Rule reversal: Ecogeographical patterns of body size variation in the common treeshrew (Mammalia, Scandentia)

Abstract There are a number of ecogeographical “rules” that describe patterns of geographical variation among organisms. The island rule predicts that populations of larger mammals on islands evolve smaller mean body size than their mainland counterparts, whereas smaller‐bodied mammals evolve larger size. Bergmann's rule predicts that populations of a species in colder climates (generally at higher latitudes) have larger mean body sizes than conspecifics in warmer climates (at lower latitudes). These two rules are rarely tested together and neither has been rigorously tested in treeshrews, a clade of small‐bodied mammals in their own order (Scandentia) broadly distributed in mainland Southeast Asia and on islands throughout much of the Sunda Shelf. The common treeshrew, Tupaia glis, is an excellent candidate for study and was used to test these two rules simultaneously for the first time in treeshrews. This species is distributed on the Malay Peninsula and several offshore islands east, west, and south of the mainland. Using craniodental dimensions as a proxy for body size, we investigated how island size, distance from the mainland, and maximum sea depth between the mainland and the islands relate to body size of 13 insular T. glis populations while also controlling for latitude and correlation among variables. We found a strong negative effect of latitude on body size in the common treeshrew, indicating the inverse of Bergmann's rule. We did not detect any overall difference in body size between the island and mainland populations. However, there was an effect of island area and maximum sea depth on body size among island populations. Although there is a strong latitudinal effect on body size, neither Bergmann's rule nor the island rule applies to the common treeshrew. The results of our analyses demonstrate the necessity of assessing multiple variables simultaneously in studies of ecogeographical rules.


| INTRODUCTION
Intraspecific geographical variation often presents vexing challenges to taxonomists, but such variation is essential for evolution and provides opportunities for insights into its underlying mechanisms.
Several environmental factors are known to drive geographical variation in morphology among mammals, including temperature gradients and isolation on islands or island-like features (e.g., Millien et al., 2006). The magnitude of the effect of these factors on morphological variation may differ across species' traits (e.g., Souto-Lima & Millien, 2014;Teplitsky & Millien, 2013). The resulting patterns of variation form the basis of a number of ecogeographical "rules" that attempt to describe such patterns and/or infer cause.
Island area and distance from the mainland-the latter often used as a proxy for the degree of isolation-are also correlated with body size on islands, with greater magnitude of size change expected on smaller, more isolated islands (Heaney, 1978;Lomolino, 2005;Millien, 2011).
Bergmann's rule describes the pattern in which species of a genus in colder climates (generally occurring at higher latitudes) have larger F I G U R E 1 Map of the Malay Peninsula and offshore islands discussed in the text (modified from Sargis et al., 2017). Black circles represent mainland localities (see Appendix 1) and this rule has been extended to populations within a species (e.g., Ashton, Tracy, & Queiroz, 2000;Bergmann, 1847;Freckleton, Harvey, & Pagel, 2003;Mayr, 1956;Meiri & Dayan, 2003;Millien et al., 2006).
Treeshrews (order Scandentia) are small-bodied mammals (adults weigh less than 315 g; Sargis, 2002) distributed across much of Southeast Asia, including on numerous islands (see Lyon, 1913;Roberts, Lanier, Sargis, & Olson, 2011). They are seemingly poor overwater dispersers, so their distribution on islands is presumably attributable to vicariance resulting from sea-level fluctuations in most cases (Olson, Sargis, & Martin, 2005;Roberts et al., 2011). Although they have been included in some taxonomically broad tests of the island rule (e.g., Lomolino et al., 2013;Meiri et al., 2008) into account in these meta-analyses. Furthermore, Bergmann's rule has never been tested in treeshrews, nor has the possible interaction between these two rules (e.g., Fooden & Albrecht, 1993).
The common treeshrew, Tupaia glis (Diard, 1820), is widespread throughout the Malay Peninsula, from about 1.25º to about 10°N latitude, and occurs on a large number of adjacent offshore islands that extend its distribution south to about 0.8°N latitude (Figure 1; Sargis, Woodman, Morningstar, Bell, & Olson, 2017). These are primarily continental islands that were connected to one another and to the mainland during the Pleistocene and became isolated as sea level rose following the Last Glacial Maximum (Voris, 2000). The common treeshrew averages about 152 g (Sargis, 2002), making it a particularly suitable taxon for testing the island rule, as it is between the hypothesized "optimal" mammalian body masses of 100 and 1000 g predicted by Brown, Marquet, and Taper (1993) and Damuth (1993), respectively. As part of our ongoing study of treeshrews, we investigated how island size, maximum modern sea depth between the island and the mainland, and distance from the mainland relate to body size of insular T. glis populations, while controlling for the additional effect of latitude on body size in this species.  (Sargis, Woodman, Morningstar, et al., 2013;Sargis, Woodman, et al., 2014;Sargis, Campbell, et al., 2014;Sargis et al., 2017) from adult skulls (those with fully erupted permanent dentition) using digital calipers. Total length and body weight were recorded from skin tags or the original field notes of collectors. All craniomandibular measurements are in millimeters and were measured to the nearest 0.01 mm; they are tabled in Appendix S1. Our sample includes the holotypes of 10 species or subspecies. Summary statistics of craniomandibular and external variables, including mean, range, 95% confidence interval, standard deviation, coefficient of variation, and percent not available, are presented in Table 3.

