Interpopulation variation in predator foraging behaviour promotes the evolutionary divergence of prey

Authors


Hirokazu Toju, Ecology Sciences, Department of Biology, Faculty of Science, Kyushu University, Hakozaki 6-10-1, Higashi-ku, Fukuoka 812-8581, Japan. Tel.: +81 92 642 2624; fax: +81 92 642 2645; e-mail: hiro.toju@gmail.com

Abstract

Despite intensive investigation of the role of predation on evolutionary processes, few studies have questioned the possibility of the evolutionary divergence of prey populations in response to interpopulation variation in predator foraging behaviour. In an interaction between a seed-predatory insect, the camellia weevil (Curculio camelliae), and its host plant, the Japanese camellia (Camellia japonica), I tested whether the evolutionary differentiation of the plant's defensive trait, pericarp thickness, was related to the interpopulation variation in the foraging behaviour of female weevils. I found that the preference of weevils for the plant fruit based on pericarp thickness varied across 13 populations in Japan. Importantly, variation in weevil behaviour explained interpopulation variation in pericarp thickness and the direction/strength of natural selection on the trait. Overall, I show that adaptive foraging of predators can result in the evolutionary divergence of predator–prey interactions.

Introduction

Predation is one of the most important driving forces of organic evolution (Vermeij, 1987; Abrams, 2000). Evolutionary diversification of the defensive traits of prey has been a central focus of studies of predation. Many authors have reported that variation in predation regimes, e.g. predator density or habitat features, is an important determinant of the adaptive radiation of prey species or populations (Endler, 1980; Reznick & Endler, 1982; Brakefield, 1987; O'Steen et al., 2002; Rundle et al., 2003; Langerhans et al., 2004; Eklöv & Svanbäck, 2006). Nevertheless, few studies have tested whether the evolutionary divergence of prey populations is related to interpopulation variation in the foraging behaviour of their predators.

During the past three decades, behavioural ecologists have clarified that predators, parasitoids and herbivores exhibit adaptive behaviour in searching for and choosing their prey or hosts (see Stephens & Krebs, 1986), although they have not questioned the effects of such behaviour on the evolution of the victims. For example, studies on foraging theory have predicted and demonstrated that predators and parasitoids search for patchily distributed prey or hosts in ways that maximize foraging efficiency (e.g. average rate of energy intake; Charnov, 1976; Iwasa et al., 1981; Wajnberg et al., 2000, 2006). During adaptive foraging, foragers are expected to regulate the duration of foraging at a patch depending on the quality of that patch or the quality of prey or hosts therein (Hubbard & Cook, 1978; Green, 1984; Lima, 1984). Similarly, female insect seed predators or herbivores select oviposition sites (Rausher, 1979; Liu et al., 2005; see also Craig et al., 2000) to increase the survival rate or performance of offspring (Jaenike, 1978; Thompson, 1988; van Alphen et al., 2003; Scheirs et al., 2004).

Importantly, such adaptive selectivity or preference of predators can lead to relationships between the expected number of attacks and genetically determined traits of prey, resulting in evolutionary changes in prey traits. If there is interpopulation variation in genetically determined foraging tactics (see Jaenike, 1989; Fox et al., 2004; Wajnberg et al., 2004) or variation in biotic or abiotic conditions affecting the behaviour of predators (Waage, 1979; Lima, 1984; Haccou et al., 1991; Driessen & Bernstein, 1999; Outreman et al., 2005), the prey may show interpopulation differentiation of defensive traits that are involved in interactions with their predators. This potential effect of predator foraging behaviour on prey evolution can be tested in the context of a geographically structured interaction involving a specific predator and prey. Recently, several studies revealed that defensive traits of prey vary greatly across populations (Brodie et al., 2002; Benkman et al., 2003; Toju & Sota, 2006a, b, c; sensuThompson, 1999), and such systems are excellent for testing the impact of predator behaviour on the evolution of prey traits (Schilthuizen et al., 2006).

