Dispersal as a part of demographic strategies
Our analyses revealed a strong association between demography and the ability to disperse, or, more generally, to move. Although fecundity, maturation time and adult survival has not been reported in the same unit of time (hence some non-conventional associations among demographic traits arose), the first axis grossly describes the well known slow-fast continuum found in vertebrates (Gaillard et al. 1989; Clobert et al. 1998) where r-species are characterised by a high fecundity. Therefore, the high impact of demographic axis 1 on mobility measurements reinforces the idea that a high mobility is part of the r-strategy. Migration has been mentioned in this strategy, as a means to temporarily escape from unfavourable conditions (Southwood 1988). Our results suggest that dispersal is another spatial mechanism involved in the r-strategy since dispersal kernels, indicative of the frequency of long-distance dispersal and of the mean distance of inter-patch movements, strongly correlated with the axis 1 of demography. This axis also marginally impacted gene flow (Table 5).
However, to what extent this relationship is causal remains to be elucidated. The comparative method used here is a powerful tool for detecting associations among traits across species (Pagel & Harvey 1988), but the interpretation requires caution as associations may be indirect or even non-adaptive, resulting, for instance, from epistasy or pleiotropic effects (Rose 1982). Dispersal, like migration, is supposed to evolve under conditions of habitat instability (Gadgil 1971; Roff 1975; Comins 1980; McPeek & Holt 1992; Travis & Dytham 1999). In butterflies, dispersal may thus have been selected by the same environmental pressures that shaped life-history strategies. In agreement with this hypothesis, Dennis et al. (2004) showed that butterfly mobility correlated with habitat disturbance (as measured by the host plants’ generation time). They hypothesised that this pattern emerged from butterflies and their host plants having evolved under common environmental conditions. However, alternatively, migration could be the causal link between demography and dispersal. In this scenario, enhanced flight performances in migratory species might cause an apparent link between dispersal and demography, even if only migration is selected for by habitat instability. There is support for this hypothesis in birds, where morphological traits (wing shape) relate to both migration and dispersal (Dawideit et al. 2009). Contrarily to that observed in birds, we found a direct effect of morphology (wing length) on dispersal, but not on migration. Nevertheless, in butterflies, there is some correlation between dispersal ability and migratory tendency (Stevens et al. 2010b). An indirect relationship might have arisen, but only if morphological or physiological traits other than wing length link dispersal to migration, and caused dispersal to evolve as a consequence of the evolution of migration. Hence, for the moment, we cannot discriminate between these two non-mutually exclusive hypotheses: dispersal evolved directly as a part of the r-strategy in response to habitat instability, or, alternatively, dispersal is facilitated in species with attributes allowing migration.
The positive relationship of mobility to r-selected traits invalidated the existence of a general mobility-fecundity trade-off in butterflies. Fecundity was positively correlated with demographic axis 1, which in turn positively correlated with mobility. Although the oogenesis-flight trade-off was evidenced in several insects (reviewed in Denno et al. 1989), there are also examples of high movement capacity associated with rapid development, early reproduction and high fecundity like we showed here (Lavie & Ritte 1978; Hanski et al. 2006). The consequences of this result are discussed further (see below: Evolutionary consequences).
The effect of demographic axis 2 on the frequency of long-distance dispersal was only significant when the effect of demographic axis 1 was accounted for, which possibly indicates a disproportional contribution of those traits that contributed strongly to the second axis when compared with the first (i.e. larval growth rate, adult lifetime, and maturation time). These life-history traits are positively correlated with the second axis of demography, which in turn correlated negatively with the dispersal measurement. As such, our results indicate that long-distance dispersing species tend to be fast developing species. This was counterintuitive, as a long maturation could also have increased the frequency of long-dispersal movements. The opposite trend observed here corroborated the hypothesis that dispersal in butterflies probably evolved in part as a response to habitat instability within the ‘fugitive species’ syndrome (Tilman 1994).
