Snail Collecting, Identification and Maintenance

Individuals from 20 land snail species belonging to the Helicoidea superfamily were collected in France and Italy during the spring and summer of 2012 (Table 1, Fig. S1, Supporting information). All species were collected in compliance with local and EU laws. Taxonomic identification was based on European field guides (Kerney & Cameron 1999) and faunas (Germain 1930; Cossignani & Cossignani 1995; Gargominy & Ripken 2011; Welter-Schultes 2012), using shell morphology and, when necessary, genitalia characteristics. Based on their size, snails have been classed as either ‘small’ (minimal greater shell size <15 mm) or ‘large’ (minimal greater shell size ≥15 mm) species (Table 1), with experimental conditions slightly differing between the two groups to ensure the movements of all species were measured with similar accuracy. Before experiments, snails were kept in controlled conditions (18 °C, 80% R.H., 16:8 L:D photoperiod). They were housed by groups of five conspecific individuals in polythene boxes (9425 cm³ and 1334 cm² of available surface in the ‘large’ group, 471·2 cm³ and 345·6 cm² of available surface in the ‘small’ group) covered with a plastic mesh. Box sizes were chosen so that the mean snail mass per available m³ (Daguzan et al. 1982) did not differ significantly between boxes containing ‘small’ or ‘large’ species (median in small boxes = 0·58 g dm⁻³; in large boxes = 0·67 g dm⁻³; Wilcoxon rank-sum test, W = 732, P = 0·99). To provide a moist substrate and a source of water for the snails, the bottom of each box was lined with synthetic absorbent foam kept saturated in water. Snails were fed ad libitum with a composite snail food (cereal powder supplemented with calcium; Ets Berton, France). This resource was eaten readily by all species, by contrast with previously tested non-helicoid snail species (Ansart et al. 2014). Each snail was individually marked on the shell either by a plastic numeric coloured mark glued with cyanoacrylate (in the ‘large’ group) or by a coloured paint dot (in the ‘small’ group) (Henry & Jarne 2007).

Movement Data Acquisition

Individual movements were recorded by time-lapse photography in an empty arena lined with synthetic foam saturated in water (Fig. 1). To limit the influence of mucus trails and odours on activity (Dan & Bailey 1982), this lining was thoroughly washed after each monitoring session and changed between species, and the room ventilated. The room was under red light, as helicoid snails are mainly active at night (Cook 2001) and lack of photoreceptors for wave lengths higher than 580 nm (Chernorizov et al. 1994). Images were captured with USB cameras fixed to the ceiling (Hercules^® 640 × 480 camera; Carentoir, France, in the ‘small’ group and Logitech^® 1280 × 960 pixels camera, Morges, Switzerland, in the ‘large’ group), which allowed the coverage of a square of side length 170 cm at a 0·20 cm per pixel resolution for the ‘large’ group (50 cm side length and 0·10 cm per pixel resolution for the ‘small’ group). Resolution was higher in the ‘small’ group as small, whitish species were more difficult to pinpoint on the arena floor. In each experimental session, five conspecific snails held within a same box (see above) were tested at the same time. At the beginning of the experimental session, both the snails and the synthetic lining (hereafter termed the ‘patch’) were taken out of their box and put at the centre of the arena. This system allowed the separation of the arena in two zones (Fig. 1): snails began the monitoring on the patch, which was a familiar substrate in which mucous trails and other familiar odours were present and then could freely move to the unfamiliar area (the rest of the arena), with the same physical substrate but no indices of snail presence. The boundary between the patch and the unfamiliar area was marked by a small step (3 mm high), and both zones were deprived of any food source. Snails were then followed at a frame rate of two pictures per minute during 1 h or until they left the camera field. Preliminary tests have shown this set-up to be a good trade-off between the need to avoid autocorrelation between successive steps (Turchin 1998) and the need to have enough time intervals per trajectory and enough individuals (Table 1) to do a statistically meaningful analysis.