Here, I evaluated the hypothesis that interpopulation variation in the adaptive foraging behaviour of predators mediates the evolutionary divergence of their prey by focusing on a system involving a seed-predatory insect, the camellia weevil (Curculio camelliae Roelofs; Curculionidae: Coleoptera; Fig. 1a), and its host plant, the Japanese camellia (Camellia japonica L.; Theaceae; Toju & Sota, 2006a, b, c; see also Thompson, 2005). The weevil is an obligate predator of the seeds of Japanese camellia, and females excavate the thick pericarp of the fruit with their long rostrum to oviposit into the seed. Previous studies (Toju & Sota, 2006a, b, c) showed that these species co-occur throughout a large portion of the host plant's range, and the weevil imposes geographically varying selection pressure on the defensive trait of the plant (pericarp thickness). Thus, this system provides an ideal opportunity to examine how natural selection due to species interactions varies across populations (see Thompson, 2005; Toju & Sota, 2006c).

Figure 1.

 The study organisms and localities. (a) A female of the camellia weevil drilling into the fruit of the Japanese camellia. (b) Thirteen study sites.

I had three main objectives. First, I showed that there is variation among populations in female preference for Japanese camellia fruits based on pericarp thickness. I use preference to refer to the shapes of functions representing the relationship between the number of attacks (excavation attempts) per fruit and pericarp thickness of the attacked fruit. Second, I tested whether the interpopulation variation in the preference of weevils for Japanese camellia fruit is responsible for variation in the direction and strength of natural selection on pericarp thickness. Lastly, I investigated the causes of interpopulation variation in the foraging behaviour of the weevil to further explore the factors promoting variation among populations in pericarp thickness. Overall, this work provides a novel example of the link between interpopulation variation in predator foraging behaviour and the evolutionary divergence of prey.

Material and methods

Study system

The camellia weevil (C. camelliae) is an obligate seed predator of the Japanese camellia (C. japonica); the larvae feed exclusively on the seeds of the host plant (Okamoto, 1988). The female weevil excavates the thick pericarp of Japanese camellia fruit using its extremely long rostrum and then inserts its ovipositor into the excavated hole to lay an egg into a seed. She is considered to deposit a single egg for each of successful excavations because a larva consumes 1.2 seeds on average within each fruit, which usually contains up to 10 seeds (H. Toju, unpublished). The probability of a female successfully excavating a pericarp is positively correlated with her rostrum length and negatively correlated with pericarp thickness (Toju & Sota, 2006a). It is noteworthy that the direction and strength of natural selection on the pericarp thickness of Japanese camellia varies geographically and thus the plant trait diverges greatly among populations (Toju & Sota, 2006a, b, c).

Predator behaviour and divergence of prey traits

Because the success of pericarp excavations is dependent on the thickness of the pericarp of attacked fruit (Toju & Sota, 2006a), female weevils that avoid fruits with thick pericarps should have higher fitness because of increased oviposition success. However, because fruits of the Japanese camellia are patchily distributed and movement between patches is expected to incur either a time cost or increased predation risk (see Stephens & Krebs, 1986), females should also excavate fruits with thick pericarps. Moreover, depositing eggs in fruits already used by another female reduces the survival and growth of larvae, so the weevils should choose fruits in which density of larvae is expected to be low. Interestingly, a congener of the camellia weevil, the chestnut weevil (Curculio elephas), cannot determine whether its host fruit contains eggs of another female (Desouhant, 1998). Thus, female camellia weevils may avoid the aggregation of their larvae based on some external features of the host fruit, such as fruit diameter, rather than by determining if another female has already oviposited there.

To reveal the interpopulation variation in the preference of the camellia weevil for Japanese camellia pericarp thickness, I used specimens from 13 of 17 populations analysed in a previous study (Table 1, Fig. 1b; Toju & Sota, 2006a). Note that the remaining four populations (Shodoshima, Arafune, Taiji and Ushibuka) were excluded because of small sample sizes (< 300) of holes made through pericarps by camellia weevils (trial holes). For each of the 13 populations, up to eight fruits were randomly sampled from each tree. The number of trial holes and the number of weevil larvae on seeds were counted for each fruit. Also, for each fruit, I calculated the pericarp thickness as the mean of four measurements along cross-axes of a longitudinal section of the fruit. Pericarp thickness was measured to the nearest 0.01 mm using digital callipers.

