INTRODUCTION

Cannibalism between juveniles is common in carnivorous animals (reviewed in Elgar and Crespi, 1992) and is assumed to be caused by competition for food (Fox, 1975; Polis, 1981). Many larval salamanders that compete intensely for food are cannibalistic (reviewed in Crump, 1992). Larger head and mouth width are advantageous to cannibals, because their predation is gape-limited (Kusano et al., 1985; Ohdachi, 1994). The disproportionately large head, wide mouth, and elongated vomerine teeth in cannibalistic morphs of Ambystoma larvae (Powers, 1907) are induced by high rearing density (Collins and Cheek, 1983), as well as by factors such as kinship (Pfennig and Collins, 1993) and diet (Walls et al., 1993). My previous studies showed that large head size among Hynobius retardatus larvae is also induced by high larval density (Nishihara, 1996a, b). The relative head widths to body size of feeding stage (stages 48 to 54; Iwasawa and Yamashita, 1991) larvae collected from Erimo, in Hokkaido Prefecture, where the egg density is high and cannibalism occurs, are larger than those of larvae collected from Bankei, in Hokkaido Prefecture, where density is low and cannibalism is not observed (Nishihara, 1996a). In captivity, the heads of feeding stage larvae from both populations grow to be proportionately larger when reared under high density conditions even when physical interactions are prevented (Nishihara, 1996a, b).

Faster growth of head size during pre-feeding stage is predicted to be favored in a larval salamander population with frequent cannibalism. This prediction is based on two hypotheses. First, a larger head in pre-feeding stage could reduce the probability of being caught and swallowed by conspecific cannibals because of the gape limitation. The larvae of Hynobius nebulosus have high mortality rates just after hatching due to predation and cannibalism (Kusano, 1981), and this is also predicted to be the case in H. retardatus. Second, faster growth of head size during pre-feeding stage could increase the probability of being able to cannibalize other individuals once the larva reaches the feeding stage. Reilly et al. (1992) argued that the time when a wider head is most important to a cannibal morph is during the first cannibalistic event among the similar-sized small larvae. These two hypotheses are not mutually exclusive, and if at least one of them is supported, I should predict that the relative growth rate of head size to body size during pre-feeding stage is larger in a population with frequent cannibalism than that with infrequent cannibalism.

In this paper, I first consider the above hypotheses on pre-feeding stage larvae having the advantage of possessing larger heads. I test whether pre-feeding stage larvae are more vulnerable to cannibalism than feeding stage larvae, whether the larvae with relatively smaller heads to body size are more vulnerable to cannibalism than those with relatively larger heads, and whether those with relatively larger heads are more cannibalistic than those with relatively smaller heads. These are tested in natural habitat where cannibalism is frequent. Second, I analyze whether my prediction on larval head growth pattern is correct. I compare the pre-feeding stage growthpatterns under the laboratory controlled experiments (Nishihara, 1996a, b) between populations with high and low frequency cannibalism. I test whether the effect of population on relative growth rate of head width to snout-vent length is significant.

MATERIALS AND METHODS

RESULTS

Field observations

Eighteen cannibal/victim pairs were observed at the pond in Erimo. One victim had been caught by its tail, and its head was stuck in its cannibal's mouth, and two other victims were not able to be pulled from the cannibal's oral cavity. These three pairs were excluded from the following analysis. The developmental stages ranged from stages 46 to 52, 42 to 47, and 41 to 51 for the cannibals, victims, and randomly collected larvae, respectively.

Seven out of 15 victims and 0 out of 15 cannibals were at pre-feeding stage (stages 42 and 43), and 4 out of 30 randomly collected larvae were at pre-feeding stage (stages 41 and 43). The proportion of the number of pre-feeding stage larvae to the feeding stage larvae was significantly greater for the victims than for the randomly collected larvae (χ² = 14.42, df = 1, P < 0.01).

Snout-vent length of the cannibals was larger than that of randomly collected individuals (t = −5.85, df = 43, P < 0.01). Whereas that of victims was not significantly different from that of randomly collected larvae (t = 1.15, df = 43, P = 0.26). Relative size of head width to snout-vent length is compared between the victims and the randomly collected individuals within the range of victims' stages (42 to 47) (Fig. 1A). ANCOVA revealed that head width after removing the effect of snout-vent length is smaller in victims than in randomly collected individuals (heterogeneity of slopes, F = 0.69, df = 1,33, P = 0.41; elevation, F = 20.30, df = 1,34, P < 0.01). Figure 1B shows the comparison of relative size of head width to snout-vent length between the cannibals and the randomly collected individuals within the range of cannibals' stages (46 to 52). ANCOVA revealed that there is no significant difference between cannibals and randomly collected individuals in head width after removing the effect of snout-vent length (heterogeneity of slopes, F = 0.04, df = 1,34, P = 0.84; elevation, F = 2.37, df = 1,34, P = 0.13).

