A series of selection analyses showed herein that natural selection driving interspecific interactions could vary among geographic populations separated by only several kilometers. Results further suggested that interpopulation variation in trait costs or climatic environment could promote the geographic differentiation of coevolutionary processes. Thus, these findings detailed below illuminate the possibility that evolutionary interactions between species are spatially far more dynamic than ever thought.
Potential of a coevolutionary arms race
Before considering weevil attacks as a major driving force of pericarp evolution, alternative factors that may also promote the evolution of thick pericarps should be considered. Importantly, camellia pericarps dehisce after seed maturation and seeds then fall to the ground; hence, pericarps are not an adaptation for attracting animals to disperse seeds (see ). Therefore, the adaptive function of pericarps is likely to be protecting seeds against abiotic or biotic environmental factors before seed dispersal. For example, desiccation or cold/frost damage to seeds might favor thicker pericarps. However, given that a sympatric congener of the Japanese camellia, Camellia sasanqua, bears fruit with a very thin pericarp (ca. 1 mm), thick pericarps should not be necessary to defend Japanese camellia seeds against abiotic environmental stresses. In addition, thicker pericarps are found in regions at lower latitudes or lower altitudes, where climatic conditions are relatively moderate [13, 38]. Alternatively, the evolution of thick pericarps may be driven by organisms that attack camellia seeds. Apart from weevils, the larvae of an unidentified lepidopteran species infest camellia seeds very rarely; however, its impact on the evolution of pericarp thickness is thought to be negligible (see Methods).
In contrast to abovementioned alternatives, the adaptive significance of having thick pericarps to defend against the camellia weevil is evident from the results. These clearly showed that the camellia weevil is a very important mortality factor for camellia seeds (Fig. 2C), and that camellias defend against weevil attacks, that is, excavation of pericarps, more effectively when they have thicker pericarps (Fig. 3; Table 2). Because this negative relationship between camellia pericarp thickness and the probability of successful weevil attacks has also been supported by a laboratory experiment , it is highly probable that camellias have evolved thicker pericarps to defend their seeds against weevils. Indeed, the proportion of surviving camellia seeds, which represents fitness of camellia individuals, was found to be positively correlated with pericarp thickness within Yakushima Island (Table 3). It is also noteworthy that the Japanese camellia has significantly thicker pericarps on islands with the camellia weevil than on islands where the weevil is absent (H. Toju, H. Abe and S. Ueno, unpubl.). Overall, these results indicate that defending seeds against the camellia weevil is the most plausible explanation for the evolution of thick camellia pericarps. In the following discussion, I postulate that the pericarp thickness of the Japanese camellia can evolve through natural selection exerted by the camellia weevil.
Spatial scale of the geographic selection mosaic
Previous studies [13, 28, 38] have shown that camellia pericarp thickness and weevil rostrum length varied remarkably within Yakushima Island (Table 1). At the extreme, the camellia and weevil traits were differentiated between populations separated by only several kilometers, and the sizes of these defensive and offensive traits were correlated across populations within the island . In addition to phenotypic variation, the success and severity of weevil attacks on camellias varied geographically within Yakushima Island (Figs. 2 and 7). These results suggest that the evolutionary interaction between the two species also varies at the same spatial scale.
The results first indicate that the benefit of having thicker pericarps in resisting severe weevil attacks was ubiquitous among camellia populations within Yakushima Island. In most populations, excluding Fukagawa, where the sample size of trial holes was small (Fig. 2), thicker pericarps were effective in defending against weevil excavations (Fig. 3). In addition, in the same five populations, camellia individuals with thicker pericarps were more likely to protect seeds from oviposition of weevils (Table 3; Fig. 5).
Nonetheless, several lines of evidence indicate that the strength of directional selection for thicker pericarps varies geographically. First, the camellia pericarp thickness at which weevil excavations are successful by the probability of 50% (i.e., BS
50) differed among populations (Fig. 4A), suggesting that the degree of pericarp evolution necessary to defend against weevils is geographically differentiated. This is presumably due to differences in the mean rostrum length of sympatric female weevils (Fig. 3). Second, natural selection analyses showed that camellias with thicker pericarp were favoured in five populations, while no significant selection was observed in the remaining population, Fukagawa (Table 3).