Table 1.   The interpopulation variation in the preference of the camellia weevil for the Japanese camellia fruit based on pericarp thickness.
Locality
Latitude (°N)Longitude (°E)Rostrum length (mm ± SE)Pericarp thickness (mm ± SE)No. of fruit, N (tree)Linear regressionQuadratic regression
Coeff. (SE)tPCoeff. (SE)tP
  1. The result of the linear/quadratic regressions of the number of trial holes (per fruit) made by camellia weevils on the pericarp thickness of each attacked fruit was shown. Also shown are the means of the rostrum length of the camellia weevil and the pericarp thickness of the Japanese camellia (see Toju & Sota, 2006a for details).

  2. *The random effects of individual trees and fruits could not be incorporated in models (see text).

Kutsuki35.37135.9110.42 ± 0.616.07 ± 0.8742 (24)1.50 (1.91)0.80.4433−0.195 (1.256)−0.20.8783
Kyoto35.02135.819.89 ± 1.026.30 ± 1.6520 (18)8.34 (2.35)3.60.17451.094 (1.349)*0.80.4280
Jurinji34.78135.3110.31 ± 0.936.13 ± 1.2060 (32)3.19 (1.82)1.80.0901−0.251 (0.761)−0.30.7437
Nara34.69135.8810.05 ± 0.876.66 ± 1.3874 (36)3.86 (0.74)5.2< 0.00010.846 (0.387)2.20.0356
Kiikatsuura33.65135.9910.66 ± 1.177.73 ± 2.4726 (20)−0.33 (0.80)−0.40.6938−0.078 (0.207)−0.40.7238
Kiioshima33.47135.869.61 ± 0.597.52 ± 1.4742 (18)2.86 (1.89)1.50.14461.623 (1.551)1.00.3069
Usa33.43133.4613.63 ± 0.9111.65 ± 2.8233 (23)0.09 (0.53)0.20.8715−0.152 (0.145)*−1.10.3010
Muroto33.25134.1910.06 ± 0.667.77 ± 1.6930 (23)2.42 (0.83)2.90.02720.602 (0.360)1.70.1557
Ashizuri32.73133.0312.98 ± 0.9312.80 ± 2.4599 (66)1.57 (0.81)1.90.0608−0.813 (0.149)−5.4< 0.0001
Reihoku32.53130.0311.68 ± 0.9411.89 ± 2.6161 (38)3.90 (1.10)3.60.0018−0.698 (0.276)−2.50.0194
Takahama32.34129.9811.48 ± 0.7011.13 ± 2.6882 (37)−0.63 (0.68)−0.90.3617−0.420 (0.144)*−2.90.0045
Yahazu30.46130.5014.54 ± 1.8712.49 ± 3.3998 (41)1.46 (0.62)2.40.0212−0.169 (0.129)−1.30.1941
Hanyama30.38130.3919.48 ± 1.8520.41 ± 3.9950 (21)−0.03 (0.67)0.00.9612−0.260 (0.124)*−2.10.0405

