Discrete or indiscrete? Redefining the colour polymorphism of the land snail Cepaea nemoralis

Biologists have long tried to describe and name the different phenotypes that make up the shell polymorphism of the land snail Cepaea nemoralis. Traditionally, the view is that the ground colour of the shell is one of a few major colour classes, either yellow, pink or brown, but in practise it is frequently difficult to distinguish the colours, and define different shades of the same colour. To understand whether colour variation is in reality continuous, and to investigate how the variation may be perceived by an avian predator, we applied psychophysical models of colour vision to shell reflectance measures. We found that both achromatic and chromatic variation are indiscrete in Cepaea nemoralis, being continuously distributed over many perceptual units. Nonetheless, clustering analysis based on the density of the distribution did reveal three groups, roughly corresponding to human-perceived yellow, pink and brown shells. We also found large-scale geographic variation in the frequency of these groups across Europe, and some covariance between shell colour and banding patterns. Although further studies are necessary, the observation of continuous variation in colour is intriguing because the traditional theory is that the underlying supergene that determines colour has evolved to prevent phenotypes from “dissolving” into continuous trait distributions. The findings thus have significance for understanding the Cepaea polymorphism, and the nature of the selection that acts upon it, as well as more generally highlighting the need to measure colour objectively in other systems.

reflectance spectrum and a reference (a very low value of double cone quantum 179 catch, 0.001), corresponding to a dark spectrum, and using the same noise-to-signal 180 ratio.

181
Analysis of morph frequencies 182 We investigated evidence for effects of location and banding on the likelihood that a 183 snail belonged to a particular morph, using generalised linear mixed effects models 184 (GLMMs) with binomial errors. Each morph was considered separately, with each 185 snail to be scored as belonging to the focal morph (1) or not (0). The three analyses 186 are not independent, since each snail can only belong to one morph. Banding 187 pattern was fitted as a fixed factor, whilst the effect of geographic location was 188 examined at three spatial scales. Variation in morph frequency at a local level was 189 modelled with random effect for site. Variation at a regional level was considered by 190 fitting a fixed effect of geographic region. Finally, continental scale variation was 191 modelled by looking for fixed linear and quadratic effects of latitude and longitude.

192
The fact that region and latitude/longitude are partially collinear was reflected in the 193 model-fitting procedure. We first fitted a saturated model with all main effects, except 194 for region, and their two-way interactions (excluding interactions involving quadratic 195 effects). Then, fixed terms were removed in a stepwise fashion, testing the effect of 196 deletion using likelihood ratio tests, until only significant terms remained. Effects of 197 latitude/longitude were then substituted with an effect of region and we compared the 198 Akaike Information Criterion (AIC) of the resulting models, to test if region was better 199 at capturing any large-scale geographic variation. Testing random effects in 200 generalised linear mixed models is problematic, so we compared the AIC of the 201 saturated GLMM with that of a generalised linear model without the random term for 202 site to provide an approximate test of the importance of site.

204
Variation in colour 205 We measured the individual reflectance spectra of 1172 shells, mainly collected from 206 across Europe (Table S1; Figure 1) and then transformed them into visual space   Table 1). The next best fitting model also recovered three clusters (VEV; BIC -232 15749.3; P < 0.001 compared with 3 rd best model) and the third recovered four 233 clusters (EEV; BIC -15755.5; the 4 th cluster contained only 16 individuals).
In our analysis, we found that chromatic variation in shells is continuously 299 distributed in visual space, meaning that there are no wholly discrete colours ( Figure   300 2). Perhaps surprisingly, we found that the most variable chromatic axis (PC1; 87%) 301 that would be visible to a bird reflects the degree of saturation, or purity of colour.

305
Despite the lack of discrete colours, density-based clustering recovered three 306 main shell types, which roughly correspond to human-perceived yellow, pink and 307 brown (Table 1; Figure 3). Brown shells were more common according to the 308 objective analysis than perceived by humans, with the frequency higher again when 309 using woodland shade as an illuminant. Therefore, prior studies that (necessarily) 310 used changes in frequencies of human-perceived colours to understand natural 311 selection on snail shells may have missed a significant part of the picturenot only 312 may birds use both achromatic and chromatic cues to differentiate morphs, but they 313 should also be able to perceive chromatic differences to a much finer precision than 314 a simple trivariate yellow, pink or brown categorisation that humans are obliged to 315 use in qualitative surveys. Of course, this does not mean that birds react to the many 316 morphs equallyit is possible that they categorically perceive a continuous variable  The effects of geographic location and banding pattern on variation in the 320 reflectance spectrum of snails were also examined, the initial aim being to develop Cepaea snails became a well-studied system. However, while scoring the shell 343 colour into different, discrete types is straightforward in offspring of individual crosses 344 in the lab, the acknowledged reality is that it is sometimes difficult to classify shells 345 consistently (e.g. see Table 1), especially when they are apparently intermediate in