Individual coordinates over time were extracted from the recorded images using the Manual Tracking plugin in the imagej software (Abramoff, Magelhaes & Ram 2004; Cordelières 2005). Because the head of a snail can turn and retract without meaningful movement from the rest of the body, its position cannot be used to describe the individual displacement over time; we instead used the coordinates of the foremost point of the shell. The r package adehabitatLT (Calenge, Dray & Royer-Carenzi 2009) was then used to transform these coordinates into trajectories, that is a succession of steps of various lengths and turn angles (Turchin 1998). When the recorded distance between two successive relocations was lower than the image resolution (i.e. 0·20 cm in the ‘large’ group, 0·10 cm in the ‘small’ group), telling actual movement apart from location errors due to camera noise was considered impossible. In this case, we replaced the coordinates of the relocation i by the coordinates of the relocation i-1.

Habitat Specialization

Association of each species with the six top level terrestrial natural habitats in the EUNIS classification was recorded (B: Coastal habitats, D: Mires, bogs and fens, E: Grasslands and lands dominated by forbs, mosses or lichens, F: Heathland, scrub and tundra, G: Woodland, forest and other wooded land, H: Inland unvegetated or sparsely vegetated habitats; Louvel, Gaudillat & Poncet 2013). This excludes anthropic habitats such as agricultural and constructed/industrial habitats (I and J in the EUNIS classification, respectively). We used the fuzzy coding system proposed by Falkner et al. (2001), where 0 = species not recorded in the habitat, 1 = minor association, 2 = moderate association and 3 = strong association between the species and the habitat. Values of species–habitat association were obtained directly from Falkner et al. (2001) data base, completed by a combination of field guides and faunas (Cossignani & Cossignani 1995; Kerney & Cameron 1999; Welter-Schultes 2012), research articles (Yildirim, Kebapçi & Gümüş 2004; Aubry et al. 2005) and our own personal observations (Table S1, Supporting information). This species–habitat matrix was then used to calculate the Paired Difference Index (PDI), which is a robust specialization index taking into account not only the number of resources used by a species, but also the strength (relative or absolute) of the association between the species and its resources (Poisot et al. 2012). PDI values can range from 0 (complete generalist, i.e. identical non-zero performance in all habitats) to 1 (complete specialist, i.e. non-zero performance is only recorded in one habitat). Given their size, snail movement might be less influenced by macrohabitat requirements than by microhabitat features such as local vegetation, moisture, soil type or shading. However, the latter are much less extensively documented than the former in European land snails (microhabitat preferences are only available for 14 species in our 20 species sample, Falkner et al. 2001), and PDI values based on macrohabitat and microhabitat preferences are correlated (Pearson's r = 0·75, P = 0·002, for the aforementioned 14 species; see also Aubry et al. 2005). Therefore, we used PDI values based on macrohabitat preferences as a measure of ecological specialization.

Phenotypic Traits

During the marking process, snails were all weighted to the nearest 0·1 mg (CP224S balance, Sartorius, Göttingen, Germany). After movement monitoring, they were killed by freezing at −20 °C for ‘large’ species, or by drowning followed by freezing for ‘small’ species. Preliminary tests showed that prolonged drowning alters the quality of soft tissues, especially the foot, but strongly increases the probability of dying with an extended soft body, making the extraction of the body from the shell much easier in small species. However, as a consequence of tissues alteration during drowning, some snails were unusable for foot dissection, meaning that the number of feet analysed was lower than the number of snails in some species (Table S2, Supporting information). After thawing, soft tissues were removed from the shell, and the foot separated from the rest of the somatic part by cutting the elongated body from immediately below the mouth to the mantle cavity. Feet were then dried at 40 °C for 3 days before being weighted.

Molecular Data and Tree Construction

The methods followed here have been described in details in Guiller et al. (2001) and Ansart et al. (2014). Therefore, we only present a general outline of the protocol.

Statistical Analyses

All statistical analyses were carried out using r (version 3.0.2, R Development Core Team 2014).

