent of the blue coloration to the same female. Maternal half-sibships provide a powerful means to detect paternal genetic effects on offspring as they control for other potentially confounding variables. We assayed the ability of offspring to survive an ecologically realistic test of fitness by exposing them to predation by the larvae of the predatory water beetle Dytiscus marginalis. Although sire's coloration did not influence tadpole body size, it did affect their ability to survive the predation trial. Offspring of bright blue males had higher survival than those of dull males when exposed to large predators, which were more voracious predators than smaller ones. Our results indicate that paternal secondary sexual traits provide information about genetic effects on offspring fitness in this species, but suggest that these effects may be context-dependent. Variable selection caused by contextual dependence may have important consequences for the evolution of female choice rules, and for the maintenance of genetic variation for both male trait and female preference.
One important next stage in the development of understanding of mate choice for genetic benefits is to determine whether selection favours the choice of ornamented males under all conditions (regardless of direct selection on female choice). Most traits that have been investigated using appropriate experimental designs show genotype–environment interactions (GEI; Lynch & Walsh, 1998). This suggests that it may not be safe to assume that a demonstration of genetic benefits associated with choice of an ornamented male can be generalized across the wide range of environments that animals encounter (Sheldon, 2000; Qvarnström, 2001). Yet, there have been few attempts to assess how GEI influences selection on female choosiness (for exceptions see Jia & Greenfield, 1997; Lesna & Sabelis, 1999; David et al., 2000; Jia et al., 2000). One difficulty in performing such tests (in common with tests of genetic benefits models in general) is the problem of eliminating potentially confounding variables (for example, parental effects or maternal genetic effects, all of which could interact with aspects of the environment). An experimental design that offers considerable potential is the use of groups of maternal half-siblings produced within the same breeding event, as maternal half-siblings differ only in the genes that they inherited from their fathers (Barber & Arnott, 2000). This design (resulting either from natural cases of multiple paternity within families, or experimental crosses) has been used in six studies that have provided evidence for genetic benefits from mate choice (Sheldon et al., 1997; Welch et al., 1998; Johnsen et al., 2000; Barber et al., 2001; Doty & Welch, 2001; Wedekind et al., 2001).
In this paper, we used an artificial breeding experiment to investigate whether a variable, sexually dimorphic, coloration developed by breeding male moor frogs acts as a signal of the genetic quality of the sire. We mated individual females with two males that differed substantially in the extent of their blue coloration in order to maximize within family variation in the characteristic of interest in the sires. Resultant larvae were reared in the laboratory, their growth measured, and then subjected to a behavioural assay of their ability to escape from a known important predator of tadpoles.
The moor frog is widely distributed throughout Fennoscandia, and breeds in a large range of enclosed waterbodies. Males form large mating aggregations (up to 1000 individuals) in early spring to which females are attracted to mate. During the mating period, some males develop a conspicuous blue dorsal coloration (Fig. 1a) which is highly variable among individuals (cf. Fig. 1b), but females are always dull brown or blackish-brown (Fog et al., 1997). The eggs hatch about 2 weeks after fertilization, and the larval period lasts about 50–60 days. Predation pressure on anuran larvae is usually very high (Alford, 1999), and although we do not have direct data on predation pressure on moor frog larvae, their occurrence in predator-rich ponds together with lack of chemical defences strongly suggests that they are no exception. The importance of predation on larval life history is also illustrated by the fact that the moor frog larvae are known to develop induced defences (higher tail fins and tail muscles) in the presence of invertebrate predators (Lardner, 1998).
Collection of material for breeding
We collected adult moor frogs (52 males and 15 females) from a large population in a pond at Lindrågen (59°28′N, 13°31′E), Värmland, Sweden, on 24 April 1999. The frogs were collected opportunistically after dark from a large mating aggregation and transported to the shore. We collected both males and females from pairs in amplexus, in addition to solitary individuals. Once at the bank, amplexing pairs were separated and frogs kept temporarily in single sex groups for up to 30 min before processing. Adults were measured [snout-vent length (SVL), forearm length] and weighed, and males were scored for the extent and intensity of blue dorsal coloration using the standard scheme in Fig. 1b. Each area was given a score from 0 (least) to 5 (most) for both of these dimensions of colour. Our use of the human visual system to assess the coloration of the frogs is subject to the criticism that it assumes that variation in colour is perceived in the same way by the scorers as it is by the frogs. Although some evidence suggests that amphibian visual systems are not dissimilar to ours in terms of the number and wavelengths of peak sensitivity (Wilczynski, 1992), this remains an assumption of our study. However, the procedure that we followed, which was to make contrasts between two groups of males selected to be maximally different in terms of coloration (see below), should be relatively robust to different ways of measuring and perceiving colour.