Adult lifetime positively correlated with both axis1 and axis 2 of demography; as such we cannot directly assess the relationship of adult lifetime with dispersal (as those axes had opposite effects). This would require the collection of survival data within the right time scale to better establish its link with other demographic parameters as well as with dispersal. The hypothesis that lifetime should be traded-off against dispersal ability at the inter-specific level is thus still an open question. There is, however, empirical support for such a trade-off in the Glanville fritillary butterfly (Hanski et al. 2006).
The fact that demographic axis 3 did not correlate with mobility was surprising, but can be linked to the fact that this axis only accounted for 15% of the total variance in life-history traits. This can also be the net result of the antagonistic forces relating dispersal to capital breeding (Table 1): direct costs of movement may penalise capital breeders, as increased relative abdomen mass will reduce flight ability (Jervis et al. 2005). On the other hand, capital breeders could be advantaged as they have more opportunity to mate before dispersal, and to allocate adult-acquired resources to flight versus egg maturation (Dennis et al. 2003).
Contrary to what we expected, the length of flight period had no significant effect on dispersal. Hence, the hypothesis that dispersal costs might be reduced with an increased window for dispersal did not find support in butterflies.
Dispersal and morphology
Dispersal allometry is variable in butterflies, depending on which mobility, measurement is considered. Wing length strongly related to the frequency of long-distance dispersal, more loosely with the mean dispersal distance and the dispersal propensity, and not at all with gene flow (Table 5). The assumed causal relationship between dispersal and wing length comes from studies where the potential effect of other traits was not controlled for. However, as already mentioned, body size correlates with many life-history traits through allometry (Blueweiss et al. 1978; Wiklund et al. 1987; Gaillard et al. 1989; Clobert et al. 1998; Garcia-Barros 2000). As we controlled for the effect of several of these traits, the effect of wing length may have been reduced.
Wing length is, however, strongly related to the frequency of long-distance dispersal of butterflies, even when controlling for the effects of other life-history traits. The body size effect on this dispersal measurement thus goes beyond life-history allometry. Rather, we suspect that the metabolic cost of flight is the key behind this pattern: as the cost of flight per unit weight is very constant (Tucker 1970; Schmidt-Nielsen 1972), large species are probably able to move longer distances at relatively low metabolic costs. This clearly requires further research.
Dispersal and specialisation
Species specialisation had noticeably weak relationships with dispersal, with (low) correlations only between the first axis of the specialisation-PCA and dispersal efficiency (gene flow). However, this result is important, as it indicates that the advantage to generalists resides in the successful transition from individual movement to effective gene flow. This relationship is thus probably attributable to the deferred costs, i.e. paid at or after immigration into the new habitat patch (Stamps et al. 2005; Bonte et al. 2011). Several mechanisms might increase the deferred dispersal costs for specialists. Attrition during transfer may be higher for species with narrow thermal tolerances, and attrition can reduce individual’s attractiveness to potential mates or diminish life expectancy or fecundity (Bonte et al. 2011); species with narrow habitat selection may also suffer reduced fitness after immigration into sub-optimal habitats, while more habitat types are optimal for habitat generalists; and attrition costs may also be higher for species accepting few nectar sources, which may result in fewer feeding opportunities during transfer. Adult feeding generalism is, however, associated with larval dietary breadth in butterflies (positive and negative relationships were shown, depending on context and analytical procedures: Stefanescu & Traveset 2009), a trait with hardly any effect on specialisation axis 1 (but rather on the second axis, not retained in our best models) and independent from adult habitat range (Appendix S1). This discredits this last hypothesis that habitat generalists may benefit from more en route nectaring. Specialisation axis 1 is dominated by adult habitat range and adult thermal tolerance, with generalists having high scores. The advantage to generalists thus probably results from differences in performance after immigration related to thermal tolerance or habitat selection, a question that could be solved by the confrontation of field data on movement rates and individual performances, to genetic data informing the genetic components of dispersal.
It is noteworthy that specialisation, although uncoupled with dispersal movements and only loosely related to migration and gene flow, has a large impact on expert score. This may indicate that expert scoring is influenced by species specialisation. It is reasonable to think that an expert may overestimate the mobility of species seen in a large variety of habitats, and flying under wide ranges of temperatures, two traits positively correlated with the first axis of the specialisation-PCA.