Classical statistical methods assume that data points are independent, an assumption that can be violated in comparative analysis data sets, owing to the common ancestry of species (Felsenstein 1985; Revell 2010). To take into account the potential effect of phylogeny in our analyses, we used phylogenetic generalized least squares regressions (PGLS) to analyse the interspecific correlations between movement traits and specialization. To know to which extent phylogeny has to be accounted for, measures of the phylogenetic signal (‘the tendency of related species to resemble each other more than species drawn at random from the same tree’, Münkemüller et al. 2012) are needed. In this study, we used Pagel's λ, a scaling parameter applied to the internal branches of a tree (λ = 0: independence of data, i.e. ‘star’-like tree; λ = 1: trait evolution follows a Brownian motion (BM); 0 < λ < 1: phylogenetic signal, with an evolution model different of BM; Revell 2010; Münkemüller et al. 2012), and calculated it for each variable separately. We then used the pgls() function in the r package caper (Orme et al. 2012) to simultaneously estimate regression parameters and phylogenetic signal in residuals by maximum likelihood. As demonstrated by Revell (2010), this procedure performs better than a priori tests of phylogenetic signal to diagnose whether a phylogenetic correction is needed. Moreover, in the absence of phylogenetic signal, it performs identically or near-identically to classical OLS regression methods. All these analyses were done using mean species values.

We investigated the possible linear relationships between habitat specialization (PDI) and the importance of locomotor organs, that is foot dry mass. To account for interspecific differences in size, the species’ mean fresh body mass was included as a covariate. Body mass and foot dry mass were both log₁₀-transformed. Then, we analysed the effect of specialization on movement characteristics proper. Three movement variables were considered. First, the exploration probability was defined as the proportion of individuals of one species crossing the boundary between the familiar and unfamiliar areas (logit-transformed to comply with the normality conditions of PGLSs) and interpreted as a proxy for emigration tendency. The two other variables were linked to movement during explorations. We measured the speed at which snails moved during exploration (path length divided by exploration duration, hereafter ‘movement speed’). The mean path sinuosity during exploration was also calculated, following the formula proposed by Benhamou (2004). Together, these two variables influence the speed at which individuals diffuse away from a given site: speed being equal, animals moving in less tortuous paths disperse further; conversely, when two animals have the same sinuosity, the fastest will disperse further in a given time interval (Codling, Plank & Benhamou 2008). Indeed, in Cornu aspersum, interindividual differences in movement identified using this experimental set-up matched differences in dispersal patterns observed in semi-natural and natural settings (Dahirel et al. 2014, M. Dahirel, unpublished data). For each movement variable, our linear regression models contained two explanatory variables: habitat specialization (PDI) and the species’ mean fresh body mass (log₁₀-transformed). Non-significant terms (P > 0·05) were dropped from these models before results presentation or interpretation.

Phylogenetic Tree

The general typology of the tree was similar to the Helicoidea subtree of Ansart et al. (2014) (Fig. 2), with a first genetic gap separating two clades, one clustering the Helicidae and Elonidae families and one corresponding to the Hygromiidae sensu lato (Kerney & Cameron 1999; Bouchet et al. 2005). As in Ansart et al. (2014), tree resolution did not allow to completely resolve phylogenetic relationships within the Elonidae/Helicidae clade and polytomies appeared. The Hygromiidae s.l. side of the tree is fully resolved, but the monophyly of the Hygromiidae sensu stricto (minus Helicodontidae and Cochlicellidae) is not supported here (as suggested by Bouchet et al. 2005).

Phylogenetic Effect

Pagel's λ for log₁₀(body mass), foot dry mass and movement speed was equal to 1 and significantly different from 0 (P < 0·001), indicating phylogenetic signal for these traits, and did not differ from 1 (P = 1), indicating strong support for a BM model of evolution. For PDI, exploration propensity and path sinuosity, the estimated λ was equal to 0, but due to low sample size, without being significantly different from 1. For all dependent variables in PGLS models presented below (i.e. foot dry mass, exploration probability, speed and sinuosity), the phylogenetic signal in residuals, after taking into account the effect of dependent variables, was also estimated to be 0 and non-significantly different from 1.

Correlations Between Movement Traits and Specialization

Traits linked to movement during exploration were significantly correlated: faster than average species moved in straighter paths than slower species and also had bigger feet, after accounting for body mass (Table 2). On the other hand, correlations between exploration probability and other movement-related traits, including foot mass, were not significant.