It is important to note that the evolution of thick pericarps may be restricted by the cost of having such a trait, despite the clear benefit of thick pericarps in defending against attacks by camellia weevils. For example, if the amount of photosynthetic products is limited, producing thick pericarps may reduce the number of fruits or seeds that can be produced. Therefore, I examined the geographic variation in natural selection pressures, using the number of surviving seeds as a measure of fitness. This should incorporate the effects of potential tradeoffs between the pericarp thickness of individual fruits and the number of fruits produced. Quantitative evaluation of natural selection indicated that natural selection pressures exerted on camellia pericarp thickness varied within Yakushima Island (Fig. 6; Table 4). Relatively strong directional selection (β
> 0.5; cf., ) was detected for pericarp thickness in Shiratani and Ohko-rindoh, while thicker pericarps were not favored in Kawahara and Fukagawa. Because camellias have considerably heavy pericarps in Kawahara compared to those in other populations (Table 5), producing pericarps would be more costly in the population. Indeed, the number of fruits produced by camellia individuals decreased with increasing pericarp thickness in Kawahara, although the correlation was marginally significant (Table 5). On the other hand, much less resources were allocated to pericarps in Fukagawa compared to other populations and the resource costs of producing pericarps may not explain nonsignificant directional selection observed in this population (Tables 4 and 5). Alternatively, the low frequency of weevil attacks (Fig. 2) and the resultant lack of benefit of having thick pericarps in this population (Fig. 5; Table 3; see also the opportunity for selection in Table 5) would be responsible for the lack of significant directional selection (Table 4; Fig. 6). Overall, these results indicate that the (co-)evolutionary process of the weevil-camellia interaction is structured at a spatial scale of several kilometers (e.g., Ohko-rindoh vs. Kawahara), as suggested by the phenotypic variations in the weevil and camellia traits within this island.
The extent of local adaptation, however, is influenced not only by local selection pressures but also by the levels of gene flow for both species. For the Japanese camellia, gene flow is expected to be localized at fine spatial scales because the home range of its major avian pollinator (Zosterops japonica; ) is estimated to be 0.3 ha , and seed dispersal by rodent dispersers is only by 5.8 m on average . Indeed, population genetic analyses using microsatellite markers have shown a significant positive relationship between genetic and geographic distance (i.e., isolation by distance) between adult trees at spatial scales of 100 m (; see also [41, 70, 71, 72]). In addition, for weevils, analysis of molecular variance  from the published data on mitochondrial DNA sequences  have revealed significant genetic differentiation in the six populations examined in this study (H. Toju and T. Sota, unpubl.). Pairwise interpopulation comparison of mitochondrial haplotypes of the weevils also indicated that gene flow is restricted, even between populations separated by only several kilometers (e.g., Fukagawa vs. Shiratani; H. Toju and T. Sota, unpubl.). Consequently, the evidence for localized gene flow for both species suggests that geographic selection mosaics can drive fine-scale phenotypic differentiation of camellia pericarp thickness and weevil rostrum length among populations, in spite of the homogenizing effects of gene flow (see [17, 74]).
Relationship between natural selection and levels of coevolutionary escalation
Although the geographic selection mosaic for camellia pericarp thickness indicates that the process of weevil-camellia coevolution is spatially structured, variation in the levels of coevolutionary escalation, as evaluated by camellia pericarp thickness and weevil rostrum length, were not consistent with variations in current selection pressures. That is, significant directional selection for thicker pericarps was not detected in Kawahara (Table 4; Fig. 6), in which the camellia and weevil traits have been already highly escalated (Table 1). The present direction and/or strength of natural selection, however, are not necessarily the same as those in the past . Given that the volume or weight of resources allocated to pericarps is proportional to cubed pericarp thickness, evolving per unit thickness of pericarps is expected to become increasingly costly as coevolutionary escalation proceeds. Therefore, although the benefit of having thicker pericarps is still significant in Kawahara (Table 3), considerable costs of resource allocation to pericarps may prevent further evolution (Tables 4 and 5). In fact, a cubic spline analysis showed that camellia individuals with intermediate pericarp thickness represent relatively high fitness as balanced by the benefit (Table 3) and cost (Table 5) of having thick pericarps, although relatively small sample size might prevent the detection of significant stabilizing selection (Additional file 2; ). Thus, the coevolutionary process may be at an equilibrium in Kawahara, whereas in Shiratani and Ohko-rindoh, relatively low resource costs may allow the evolution of thicker pericarps (Fig. 6; Table 5).