After processing, each individual was assigned an identification code and placed in an individual 1-L container for transport to the laboratory in Uppsala, where the frogs were stored for 5 days in the dark in a constant temperature room (4 °C), conditions under which they are inactive.
Twenty-four males were selected for experimental matings by converting the information about blue coloration of each individual to a single numerical value. We first divided the dorsal surface of the frog into nine compartments. These comprised: (1) the head; (2 + 3) the forelimbs; (4) the upper back; (5) the lower back; (6 + 7) the left and right flanks; (8 + 9) the hindlimbs. Each compartment was scored on a five-point scale, from 1 to 5, for two aspects of coloration: proportion of compartment coloured blue, and the intensity of blue (corresponding to chroma). Figure 1 illustrates males at either end of the range from bright (Fig. 1a) to dull (Fig. 1b); the male in Fig. 1a would score close to the maximum on both measures for all compartments. These 18 measurements for each individual were then converted to a first Principal Component (PC; calculated on correlation matrix; PC1 explained 73% of the overall variation, with all compartments having strong positive loading), using the measurements for all 52 males. We then used PC1 to rank the individual males from least blue (dullest) to most blue (brightest) and selected 12 males from each tail of this distribution. Pairs of males (one dull, one bright) were then created by selecting one male from each tail of the distribution that were of similar size, as determined by SVL. We thus aimed to mate a single female to two males that differed in terms of their coloration but not in terms of body size. The males assigned to the bright and dull groups that were subsequently used in the predation experiments differed significantly in PC1 for blueness (t = 13.28, d.f. = 18, P < 0.0001), but not for SVL, forearm length or lean body mass (t = 1.54, d.f. = 18, P = 0.14).
Artificial crosses were performed following the procedure described in Meriläet al. (2000). Briefly, for each mating unit (one female and two males), the males were anaesthetized with an overdose of MS-222, their testes were dissected out and minced in amphibian Ringer (Rugh, 1962) on individual Petri dishes to produce a sperm solution into which the eggs from the female were stripped. The order of stripping of females eggs on the sperm solution was randomized and changed in the course of stripping to avoid possible bias caused by order of eggs in the oviduct. After 20-min stripping, the sperm solution was discarded and the eggs covered with reconstituted soft water (RSW; APHA, 1985). RSW was used also in all subsequent rearing of eggs and tadpoles to ensure homogenous water quality in the experimental units.
After crossing, more than 200 eggs were reared in 1-L plastic vials, with mesh walls to allow water exchange, placed in two large aquaria (144 × 53 × 15 cm) connected with a circulating water system (volume 300 L). After hatching the tadpoles were first reared in groups of 30 individuals in 1-L vials placed in the same large aquaria with circulating water. After 8 days, the tadpoles were transferred to plastic net cages (22 × 22 × 15 cm) placed in two aquarium systems each consisting of two large aquaria (144 × 53 × 15 cm, 12 cages in each aquarium) connected with water circulation. In each cage there were 50 tadpoles, all originating from the same full-sib family, and each full-sib family was replicated twice (once in both aquarium systems). The tadpoles were fed slightly boiled chopped lettuce ad libitum and temperature was maintained at constant 15 °C. At all times, eggs and tadpoles were identified by a two digit number representing their maternal half-sibship. Thus all rearing, measurements and behavioural assays were conducted without immediate knowledge of the characteristics of the tadpoles’ sire.