Body mass being equal, habitat specialist species (high PDI values) had smaller feet than more generalist species (PGLS, P = 0·033, Fig. 3). There were exploring snails in all 20 species; the probability of exploration ranged from 0·15 (Helix ligata) to 0·90 (Arianta arbustorum). Habitat specialist species were less prone to explore than generalists (R² = 0·23, P = 0·018, Fig. 4). Mean movement speed during exploration ranged from 0·46 cm min⁻¹ in Helicodonta obvoluta to 4·11 cm min⁻¹ in Cepaea hortensis and was positively correlated with log₁₀(body mass); size being equal, habitat generalists were on average faster (Fig. 5, R² = 0·49, P_bodymass = 0·004, P_PDI = 0·012). The sinuosity of explorers’ paths was positively correlated with habitat specialization (Fig. 6, R² = 0·21, P = 0·023).

Phylogenetic Constraints on Body Mass Influence Movement Capacities

Phylogenetic signal varied between traits and interestingly, all phylogenetically constrained traits were positively correlated with log₁₀(body mass), with no residual phylogenetic signal when body mass was taken into account in regression models. Physical and morphological traits of organisms, and especially body size, are known to be largely conserved through evolution (Blomberg, Garland & Ives 2003). Thus, body mass appears to be phylogenetically conserved in land snails (this is apparent on Fig. 2) and, in turn, to strongly influence dispersal-related traits. This positive relationship between body mass and movement or dispersal capacities has been observed in many other animal groups (Jenkins et al. 2007; Stevens et al. 2014) and is based, among other things, on a decrease of the energetic cost of movement per distance unit with increasing body mass (Schmidt-Nielsen 1972). By contrast, exploration propensity and path sinuosity are not correlated with body mass and neither present significant phylogenetic signal. This might be because the decisions of leaving and turning are much less constrained by size-dependent costs than movement itself (even if the energetic costs of turning are not so negligible; Wilson et al. 2013). Although all movement-related traits we studied respond to environmental pressures to a degree (see below), the fact that speed is evolutionary constrained (via body mass), when exploration propensity and path sinuosity are more labile (as are many behavioural traits; Blomberg, Garland & Ives 2003), might have important consequences for the evolution of dispersal strategies. Indeed, it implies that adjustments of dispersal to rapidly changing conditions are more likely to occur via changes in emigration probability and path straightness, than by shifts in dispersal speed. However, movement speed and its potential consequences on time costs of dispersal remain understudied to date (Bonte et al. 2012).

Behavioural and Phenotypic Traits Correlate in a Movement/Habitat Specialization Syndrome