To determine the preference of female weevils for pericarp thickness, I regressed the number of trial holes per fruit on pericarp thickness for each population. All analyses were conducted using the statistical package R, version 2.2.1 (http://www.r-project.org/). I used both linear and quadratic regression models. For the linear model, I used a generalized linear mixed model (penalized quasi-likelihood procedure) with normal error distribution and an identity-link function (glmmPQL command). Pericarp thickness was used as a fixed term, and individual trees and each fruit nested within individual trees were fitted as random terms. For the quadratic regression, I used a nonlinear mixed model in which the above two random terms were incorporated (nlme command). However, because the calculation of a nonlinear mixed model did not converge for several populations, I used a model without random terms for those populations (nls command). Note that fruits with no trial holes were excluded from the regressions because such fruits may be avoided by female weevils because of other unidentified factors. This procedure was validated because the pericarp thickness of fruits with zero trial holes was not significantly different from the pericarp thickness of attacked fruits in 12 of 13 populations (anova; α = 0.05) and even in the remaining population (Ashizuri), only three of 102 fruits were eliminated. Besides, it is noteworthy that female weevils did not oviposit at all the times when they succeeded in excavating Japanese camellia pericarps presumably because they evaluated the quality of seeds to avoid oviposition into unfavourable seeds such as immature ones. This behavioural property of camellia weevils, however, would not affect the hypothesis testing below because comparisons among fruits revealed that the ratio of the number of larvae to the number of successful excavations (i.e. probability of accepting seeds) was independent of pericarp thickness in 12 of 13 populations analysed (linear regression; α = 0.05), and even in the remaining population (Yahazu), the association was weak (P = 0.043).

The linear or quadratic coefficients of the linear and quadratic regressions were assumed to represent the shapes of the relationship between the number of attacks and pericarp thickness in each population. Thus, by comparing the coefficients among populations, I examined interpopulation variation in insect behaviour. Before population comparisons, the coefficients were standardized so that the explanatory variable (pericarp thickness) and response variable (number of trial holes per fruit) had unit variance and a mean of zero for each population. I called the standardized linear or quadratic coefficients βpref or γpref respectively.

Subsequently, the effect of the weevil's preference on the evolutionary divergence of pericarp thickness was tested. I predicted that interpopulation variations in the indicators of weevil's preference, βpref or γpref, was associated with the differentiation of pericarp thickness or that of natural selection acting on the plant trait. Therefore, across populations, the pericarp thickness of the Japanese camellia was regressed on βpref or γpref respectively. Also, the correlations between βpref or γpref and the standardized linear selection coefficient (Lande & Arnold, 1983) for pericarp thickness, which was evaluated in a previous study (Toju & Sota, 2006a), were examined. In these regressions, βpref or γpref estimates with more than 0.15 P-values were replaced with zero because their original estimates were statistically unreliable.

Realized distribution of weevil larvae

To understand the mechanisms linking the preference of weevils and the evolutionary divergence of the pericarp thickness of Japanese camellia, the association between the pericarp thickness and the realized density of weevil larvae on each fruit was visualized. For each population, the relationship between pericarp thickness and the number of weevil larvae per fruit was shown with a cubic spline (Schluter, 1988) using the software glms ver. 4.0 (Schluter, 2000). For simplicity, populations without significant preference of weevils for Japanese camellia fruits were excluded. In addition, because γpref, but not βpref, showed significant or nearly significant correlation with the interpopulation variations in pericarp thickness or the strength of natural selection on the trait (see Results), application of the cubic spline analysis was concentrated on five populations in which γpref was statistically significant at α =0.05.

Importantly, the realized distribution of weevil larvae on fruit is expected to strongly affect the variation in the fitness of Japanese camellia individuals. Therefore, to link the distribution of larvae and the evolution of pericarp thickness, natural selective pressures acting on the plant trait were also visualized. In this aim, the data of natural selection on pericarp thickness (Toju & Sota, 2006a) were reanalysed using a cubic spline for each of the five populations noted above.

Factors affecting variation in foraging behaviour

Finally, I examined the causes of the geographic variation in camellia weevil preference for fruit based on pericarp thickness. Given that the success of excavations is negatively correlated with pericarp thickness (Toju & Sota, 2006a), fruits with thinner pericarps are expected to contain more weevil larvae if all fruits are attacked an equal number of times. In such a situation, however, larvae are expected to be aggregated in fruits with thin pericarps and compete with each other there. Hence, female weevils should invest certain time in attacking fruits with thick pericarps to make the distribution of their larvae on fruits uniform against pericarp thickness and thereby avoid aggregation of offspring (cf. Hubbard et al., 1987; Visser et al., 1992; Desouhant et al., 2000; Nufio & Papaj, 2004; Plantegenest et al., 2004).