After 40-day hatching, free predation trials were conducted in opaque plastic aquaria (40 × 25 × 14 cm) filled with 10 L of water and supplemented with a few small stones to provide spatial complexity. As a consequence of logistic limitations (shortage of predators), trials were only conducted with 20 crosses (i.e. 10 sets of maternal half-sibs sired by dull and bright males). Ten larvae from each replicate of each cross were placed in each aquarium, and the predator, a late-instar larva of the large diving beetle Dytiscus marginalis was introduced into the aquarium after the larvae had acclimatized to the environment for 1 h. When in the laboratory the predators had been kept in 1-L vials and fed R. arvalis or R. temporaria tadpoles daily. However, before the trials they had been deprived from food for 2 days. Each predator was used only once in the trials. The number of surviving tadpoles in the aquarium was counted each hour after the initiation of the experiment, and the trials were terminated after 7 h. All trials were conducted in room temperature (20 ± 1 °C) under dim illumination. After the experiment, the predators were measured for their head width, body length and weighed to the nearest milligram. Before the predation trials, 10 tadpoles from each replicate were preserved in 10% formalin and their morphology was later measured. While female identity significantly influenced body length, body width, tail muscle height, tail height and tail length (F11, 32 ≥ 2.21, P ≤ 0.040), there were no differences between the offspring of blue and dull males in any of these characters (F1, 32 ≤ 0.75, P ≥ 0.393). Similarly, morphological variables corrected for general body size by using the residuals obtained from linear regressions, where the morphological variables were regressed against their first PC axis, did not differ between the male types (F1, 32 ≤ 2.51, P ≥ 0.123).
Survival of tadpoles during the predation trials was analysed using two methods. First, for each replicate, we scored the proportion of tadpoles (from the initial sample of ten) surviving to the end of the trial. These data were analysed using General Linear Modelling procedures implemented in GLMStat Version 5.1 (Beath, 2000), with binomial errors and a logit link. The second method used individual observations of tadpoles, where survival time was scored to the nearest hour during the predation trials. As these data are not normally distributed, and also strongly right-censored (many individuals survived past the end of the last observation period), statistical methods that allow for this error structure are required. We used proportional hazard models implemented in JMP Version 4.0.4 (SAS, 2001). These involve fitting maximum likelihood functions which assume parametric form for the covariates but nonparametric form for the hazard (survival) function. In both cases we added a measure of predator size to the model as a covariate, in addition to measures of the characteristics of the sire. When estimable, interactions were added to initial models and deleted when P > 0.15. Similarly, covariates with P > 0.20 were removed from the models. Maternal identity was included in all models as a general factor to control for any between-clutch effects unassociated with the father, irrespective of its statistical significance.
On average, 51% of tadpoles survived the 7-h predation trial, but the survival of tadpoles in each maternal half-sibship was quite variable (proportion surviving ranging from 0.200 to 0.667; Fig. 2, and even more variable among trials, ranging from 0 to 0.9). We therefore sought, initially, to explain variation in the proportion of tadpoles surviving the trial using maternal identity and predator size. GLMs indicated that there was a highly significant effect of predator size on the proportion of surviving tadpoles: effect of predator mass (cube root-transformed): b = −3.319 ± 0.612 SE, χ2 = 31.51, d.f. = 1, P < 0.0001. Predator mass was strongly correlated with predator length (r = 0.947, n = 61, P < 0.0001), but predator mass had a stronger effect on survival than did length (χ2 = 29.09, d.f. = 1, P < 0.0001), so was used as a covariate in subsequent analyses. A similar conclusion was reached by analysing tadpole survival using proportional hazard models: there was a highly significant effect of predator mass on the survival time of individual tadpoles (χ2 = 27.21, d.f. = 1, P < 0.0001). Maternal identity did not influence the proportion of tadpoles surviving to the end of a trial (F9, 52 = 0.92, P = 0.52).
Effect of sire's coloration
Addition of sire's coloration, coded either as a factor with two levels (blue, dull) or as a continuous variable (PC1 from scorings of colour variation) did not explain any additional variation in tadpole survival in itself, either with a GLM or proportional hazards model (χ2 = 0.75, d.f. = 1, P = 0.39). Thus, male coloration does not invariably influence the survival of tadpoles. As predator size was a major determinant of the survival of the tadpoles (accounting for approximately 75% of the deviance explained in the GLMs) we sought to determine whether tadpole survival was a function of the interaction between sire coloration and predator size. Adding a term representing the interaction between sire's coloration and predator size revealed that there was a significant interaction between the two, analysed using both GLM and proportional hazard analysis ( = 7.20, P = 0.007 and = 7.56, P = 0.006, respectively). In these analyses, we treated replicates from crosses as independent, which might mean that effect sizes were inflated due to pseudoreplication. We tested whether this was the case by fitting a mixed model, using restricted maximum likelihood (REML) estimation, which included the identity of the cross as a random effect, and with the proportion of tadpoles surviving to the end of the trial as the response variable. This model again returned a significant sire coloration × predator mass interaction (F1, 40 = 7.60, P < 0.009); residuals from this model were normally distributed (Shapiro–Wilk W = 0.984, P = 0.82), and there was no evidence of an effect of cross-identity on the proportion of tadpoles surviving (F19, 40 = 0.02, P = 1.00).