| MATERIALS AND METHODS
The craniomandibular dataset included a significant amount of missing data (15.8%) resulting from damaged or incomplete specimens that would have prevented the statistical analysis of our complete dataset. The missing cases were thus imputed using the "mice" method with predictive mean matching, as described in Clavel, Merceron, and Escarguel (2014). This method uses model estimates to fill in missing values in a dataset with multiple imputations. We performed our analyses on data averaged from 50 independent imputations.
We then performed a principal components analysis (PCA) on the 22 ln-transformed craniomandibular variables and retained the first component (PC1) for further analyses. Next, we tested for overall sexual dimorphism in PC1 on the complete dataset, with a Welch twosample mean comparison test. Sex was included as a variable in all subsequent models as a factor to account for sexual dimorphism.
To test for Bergmann's rule (i.e., a significant positive relationship between body size and latitude) and an island effect on body size simultaneously, we ran a mixed-effect model on the entire dataset (island and mainland populations) with Latitude, Source of the population (island versus mainland), and Sex as fixed factors, and Locality (13 island localities and mainland) as a random factor. We also tested whether the latitudinal trend differed between island and mainland populations (interaction term between the variables Latitude and Source), and whether sexual dimorphism was different between island and mainland populations (interaction term between the variables Sex and Source). Variance inflation factors for our fixed effects were estimated (checking for values below a threshold of 2.5), and diagnostic plots (residual plots and q-q plots) were used to evaluate the fit of the model.
We further explored the effect of a number of additional factors on the variation in body size among the 13 island populations. We used Distance to Mainland as a measure of isolation, but also included Sea Depth in this analysis. Sea Depth is relevant here in the context of the paleogeography of the islands and how they formed.
Because these islands are the result of sea level rise following the Last Glacial Maximum (Voris, 2000), sea depth between the islands and the mainland can be viewed as a proxy for time since isolation and not simply degree of geographical isolation. We used a hierar- Because of collinearity in the explanatory variables, we did not attempt to model or estimate the strength of the main effects, but instead focused on determining the sign of these effects (i.e., positive or negative) because we were interested in testing the generality of ecogeographical patterns.

| RESULTS
The first component (PC1) from our PCA of skull variables yielded an eigenvalue of 15.62 and explained 71% of the variance in the skull data (Table 4). PC1 was negatively correlated with all 22 craniodental T A B L E 3 Summary statistics for the 22 skull measurements. Abbreviations for measurements are defined in Table 2. Statistics are sample size (n), mean, range, 95% confidence interval (CI), standard deviation (SD), coefficient of variation (CV), and percent not available (%NA)  (Table 4), so higher scores represent smaller skull sizes.
We detected significant sexual dimorphism in PC1 (t = 2.40, p < .0172). Overall, males were larger and had lower PC1 scores than females. Therefore, we included Sex as a factor in all subsequent analyses.
A mixed-effect model with PC1 as a response variable revealed a significant positive effect of Latitude (t = 3.85, p < .0027; Table 5, Figure 2), indicating decreasing body size with increasing latitude.
The strength of this relationship did not differ between islands and the mainland (p = .79). As already detected with the mean comparison test, males appeared to be larger in body size (smaller PC1) than females (t = −2.48, p < .014) independent of the Source of the popula- Mahalanobis distances (all p < .001, 1,000 permutations; Figure 3).

