Knowledge of the natural variation of a trait is the basis for studying the microevolution of the trait. Our study aimed to assess the wild type shape space of the mouse mandible, which has become a major model for the evolution of morphological characters. However, dealing with wild-caught animals implies that these will differ with respect to age, size and previous diet. In addition, dealing with museum specimens could introduce further systematic differences caused by different preparation conditions, which are often not recorded. Apart of diet (see below) a systematic study on these possibly confounding factors has not been done before, but was evidently necessary, given that we record rather subtle differences. We have therefore first designed a number of tests to look at the influence of technical variables, such as preparation method of the skull, suboptimal sample size and digitization error. In addition we have looked at the influence of biological variables, such as age and size. The intuitively most important environmental factor that could influence mandible shape is diet and we have therefore paid particular attention to this. Previous studies on the influence of diet differences had found that food consistency has an effect on mandible shape [15, 16, 35–37]. But we have asked here for the first time how strongly these effects might confound measured differences between wild mouse populations. We found that there is indeed an effect of diet on mandible shape, but even in the extreme hard diet/soft diet shift experiment the effect is not more dramatic than for age or size differences, and the Procrustes distances between diet treatment groups are also smaller than the majority of intraspecific Procrustes distances. Furthermore, since house mice are generalists, which are not prone to specialize on any specific diet, we assume that extreme diet consistency differences are rare in nature (but see the case of Kerguelen mice below).
While each of these tests and experiments showed only a small influence on Procrustes distances between populations, it could well be that taking together all error sources and environmental factors would introduce more error in our data than any single error source. However, it is possible to use the shape differences between "treatment" and "control" groups in these experiments to calculate combined effects. We made use of this option to simulate a distribution of Procrustes distances under various combinations of error and environmental effects. This simulation is conservative in as far as it assumes cumulative effects of age, size, and diet treatments, whereas a smaller number of factors is likely to be relevant in real populations. Even under this conservative model, the combined error and environmental effect are unlikely to explain the average Procrustes distance between natural populations.
Given that the differences we have measured between the wild populations can apparently not fully be ascribed to technical and environmental factors, it seems safe to conclude that genetic factors contribute significantly to mandible shape differences. This conclusion is further supported by the fact that major shape differences exist between inbred strains that were derived from the same wild population at about the same time in the same laboratory (inbred strains Stlt, StrB and StrA, see below), i.e. the environmental influences should have been very similar in this case, but this is not the case: major differences are evident (see below). Furthermore, our assessment of the ontogenetic origin of shape differences between inbred strains suggests that they become manifest early in postnatal development, weeks before weaning, such that dietary differences acting later in life would only modulate already existing differences, rather than initiating them in the first place. Thus, the intraspecific diversity in mandible shape in the house mouse is apparently to a high degree genetically determined and can therefore be interpreted in an evolutionary context, involving an assessment of the role of selection and neutral evolution in shape divergence.
The role of selection
Morphological traits in adults, such as mandible shape, are expected to be directly exposed to purifying selection, since their integrity should directly contribute to the fitness of the individual. This would imply that differences between populations should be driven by positive selection to new environmental conditions. However, neutral accumulation of differences over time would also be a possibility, in particular since there appears to be little gene flow even between populations of the same subspecies . Some of the patterns that we see in our data allow to address this question.
The strongest indicator for selective constraints and only a small role for neutral divergence is the finding of relative higher similarity of shapes for M. m. domesticus populations from summer-dry habitats in Iran, Egypt and Sicily. This is of particular relevance, since these populations are at the same time the ones that are the relatively oldest ones. Based on molecular and fossil evidence, it has been shown that M. m. domesticus mice started to spread from the area of Iran into the Near East and Northern Africa approximately 8, 000 years ago, while the colonization of Sicily and Western Europe started only 3, 000 years ago . Thus, M. m. domesticus has originated in a summer-dry climate and is expected to be optimally adapted to this. If one would assume a neutral divergence of shapes, one would expect that old populations are more different from each other than young populations, but this does not appear to be the case here. Instead, the younger populations in Western Europe show more morphological divergence than the older populations from Africa and the Near East. This divergence could be due to adaptive effects in the wake of colonizing new environments with new food sources, or could be due to the fixation of new alleles during the colonisation bottlenecks. At present it is difficult to distinguish between these possibilities, but one can point out that the molecular analysis of the populations has suggested that the colonisation bottleneck was not very strong .