However, attacking fruits with thick pericarps is costly because such attacks would be more likely to result in the failure of pericarp excavations and the resultant reduction of oviposition efficiency (i.e. average rate of oviposition; cf. Charnov, 1976; Iwasa et al., 1981; Sevenster et al., 1998). Thus, it is expected that there is a trade-off between the oviposition efficiency of female weevils and the performance/survival of their larvae (Scheirs et al., 2000; Mayhew, 2001; Scheirs & De Bruyn, 2002). Therefore, I predicted that female weevils basically chose thicker pericarps, but in populations in which the average probability of successful excavations was lower, they were more likely to avoid fruits with too-thick pericarps. This prediction was tested by regressing βpref or γpref on the proportion of successful pericarp excavations (i.e. the total number of trial holes in fruit divided by the total number of holes reaching seeds for each population; Toju & Sota, 2006a) across populations.

Results

Variation in weevil's preference for pericarp thickness

Four of 13 populations examined showed significant positive relationships between the number of trial holes and the pericarp thickness of the attacked fruit (Table 1; Fig. 2). In four populations that partially overlap the abovementioned four populations, fruits with moderate pericarp thickness had the most trial holes (Table 1; Fig. 2). Although the number of trial holes showed a concave relationship with pericarp thickness in one population (Nara), the curve represented a virtually monotonic increase (Fig. 2). Overall, these results suggested that weevil's preference for Japanese camellia fruit differed among populations.

Figure 2.

 The interpopulation variation in the relationship between the number of attacks and pericarp thickness. The number of trial holes made by female camellia weevils on respective fruits was regressed on the pericarp thickness of the Japanese camellia for each population (see also Table 1). Solid lines represent significant linear regression, whereas dotted lines represent significant quadratic regression (α = 0.05 for both cases).

Then, the relationship between the pattern of weevil's preference and the evolution of Japanese camellia pericarps was tested. Regressing weevil's preference on Japanese camellia pericarp thickness revealed that γpref significantly decreased with increasing pericarp thickness (y = 8.80 − 21.5x, F1,11 = 9.2, P = 0.0115; Fig. 3a), although for βpref the relationship was not significant (y = 10.7 − 3.39x, F1,11 = 0.3, P = 0.58). Also, the correlation between γpref and the natural selective pressures acting on pericarp thickness was nearly significant (y = 0.00487 − 0.568x, F1,11 = 4.5, P = 0.0563; Fig. 3b), whereas βpref did not show association with the direction/strength of natural selection on the plant defensive trait (y = 0.0618 − 0.125x, F1,11 = 0.4, P = 0.53).

Figure 3.

 The relationship between the preference for fruit based on pericarp thickness and the divergence of the plant trait. (a) The mean pericarp thickness of Japanese camellia was regressed on the quadratic regression coefficients of Fig. 2 (standardized) (γpref; see text) across populations. Note that a negative value of γpref means that the camellia weevils preferred fruit with intermediate pericarp thickness. The solid line represents a significant linear regression. Filled circles indicate populations with significant γpref estimates (α = 0.05; Table 1). (b) The standardized linear selection coefficient (Lande & Arnold, 1983) for pericarp thickness was regressed on γpref across populations. The dashed line represents a nearly significant regression (P = 0.056).

Linking preference, distribution of larvae and natural selection on pericarp

In population Nara, where βpref and γpref were positively significant and female weevils preferred fruits with thicker pericarps (Fig. 2), there was no correlation between pericarp thickness and the number of weevil larvae per seed (Fig. 4), presumably because the probability of successfully excavating pericarps decreases with increasing pericarp thickness (Toju & Sota, 2006a). As a result, positively significant directional selection on pericarp thickness was not detected in this population (Toju & Sota, 2006a), rather, a cubic spline showed weak natural selection towards thinner pericarps (Fig. 5).