| DISCUSSION
The common treeshrew, T. glis, is endemic to Southeast Asia, where it is an abundant and common component of tropical forest mammal assemblages in the region. Although it has featured prominently in biomedical research (Fuchs & Corbach-Söhle, 2010), it remains poorly studied in the wild. Here, we conducted a review of patterns of morphological variation in the skull of this species to test the validity of two ecogeographical rules that have been described and widely tested in other mammals.

| Sexual size dimorphism
Although sexual dimorphism is variably expressed in the coat color of the common treeshrew (Steele, 1983), we found that males had significantly lower PC1 scores and are therefore significantly larger than females. A similar size disparity between the sexes has also been reported in the closely related T. belangeri (Collins & Tsang, 1987), demonstrating that sexual size dimorphism warrants consideration in future research on morphological variation in treeshrews.

| Bergmann's rule
Our results show that size variation in Tupaia glis from the Malay Peninsula and surrounding islands follows a latitudinal gradient, with additional effects of isolation (estimated by minimum distance to the mainland and maximum sea depth between the island and the mainland) and island area in the island populations. In the common treeshrew, body size decreases with increasing latitude, which is the inverse of Bergmann's rule. Although Bergmann's rule is supported by a number of empirical studies in mammals (Ashton et al., 2000;Meiri & Dayan, 2003;Millien et al., 2006), the pattern may be less apparent in temperate large mammals (Steudel, Porter, & Sher, 1994) or subterranean species that live in a more stable environment (Medina, Martí, & Bidau, 2007); however, none of these characteristics applies to the common treeshrew. In their review, Alhajeri and Steppan (2016) detected a weak positive relationship between body mass and temperature among more than 1300 rodent species, but this relationship did not hold when the phylogenetic structure in the data was considered. Instead, larger body mass was related to increasing precipitation (Alhajeri & Steppan, 2016), supporting some previous studies (James, 1970;Yom-Tov & Geffen, 2006). Alhajeri and Steppan (2016) conducted their study among species within a single mammalian order and concluded that Bergmann's rule may operate within a species (e.g., Ashton et al., 2000;Meiri & Dayan, 2003;Millien et al., 2006) rather than among species within a genus (Bergmann, 1847). Supporting this view, Albrecht (1980) (Mayr, 1956), although other factors such as resource availability (McNab, 2010) and levels of competition and predation have been invoked, leading Watt, Mitchell, and Salewski (2010) to designate Bergmann's rule as a "concept cluster."