We explored the shape of the interaction between sire coloration and predator mass by plotting tadpole survival as a function of predator size and male coloration, coded categorically (Fig. 3). This revealed that the interaction is caused by the offspring of blue males surviving better than those of dull males when the predation risk is more intense (i.e. when the predator is larger). At low levels of predation (small predator) there is little difference between the survival probabilities of the offspring of the two types of male. One trial (with the largest predator) had relatively high leverage in the interaction (marked with an asterisk). Excluding this trail from the analysis, the interaction between sire coloration and predator mass becomes marginally nonsignificant (proportional hazard model: = 3.39, P = 0.066; mixed model: F1, 39 = 3.88, P < 0.056).
We tested whether other measured characteristics of the sire explained survival when tested in interaction with predator mass. There was no effect of whether a male was in amplexus or not when captured ( = 0.65, P = 0.42), or of male SVL ( = 0.56, P = 0.46), but the interaction between male lean body mass and predator mass also explained variation in survival ( = 5.57, P = 0.02). However, when both mass and male colour were tested in the same model, only the interaction between predator mass and male colour was significantly related to survival probability ( = 4.79 P = 0.03).
When moor frog tadpoles were exposed to an ecologically realistic predation trial, we found that the most important factor explaining whether they survived was the size of the predator in the trial. Unsurprisingly, large predators caught considerably more tadpoles within the predation trial than did small predators. The effect of the coloration of the father of the tadpoles on their survival was not consistent across trials, but depended upon the size of the predator. Tadpoles that were sired by males with more intense blue dorsal coloration were more likely to survive trials with large predators than were those sired by dull males. No other characteristics of the sire of the tadpoles explained their survival chances, when the effect of coloration was controlled for.
The experiment was designed as a direct test of whether the coloration of male moorfrogs indicated anything about the genes that their offspring would inherit. As the comparisons involve maternal half-siblings, with the effect of the mother's genes and maternal effects controlled for, any effect associated with the sire must be caused by genes inherited from the father (Welch et al., 1998; Barber & Arnott, 2000). As fertilization is external, and was conducted under constant controlled conditions, we can rule out the possibility that the effects are associated with any environmental factor. We did not detect any effect of sire coloration on growth rate, size at metamorphosis or survival in the laboratory, free of predation (J. Merilä, B. C. Sheldon, A. Laurila, unpublished data). However, when we exposed the tadpoles to a predator, we detected an effect that was dependent on the size of the predator, being more marked when the predation risk was more intense. There are three interesting implications of this finding.
First, predation on larvae is the major cause of mortality for many anuran amphibians (Alford, 1999). Thus, a tadpole's ability to survive predation challenge has important implications for its chance of successfully metamorphosing. The plot of the interaction in Fig. 3 suggests that the offspring of the bluest males enjoy an approximately two-fold increase in survival probability when faced with the largest size of predator. The majority of tests of ‘good genes’ sexual selection have concentrated on effects of sires on characteristics of offspring early in their development (e.g. Hoikkala et al., 1998; Welch et al., 1998; Johnsen et al., 2000; see Promislow et al., 1998 for an exception). Obviously it is difficult to extrapolate directly to a situation where predation occurs in natural populations, but a two-fold survival advantage over the course of a day would translate to a huge difference in numbers surviving by the end of the larval period. This would necessitate very strong counter-selection during later life-history stages in order to negate the benefits obtained due to increased larval survival. It is possible that an advantage in terms of predator evasion might be counter-balanced by a decrease in growth rate, and hence a lengthening of the larval period (cf. Barber et al., 2001). We failed to detect any other effects of sire coloration on juvenile characteristics, but this may have been because the tadpoles were fed ad libitum and hence faced little stress before exposure to predation.