In addition to trait costs, two other factors may cause the negative or positive relationships between the levels of coevolutionary escalation and the strength of directional selection for thicker pericarps. One is suggested by the interpopulation variation in the difference between mean pericarp thickness and BS
50 (Fig. 4B). Because camellias have already evolved very thick pericarps to resist most weevil attacks in some populations (e.g., Hanyama and Kawahara; Figs. 3 and 4C), further evolution of pericarps may not effectively increase fitness in these camellia populations. This loss of the benefit of thicker pericarps, however, is not evident in the present data sets (see Kawahara in Table 3). The second factor is the dependence of weevil behavior, that is, the preference of female weevils for camellia fruits based on pericarp thickness, on the levels of coevolutionary escalation . A previous study showed that the preference of weevils varied among populations, and that interpopulation variation in weevil behavior is an important determinant of the geographic selection mosaic for camellia pericarp thickness . Importantly, female weevils avoid attacking fruits with relatively thick pericarps in populations in which they are subject to higher risks of failure in attacking such fruits due to low average probability of successful excavations . Given that the proportion of successful excavations decreases as coevolutionary escalation proceeds (; see also Fig. 4C; Table 1), presumably due to the mortality costs of weevils with long rostra (cf., ) and the resultant limitation in the evolution of this weevil trait, thicker pericarps might further increase the fitness of camellias. Indeed, camellia fruits with relatively thick pericarps were avoided by weevils in Ohko-rindoh and Shiratani (Toju, unpubl.), leading to further directional selection (Table 4), but were preferred in Fukagawa, preventing the evolution of thick pericarps. No significant tendency was observed in Kawahara.
Environmental dependence of the weevil-camellia coevolution
Geostatistical analyses confirmed that the attacks of weevils on camellia fruits varied at a fine scale within Yakushima Island (Fig. 7). Moreover, regression analyses showed that the proportion of successful pericarp excavations by weevils and the proportion of infested seeds increased with decreasing temperature within the island (Table 6).
The environmental clines of the weevil-camellia interaction are observed not only within Yakushima Island, but also across the Japanese archipelago [13, 28, 39]. Examinations of altitudinal gradients within this island, and latitudinal gradients over the whole of Japan, have revealed that the proportion of successful excavations and the proportion of infested seeds increase in the cooler-temperate regions (Table 6; Fig. 1G, I; ). Analyses at within-Yakushima and entire-Japan scales also showed that the size of the camellia defensive trait, i.e. pericarp thickness, decreased with increasing altitude and latitude [13, 28, 38, 39]. Together, these results suggest that climatic factors (e.g., low temperature) limit the evolution of thick camellia pericarps in populations at higher altitudes and latitudes, despite the more severe impacts of seed infestation by weevils in these areas.
A putative factor responsible for the geographic differentiation is geographic variation in "productivity." Mathematical models of predator-prey or parasitoid-host coevolution predict that environmental factors influencing the fecundity, density-dependent factors limiting population dynamics, and the costs of defensive traits of victims determine the occurrence of arms races or the levels of coevolutionary escalation ([77–79]; N. Iseki, H. Toju and A. Sasaki, unpubl.). Although these predictions have not yet been demonstrated in the wild (cf., the laboratory study of ), the influence of productivity on local coevolutionary dynamics [35, 80] is expected in the weevil-camellia system. For example, because the photosynthetic capacity of camellias increases exponentially with increasing temperature , local climate would affect the optimal resource allocation strategy  (sensu ) or fecundity of camellias, thereby leading to the spatial variation in the levels of coevolutionary escalation (N. Iseki, H. Toju and A. Sasaki, unpubl.). Consequently, findings in the weevil-camellia system support the hypothesis that spatial variation in climate is a major factor driving the geographic differentiation of ecological and evolutionary interactions between species [84–87].
In contrast to the potential dependence of the weevil-camellia interaction on temperature, the relationship between annual precipitation and the variables representing weevil attacks (Table 6) was not straightforward. Because the mean of annual precipitation experienced by sampled individuals is very high (3809.9 mm; SD = 297.0 mm), the photosynthetic activity of the Japanese camellia is unlikely to be affected by the availability of water. Thus, given the negative correlation between annual mean temperature and annual precipitation (see Methods), the association between precipitation and the nature of weevil attacks may be a statistical artifact.