The most divergent populations among M. m. domesticus are the ones that were caught on the sub-Antarctic islands of Guillou and Cochon, which are both part of the Kerguelen Archipelago where mice have arrived less than 200 years ago . They occupy a new part of the shape space, indicated by their separation on the second PCA axis in comparison to the other wild-caught mice (Figure 4a). It is indeed known that these island mice have adapted to a new diet (preference for earth worms, which are otherwise not much used by mice , i.e. the common mandible shape difference along PC2 could have been caused by positive selection. On the other hand, this does not explain why they differ so much in PC1 (Figure 4a), although they live on ecologically very similar neighbouring islands. However, these mice came actually from two genetically distinct source populations , which could explain these differences. This would therefore be a case where parallel selection has led to some common changes (represented in PC2), but within different genetic backgrounds, which are responsible for the differences represented in PC1. While this interpretation is necessarily speculative (and does also not rule out additional bottleneck effects), it also corroborates the findings that we made in our companion paper on comparisons between mandible shape QTLs obtained from different experiments .
Another indicator of a possible influence of selection on shape comes from the finding that the M. m. domesticus population that lives in sympatry with a M. spretus population (DOM SPAIN PUDEMONT) shows a shift in shape space. In the PCA plot these mice are on opposite trajectories outside of the shape space of the other mainland M. m. domesticus populations, including the ones from the summer-dry regions, as well as from Germany (Figure 4a). The two species of mice included in this dataset were caught in the same small village in a single trapping campaign, suggesting that they share the same habitat. Hence, one could interpret the divergence of the M. m. domesticus shapes as a character displacement effect, in response to the interaction with M. spretus, which is the ancestral species in the region. However, this situation will need to be further analysed before a firm conclusion is possible, since it is so far not known in which form the two species might ecologically interact with each other.
To interpret shape differences between inbred strains in an evolutionary context, which means, with reference to natural variation, it is critical to understand which level of genetic diversity they reflect. Inbred strains are genetically comparable to individuals since they represent a single genotype only. Any variance between individuals from an inbred strain should therefore be ascribed to technical and environmental variance, while the strain mean shapes represent the genotype of these "individuals". We find that the within-strain variance of inbred strains is indeed much lower than the within-population variance of wild caught mice (Figure 5). Inbred strains represent thus essentially a random sample of chromosomes drawn from specific wild populations or admixed from various origins.
We find that inbred strains, which have been derived from wild populations, differ from each other more than the mean shapes between wild populations. This includes inbred strains, which have been derived from the same natural population, as is the case for the strains Stlt, StrA and StrB in our study. The Procrustes distances between these three strains (0.0373 - 0.0446) are as large as the larger interspecific distances between wild populations. The comparison of between- and within-population variation in wild mice (Figure 5) shows that inter-individual variation within natural population is much larger in comparison to differences between populations. The opposite pattern is seen for inbred mice (Figure 5) where the within strain differences are small due to inbreeding. The mandible shapes of inbred strains appear thus indeed to represent random samples of the original wild diversity, i.e. they "behave" as if they were individuals drawn from wild populations. This interpretation does of course not preclude possible additional changes due to inbreeding and laboratory evolution.
Although the mandible shapes of the inbred strains in our study are well within the range of natural variation, their distribution in Figure 6 suggests that there may be some pattern. The variation among inbred strains seems to be limited along PC2, while PC4 separates classical strains from wild-derived ones. While the former may be ascribed to laboratory effects, it is more difficult to speculate about a reason for this latter pattern. Perhaps it is partially due to phylogeny, since the classical strains have ultimately all been derived from the same admixed base population . These questions will have to be revisited in the future.