Figure 4.

 The distribution of weevil larvae against pericarp thickness of Japanese camellia. Each panel represents the relationship between pericarp thickness and the number of larvae per seed for each population. Solid lines signify the prediction curves by cubic splines (Schluter, 1988) with ±SE (dashed lines). For simplicity, populations with significant γpref estimates (Table 1) were shown.

Figure 5.

 Visualization of natural selection on pericarp thickness of Japanese camellia. Natural selection on the pericarp thickness of Japanese camellia (Toju & Sota, 2006a) was reanalysed by cubic splines. As a fitness measure, the proportion of surviving seeds calculated for each individual was used. Solid lines represent the prediction curves (±SE; solid lines). For simplicity, populations with significant γpref values (Table 1) were shown.

In four populations where γpref was negatively significant, weevil larvae were more aggregated in the seeds of fruits with thinner (Ashizuri, Takahama and Hanyama) or intermediate (Reihoku) pericarps (Fig. 4). Because female weevils preferred fruit with intermediate pericarp thickness in these populations (Fig. 2, Table 1), there were few larvae in the seeds of fruits with thick pericarps (Fig. 4). Although fruits with thinner pericarps were not preferred there too, they contained more weevil larvae in three of four populations (Fig. 4) because female weevils were more successful at excavating such fruits (Toju & Sota, 2006a). These distributions of weevil larvae resulted in specific fitness functions for pericarp thickness in respective populations (Fig. 5). That is, thicker pericarps were favoured in three populations where fruits with thinner pericarps contained more weevil larvae (Ashizuri, Takahama and Hanyama), and disruptive selection was detected in population Reihoku, in which fruits with intermediate pericarp thickness had the most weevil larvae (Figs 4 and 5; Toju & Sota, 2006a).

Causes of interpopulation variation in preference

Analyses of the potential effect of interpopulation variation in weevil's preference for pericarp thickness revealed that γpref was significantly associated with the proportion of successful excavations (y = −0.327 + 0.504x, F1,11 = 20.8, P = 0.0008; Fig. 6), whereas βpref was not (y = 0.160 + 0.0563x, F1,11 = 0.0, P = 0.85). That is, in populations in which weevils were subject to a higher risk of the failure of pericarp excavations, they attacked fruits with intermediate pericarp thickness.

Figure 6.

 A putative factor of the interpopulation variation in the preference for Japanese camellia fruit based on pericarp thickness. The quadratic regression coefficient in Fig. 2 (standardized; γpref) was regressed on the proportion of successful excavations by the camellia weevils across populations. Filled circles represent significant regression coefficient in Fig. 2. Arrow indicates population Nara.

Discussion

To test the hypothesis that interpopulation variation in the foraging behaviour of predators can cause evolutionary diversification of prey populations, I hypothesized that female camellia weevils showed adaptive foraging behaviour on patchily distributed fruit of Japanese camellia. In the foraging processes, the camellia weevil should select the pericarp thickness of Japanese camellia because fitness of females is strongly dependent on the size of this plant trait (Toju & Sota, 2006a; see also discussions below). Indeed, I found interpopulation variation in the preference of the camellia weevil for the pericarp thickness of the host fruit (Table 1; Fig. 2).