Fig. 1

Comparison of relative head width to snout-vent length in larval Hynobius retardatus from a natural pond of Erimo population. (A) Comparison between victims and randomly collected larvae; (B) Comparison between cannibals and randomly collected larvae. Comparisons were made between larvae within the same ranges of developmental stages.

Laboratory experiments

The average snout-vent length of the hatchlings (stage 41) from the Erimo population was significantly larger (by ca. 10%) than that from the Bankei population, but their head widths did not differ between the populations (Table 1). All larvae from both populations reached the feeding stage (stage 44) by the 7th day after hatching. Both snout-vent length and head width growths during pre-feeding stage were significantly larger (by more than 40%) in larvae originating from the Erimo population than in those from the Bankei population (Table 1). Relative head width in the pre-feeding stage increased faster in the larvae of the Erimo population than in those of the Bankei population (Table 1). Two-way ANOVA showed no effect of density or interaction on any of the size and growth parameters (Table 1).

Table 1

Results of two-way ANOVA on the growth parameters (mean ± SD) of larvae from two populations under high- and low-density treatments

DISCUSSION

Observations in the field revealed that pre-feeding stage larvae in the pond were fed more heavily by their conspecifics compared to feeding stage larvae. The data also showed that larvae with proportionally smaller head widths are more vulnerable to cannibalism than those with larger heads. Although the differences of absolute head widths rather than proportional might be more important in determining the outcome of cannibalism as shown in Ohdachi (1994) and Kusano et al. (1985), the results support the hypothesis that being the most vulnerable stage to cannibalism, pre-feeding stage larvae with larger head will reduce the risk of being cannibalized. The hypothesis that wider head increases the possibility of preying on smaller conspecifics was not supported in the present result because there is no difference in proportional head size between cannibals and randomly sampled individuals. However, it is true that the absolute body size of cannibals measured as their snout-vent length was larger than that of randomly sampled larvae. Whether large individuals tend to cannibalize others or cannibalistic individuals tend to become larger than the rest of the conspecifics is unknown.

Relative head width increased in a faster rate as well as absolute head width and snout-vent length during the prefeeding stage in larvae of the cannibalistic Erimo population than in larvae of the non-cannibalistic Bankei population. This result is consistent with my prediction that faster growth in head size is favored in a population with frequent cannibalism. Erimo population has larger egg size than Bankei population (unpublished data), which might be causing the greater overall growth during the pre-feeding stage in Erimo population, although egg size is most likely determined by combinations of factors such as altitude (Takahashi and Iwasawa, 1988) and breeding habitat (Petranka et al., 1987; Woodward, 1982).

Alternative way to reduce the risk of being cannibalized is to produce larger hatchlings and/or shorten the duration of the pre-feeding stage period. Kaplan (1980) reported that Ambystoma tigrinum produces larger hatchlings and shortens the length of the pre-feeding stage by laying large eggs. The size of hatchlings cannot be interpreted as an adaptation to minimize cannibalism in this species, H. retardatus, because although the snout-vent length was significantly larger in hatchlings from the Erimo population, the head width of hatchlings which is more critical for avoiding cannibalism was not significantly different between these two populations. The length of the pre-feeding stage period was 7 days in larvae from both populations at the temperature of 17°C. The larvae were observed daily, and the range of the length of period was 6 to 7 days, indicating a highly synchronized development in the two populations despite having different egg sizes. In A. tigrinum, the duration from hatching to start feeding varies as much as 1 month among the embryos of different females (Kaplan, 1980).

I found no effect of rearing density on any of the prefeeding stage growth parameters. In my previous studies (Nishihara 1996a, b), I showed that there is density-induced phenotypic plasticity in overall growth pattern of larvae (including both pre-feeding and feeding stages) from both Erimo and Bankei populations. These results indicate that the effect of rearing density on growth and morphology of the larvae appears during the feeding stage in this species. In addition, the significant population effect on growth parameters of prefeeding stage regardless of the rearing density indicates that genetic variation of these larval traits exists among local populations of this species.

The results from the field observation suggest that the larval head morphology and growth in Erimo population are currently under selective pressure of being cannibalized, but this does not necessarily mean that these traits are under directional selection. Possible costs of cannibalism have been known such as risk of injury or death, risk of eating a relative (reviewed in Crump, 1992), or risk of consuming pathogens from victims (Pfennig et al., 1991). If any of these costs is associated with the cannibalistic characters of the larvae, they could be under balancing selection. To draw conclusions to what kind of selection the current population is experiencing, more information on these possible costs should be obtained.

Previous experiments which examined the factors inducing the cannibal morphs of ambystomatid salamanders (Collins and Cheek, 1983; Pfennig and Collins, 1993; Walls et al., 1993) had been terminated and measured when the larvae were at some stage near metamorphosis. The growth pattern of the head morphology throughout the pre-feeding and feeding stages should be considered separately. The results of the present study emphasize the significance of acquiring large head size during pre-feeding stages, i.e. the stages most vulnerable to cannibalism, supporting the hypothesis that a wide head is an adaptation against frequent cannibalism in larval H. retardatus.