All movements traits studied here were correlated to habitat specialization: generalist species were more explorative, faster and moved in straighter paths. As a consequence, they are expected to emigrate more, disperse faster and further away from familiar sites, compared to specialists (Codling, Plank & Benhamou 2008). They also had, size being equal, bigger feet, a trait positively linked to movement speed during exploration. The existence of correlations linking both behavioural and morphological traits in a common syndrome naturally leads to questions on the determinants of dispersal and the trade-offs constraining dispersal evolution (Bonte et al. 2012; Ronce & Clobert 2012). Intuitively, dispersal can be seen at odds with ecological specialization, as an increase in one trait tends to disrupt benefits brought by the other: high dispersal increases the risk of ending in a different, less favourable habitat, thus decreasing the benefits of specialization (Bonte et al. 2003); and gene flow mediated by dispersal may interrupt ongoing local adaptation (Kisdi 2002). According to evolutionary models, this situation is expected to lead to the apparition of low-dispersal specialists and high-dispersal generalists (Kisdi 2002; Nurmi & Parvinen 2008, 2011). Indeed, after manipulating dispersal levels of bacteria in experimental heterogeneous micro-landscapes, Venail et al. (2008) observed the evolution of resource specialists when dispersal was limited and generalists in higher dispersal treatments. In addition, organisms living in environments where resources are scarce are expected to have lower metabolic rates (Mueller & Diamond 2001; Vignes et al. 2012), meaning that specialists may have slower metabolisms than generalists (e.g. Le Lann et al. 2012). This might influence locomotor performance (Niitepõld et al. 2009; Réale et al. 2010) and lead to decreased dispersal capacity with increased specialization, as observed in our study. Helicoid land snails are able to withstand long periods of dormancy, up to several months (Falkner et al. 2001). If there are metabolic differences between generalist and specialist snails, it might be interesting to see whether specialists are more likely to enter or survive dormancy, leading to a negative correlation between risk spreading in space and in time (Buoro & Carlson 2014), and adding a supplementary dimension to the observed dispersal/specialization syndrome. On the other hand, the aforementioned evolutionary models also predict the existence of non-monotonous relationships between dispersal and specialization under some conditions (Kisdi 2002; Nurmi & Parvinen 2008). There are indeed cases when specialists have maintained high dispersal, which compensates for the spatial scarcity of specialists’ resources (e.g. Centeno-Cuadros et al. 2011); however, negative correlations between specialization and dispersal seem to be the rule rather than the exception (Bonte et al. 2003; Venail et al. 2008; Stevens et al. 2014; the present study). This has important consequences for species conservation, as both specialization and low-dispersal ability are traits known to increase extinction probability, especially in the face of climate and land use changes (Kotiaho et al. 2005; Botts, Erasmus & Alexander 2013). Finally, another theoretical prediction on the joint evolution of specialization and dispersal is that the former should usually evolve faster than the latter, because the effects of dispersal on fecundity are relative on the conditions in both the departure and the arrival patch (Nurmi & Parvinen 2011). While our results showed that several movement traits are evolutionary constrained and habitat specialization more labile, as predicted, this seems to be mostly linked to body size constraints on movement, which were not accounted for in Nurmi & Parvinen's (2011) model. We did not see any differences in Pagel's λ between size-independent movement traits and specialization, but our sample size strongly limited our ability to detect small differences in phylogenetic signal (Münkemüller et al. 2012).

The present study, along a few others (Bonte et al. 2003; Entling, Stämpfli & Ovaskainen 2011; Stevens et al. 2012, 2014; Mabry et al. 2013) highlights the usefulness and power of multi-species and multi-traits comparisons to understand the ecological and evolutionary drivers of dispersal. Indeed, such comparative analyses, either by using trait data bases or through standardized experimental approaches, offer the potential to unravel complex associations between dispersal and multiple other traits. Here, we have been able to gather evidence for a dispersal/specialization syndrome in European land snails involving both behavioural and morphological traits. While we used relatively coarse measures of habitat specialization, they were correlated to finer-grained measures (see Material & Methods), and previous studies indicate that measures of specialization taken at different scales give very similar results (Bonte et al. 2003; Entling, Stämpfli & Ovaskainen 2011). Further research is, however, needed to determine how the strength of this syndrome varies with local habitat, population or landscape characteristics (Stevens et al. 2013) and to which extent dispersal and specialization correlate with other life-history traits, such as dormancy (Buoro & Carlson 2014), metabolic activity (Réale et al. 2010) or reproductive performances (Ronce & Clobert 2012; Stevens et al. 2012). Partial data on snail mean clutch mass seem to support the latter, with a positive link existing between habitat specialization and clutch mass (A. Ansart, unpublished data; Fig. S2, Supporting information). Despite these advantages, phylogenetic comparative methods have rarely been used in dispersal ecology until very recently, and approaches combining experimental data and phylogenetic information to investigate the evolution of movement traits are even rarer (but see Corcobado et al. 2012). Altogether, comparative studies (Stevens et al. 2012, 2014; the present study), experimental evolution approaches (Venail et al. 2008) and evolutionary models (Kisdi 2002; Nurmi & Parvinen 2011) show that dispersal and specialization should not be considered separately, but as two elements of a more general syndrome. This has potentially important consequences for ecosystem functioning in heterogeneous environments (Venail et al. 2008) and might make specialist species doubly vulnerable to extinction (Warren et al. 2001; Reinhardt et al. 2005). Therefore, recognizing and accounting for the pervasiveness of dispersal/specialization syndromes, and dispersal syndromes in general, is critically important to understand how the dynamics and persistence of populations, species and communities will be affected by the current environmental changes.