A number of studies on foraging theory allow us to speculate on two alternative mechanisms mediating the preference of weevils for Japanese camellia fruits based on pericarp thickness. The one is that the oviposition behaviour of camellia weevils is genetically differentiated due to varying natural selection in each population (see Jaenike, 1989; Fox et al., 2004; Wajnberg et al., 2004). If camellia weevils can select preferable fruits based on some kinds of external cues, such as fruit diameter, it can be postulated that the interpopulation variation in preference (Table 1; Fig. 2) results from variation in the thresholds for determining whether to accept or reject encountered fruits. The other possible mechanism is the plastic modification of foraging behaviour. By using knowledge of the success or failure of previous excavations, camellia weevils may be able to decide whether to remain on the present fruit for further oviposition attempts (cf. Waage, 1979; Lima, 1984; van Alphen et al., 2003). For example, when a female has successfully excavated a pericarp and oviposited on one of the seeds, it is likely that subsequent excavations and oviposition on the same fruit will also be successful. Thus, remaining on the current fruit and attempting another excavation would likely be more adaptive, in terms of the number of eggs laid, than spending time to search for another fruit (incremental mechanism; Waage, 1979; van Alphen et al., 2003). In contrast, when an excavation is unsuccessful, the female is likely to fail in subsequent excavations on the same fruit. Hence, it would be more adaptive for the female to search for another fruit better suited to her excavation abilities. From the aspects of this second hypothesis, we can expect that camellia weevils can modify their foraging behaviour in response to population-specific conditions (see below; see also Waage, 1979; Lima, 1984; Haccou et al., 1991; Driessen & Bernstein, 1999; Outreman et al., 2005).

Whether genetic or plastic, the interpopulation variation in weevil foraging behaviour is structured along a gradient of the pericarp thickness of Japanese camellia, potentially explaining the interpopulation differentiation of this plant trait (Figs 3–5). This suggests that differences in the foraging behaviour of predators are important determinants of variation in natural selection on prey traits and their resulting evolutionary divergence. Although the diversification of prey traits under different predation regimes has been intensively investigated, this study is the first to illuminate the possibility that variation in predator foraging behaviour, as well as the absence/presence of predators (Endler, 1980; Reznick & Endler, 1982; Benkman et al., 2003; Langerhans et al., 2004) or habitat features (e.g. industrial melanism; Brakefield, 1987), contributes to the adaptive divergence of prey populations.

Then, what factors affect interpopulation variation in the preference of camellia weevils for Japanese camellia fruit based on pericarp thickness? I predicted that female weevils basically preferred fruits with thicker pericarps to avoid aggregation of larvae, but in populations in which the probability of successful excavations was low, they did not select fruits with too-thick pericarps because attacking such fruits usually resulted in the failure of oviposition. On one hand, in populations in which the probability of successful excavations was high (e.g. Nara; Fig. 6), female weevils attacked thicker pericarps more frequently (Fig. 2) and the distribution of larvae was uniform against pericarp thickness (Fig. 4). On the other hand, in populations in which weevils are subject to a higher risk of failing to excavate a thick pericarp (e.g. Ashizuri, Reihoku, Takahama and Hanyama; Fig. 6), fruits with thick pericarps were avoided and thus those with intermediate pericarp thickness were attacked the most frequently (Fig. 2).

These interpopulation differences in foraging behaviour (Fig. 6) are concordant with the prediction that foraging patterns of ovipositing females are determined by the balance between optimal oviposition strategies to increase the performance of their offspring and optimal foraging tactics to increase the efficiency of oviposition (Scheirs et al., 2000; Mayhew, 2001; Scheirs & De Bruyn, 2002). Thus, depending on local conditions such as the probability of successful excavations, female weevils should choose adaptively among fruit based on pericarp thickness.

In conclusion, I demonstrated that adaptive foraging by predators could vary among populations due to biotic or abiotic conditions, and this variation possibly resulted in divergent natural selection on prey traits. The link between predator behaviour and prey evolution should be taken into account when examining the evolutionary trajectories of predator–prey interactions.

Acknowledgments

I thank M. Hasegawa, H. Hirano, H. Nishi and Y. Notsu for the collection of C. camelliae. Primate Research Institute (Kyoto University), Field Science Education and Research Center (Kyoto University), Amakusa Marine Biological Laboratory (Kyushu University), Saru-goya Foundation, G. Hanya, R. Tsujino and M. Yokoo have supported fieldwork for sampling. I am also grateful to O. K. Mikami, T. Sota and anonymous reviewers for productive comments on the manuscript. This study was partly supported by the Japan Society for the Promotion of Science (the Research Fellowship for Young Scientists; No. 1702263).

Ancillary