The second interesting aspect of our finding was that the genetic effect of sires was more pronounced (indeed, only detectable) when the predation risk was intense. This result implies that the genetic benefit to a female from mating with a blue male depends on the conditions that her offspring will be reared in, and by extension, that selection on any female preference would depend upon the environment that her offspring encounter. The same is true for the genes underlying the relationship between the male coloration and the escape behaviour of the tadpoles. Three previous studies have shown that the genetic effect of sires depended on the conditions that their offspring were reared in (Jia & Greenfield, 1997; Lesna & Sabelis, 1999; Doty & Welch, 2001). Environmentally related variability in the outcome of gene-dependent choice in females may lead to plasticity in female choice rules. This might take the form of following particular rules under particular conditions (e.g. Qvarnström et al., 2000; Alonzo & Sinervo, 2001) or, if the relationship between sire characteristics and offspring fitness is unpredictable, lead to indiscriminate multiple mating as a form of bet-hedging (see Qvarnström, 2001, for further discussion). Current reviews of the fitness benefits of multiple mating by females tend to stress the importance of explanations based on genetic compatibility (e.g. Jennions & Petrie, 2000; Tregenza & Wedell, 2000), perhaps in interaction with the developmental mode of offspring (Zeh & Zeh, 2001). One alternative worth exploring is that the relationship between male ornamental characters and offspring fitness may be sufficiently unpredictable, in current environments, to greatly reduce the unanimity of female choice.
The main problem with a demonstration of a paternal genetic effect on offspring associated with a character under sexual selection remains that of explaining how genetic variation is maintained in natural populations. Levels of genetic variance are determined by the balance between selection on the one hand (removing variation) and mutation and gene flow (which regenerates it) on the other. Theoretical work suggests that sexually selected traits may have a large mutational variance if they are influenced by large numbers of loci (Rowe & Houle, 1996; Houle, 1998). If it is also true that if selection is variable in intensity, then explaining the maintenance of genetic variance in good genes systems may be less difficult, as the regenerative effects of mutation may be relatively greater, and the degenerative effects of selection relatively smaller, than has been believed. However, to show empirically that selection and mutation maintain standing genetic variance in natural populations will be a challenging task (see also Kotiaho et al., 2001).
Our results also suggest that blue coloration should be the target of female choice in moorfrogs, although it is possible that the degree to which females choose males on this trait may depend upon the circumstances of their breeding attempt (see Qvarnström et al., 2000 for a related demonstration in a wild bird population). There is limited evidence that blue coloration may be a target of female choice in this species. In a field sample which included the males collected for this study, we found that males in amplexus tended to be brighter blue than those that were not in amplexus (t = 1.84, d.f. = 86, P = 0.07) and Hedengren (1987) found that brighter blue coloration was associated with increased mating success in a different Swedish population. One difficulty in assessing these findings is that the coloration of males is highly labile, so that it is possible that part of the effect mentioned is caused by amplexus rather than the reverse. In addition, we do not have detailed knowledge of the visual physiology of this species, so are forced to assume that the variation in colour that we have scored is perceived in the same way by frogs. However, there is extensive evidence that both anuran and urodele amphibians, including congeners of Rana arvalis, have rather good colour vision (e.g. Harosi, 1982; Sillman, 1987; Przyrembel et al., 1995). Interestingly many amphibians show a phototactic response to blue light in preference to red or green light (Kicliter et al., 1981; Kicliter & Goytia, 1995; see Wilczynski, 1992 for a general account of amphibian visual physiology). A further possibility is that expression of the blue coloration, rather than being the direct target of female choice, is correlated with an unknown trait that is the target of female choice. However, this would raise the question as to what the function of the striking variation in coloration is (Fig. 1a,b).
Although definitive evidence from choice trials is lacking, our results suggest that studies of sexual selection in anurans should be extended to cover the possibility that coloration is the target of female choice, particularly in sexually dichromatic species (see Summers et al., 1999 for direct evidence of colour as the target of female choice in anuran amphibians). Because of the relative ease with which controlled breeding designs can be conducted (facilitated by large clutch sizes and external fertilization), amphibians offer some excellent opportunities for further investigation of the genetic effects of mate choice.
We are grateful to Sven-Åke Berglind, Maria Järvi-Laturi, Satu Karttunen, Karoliina Räsänen, Katja Räsänen, Fredrik Söderman, Ralph Tramontano and Kaarina Varkonyi for their help with fieldwork and rearing tadpoles, and to the Swedish Natural Sciences Research Council and British Ecological Society for funding the work. BCS is a Royal Society University Research Fellow. The experiments were conducted with the permission of the Ethical committee of Uppsala University.