| Island rule
The single variable most strongly related to body size in our study is latitude. Tupaia glis is a generalist, feeding on arthropods, fruits, leaves, seeds, and small vertebrates (Nowak, 1999). With a mean weight of 152 g (Sargis, 2002), the common treeshrew falls well within the range of "small" mammals (Merritt, 2010). For small-bodied species, the island rule predicts the evolution of larger body size on islands (Foster, 1964;Lomolino, 1985Lomolino, , 2005Van Valen, 1973). Here, only two of the island populations of common treeshrews exhibit larger body size than on the mainland as predicted by the island rule (e.g., populations from Batam and Bintan islands, both located at the most southern latitude in our study area). However, when we controlled for latitude, none of the populations from the offshore islands around the Malay Peninsula differ in size from the mainland population. The island rule, like Bergmann's rule, may prove to be taxon-specific in mammals (Meiri et al., 2008), with species within a given order typically following a common pattern. For example, rodent species, with some exceptions, typically evolve larger body size on islands (Durst & Roth, 2015), and patterns found among primates support the island rule as well (Bromham & Cardillo, 2007;Welch, 2009). In contrast, the apparent lack of support for the island rule in the common treeshrew may prove to be the common pattern throughout Scandentia.
This contrast is particularly interesting given the close relationship of treeshrews and primates in the supra-ordinal grouping Euarchonta (e.g., O'Leary et al., 2013). Further testing of the island rule across Scandentia and in Dermoptera (colugos), another euarchontan order that has a similar Southeast Asian distribution, should provide unique insight into these patterns of insular body size variation.
When considering solely insular populations, we found that the secondary driver of T. glis body size, after latitude, is maximum sea depth between the mainland and islands: body size of populations on islands separated from the mainland by deeper seas is typically larger. Island area has a tertiary effect on body size: The smaller the island, the smaller the individuals on that island. Hence, common treeshrews are smaller on smaller islands, and the pattern is strongest for populations separated from the Malay Peninsula by shallower seas.
Body size of common treeshrews living on islands is positively correlated with island area, as generally predicted for mammals and other vertebrates (Heaney, 1978;Lomolino, 2005). This correlation of body size with island area was documented by Heaney (1978) for another Southeast Asian mammal, the Asian tri-colored squirrel (Callosciurus prevostii; ~350-400 g [Thorington, Koprowski, Steele, & Whatton, 2012]). Heaney (1978) predicted that mammals that are slightly smaller than the Asian tri-colored squirrel, such as the common treeshrew, should increase in body size as island area increases, a prediction supported by our study. This hypothesis may not apply to larger F I G U R E 3 Bivariate plot of the first two axes from canonical variate analysis of the 22 skull variables from the 13 island populations grouped by region as described in Table 1 (see also species that are more resource-limited. Schillaci et al. (2009) found that body size is not related to island area in the long-tailed macaque (M. fascicularis; ~2.5-8.3 kg [Fa, 1989;]) from this region; in fact, longtailed macaque populations from Singapore and Bintan (see Table 1 for areas) both exhibit insular dwarfism, possibly related to food limitation and high population density (Fooden & Albrecht, 1993;Schillaci et al., 2007).
Such variation in body size patterns might be expected in a single species distributed among several island groups (see below ;Fooden & Albrecht, 1993), especially a species that would fit the "intermediate" category in Heaney's (1978, figure 3) size classification, such as T. glis. As Heaney (1978) suggested, mammal species of intermediate size may (i) not vary in their body size pattern, (ii) always converge on the pattern of either a large or small mammal, or (iii) converge on the pattern of either a large or small mammal depending on the conditions. He concluded that the Asian tri-colored squirrel demonstrated the third option (Heaney, 1978), and this may be the case for the common treeshrew as well, given the assumption of vicariance.

| Island group differentiation
The relative effects of island area and isolation acting on body size evolution in island populations of common treeshrews are difficult to tease apart. Factors that could influence body size include the relative timing of establishment on these islands and the different source populations. These factors could account for the clearly distinct morphology among the different island groups (Figure 3), irrespective of island area or degree of isolation. Fooden and Albrecht (1993) demonstrated similar variability among island groups and among islands within island groups in M. fascicularis throughout Southeast Asia, where different island populations variably exhibited a decrease, increase, or no change in skull length. Such variation was found both among and within island groups, and Fooden and Albrecht (1993, p. 533) attributed concordance among island populations to "common ancestry, parallel adaptation to local environmental conditions, or coincidence." These factors may also apply to the variation we found both among and within the western, eastern, and southern island populations in our study (see also Sargis et al., 2017), again, with the assumption of vicariance. Closer comparison of habitat and other conditions on these islands may reveal some critical thresholds in island size that affect the magnitude and direction of change in body size.

| CONCLUSIONS
Ecological factors such as resource limitation, intraspecific/interspecific competition, predation, and parasitism (Heaney, 1978;Lomolino, 2005) are all operating after the establishment of a population on an island. Unfortunately, this history for the islands in our study is not known, but this may have affected the patterns we documented here.
Future studies of the phylogeography of T. glis and other species on the mainland and offshore islands may provide relevant insight into the establishment of treeshrew populations on these islands as well as the demographic consequences. Furthermore, modern biological surveys of these islands have the potential to provide critical data on variation in species richness and population density that would allow a more thorough assessment of resource availability, ecological release from predation and parasitism, and both inter-and intraspecific competition. Finally, our study demonstrates the need for simultaneously testing potentially nonindependent ecogeographical patterns in broadly distributed taxa whose morphology may be influenced by multiple factors, a likely scenario for many species.