Dispersal and routine behaviours
Surprisingly, and contrary to our expectations, when routine movements affected dispersal, this was independent from ecological specialisation (interactions not shown in Table 5 were not significant). In line with our expectations, single-egg layers realised longer dispersal distances than egg-batch layers, probably because single-egg layers had to move more often to select their oviposition sites. Interestingly, the oviposition strategy only correlated with short-distance dispersal. Indeed, the mean dispersal distance was extracted from negative exponential kernels that best fit at relatively short distances (Baguette 2003). This might indicate that only small-scale inter-patch movements can be realised as a by-product of routine movements. This difference in patterns related to short- and long-distance dispersal movement is in accordance with the findings of Hovestadt et al. (2011), showing the presence of mixed dispersal kernels in the butterfly Maculinea nausithous, which they hypothesised was the outcome of a mixture of two distinct processes: daily routine movement and genuine dispersal (Van Dyck & Baguette 2005). Such routine movements did not impact on the frequency of long-distance movements or gene flow, both of greatest importance for species spread, species persistence and metapopulation functioning (Baguette 2003; Schtickzelle et al. 2005a; Trakhtenbrot et al. 2005).
Dispersal directly interacts with the adaptive response of species to environmental changes, as it is responsible for the spatial redistribution of genotypes (Ronce 2007). This meta-behaviour now faces increased selective pressures because of the conjunction of an increasing impact of those global changes that require a spatial response through increased dispersal, and the ubiquity of dispersal costs (Bonte et al. 2011). The fact that dispersal life-history trait relationships are highly variable among the dispersal measurements considered challenges and the hypothesis that all behaviours related to dispersal have evolved jointly into a real dispersal syndrome (Clobert et al. 2009); rather, each element in the dispersal process has probably evolved partly independently from the others in response to uncoupled selection pressures (Baguette & Van Dyck 2007; Clobert et al. 2009), and will probably continue to do so in the future. We think that this pattern has resulted from the partly independent costs associated with the various dispersal steps (Baguette & Van Dyck 2007; Bonte et al. 2011).
Our analyses ignored within-species variation in dispersal behaviours and in other traits. However, these may be quite high (see Stevens et al. 2010a on the importance of intraspecific variation in dispersal). The strong correlations among traits may also constrain dispersal at the within-species level. For instance, Schtickzelle et al. (2006) showed contrasting dispersal behaviours in the butterfly Boloria eunomia, a pattern that paralleled the level of fragmentation in suitable habitats. It would be interesting to investigate how other traits behave along such a gradient of habitat fragmentation. If other trait values are selected for as a by-product of selective pressures imposed on dispersal behaviours, this may have strong effects on processes like local adaptation and speciation. Likewise, other environmental conditions may change the cost-benefit value of dispersal, like population density (Konvicka et al. in press), host plant distribution or climate. Investigating how the co-variations among traits vary according to these conditions at the population level certainly deserves further empirical investigation.
Dispersal consists of several behaviours, from the decision to leave, through the ability to move safely through inhospitable habitats, to navigate towards a suitable patch, to the settlement and the recruitment into this patch (Stenseth & Lidicker 1992; Ims & Yoccoz 1997). We showed that these components of dispersal may be partially decoupled in evolutionary and ecological times. Although related to movement rates, wing length has no direct effect on dispersal efficiency (i.e. gene flow) in butterflies. We observed the reverse for specialisation, which was related to gene flow, but not to individual movements. Accordingly, we suggest that individual movements and gene flow, two components of the dispersal process, might be uncoupled under some circumstances. Furthermore, ordinary movements may result in small-scale dispersal, but have no significant effect on long-distance dispersal and gene flow. Hence, ecological or evolutionary changes in ordinary movements might impact local dispersal, but will probably have no effect on spatial gene flow, especially for long distances. Likewise, if the relationships between the frequency of long-distance dispersal and the second demographic axis (Table 5) is causal, or at least direct, an evolutionary change in development rate, like for instance, in response to climate change (Parmesan 2006), may result in a side-effect on the ability to move long distances.
Our comparative study helps identify which life-history traits co-vary with which dispersal traits; however, this study does not identify the causal mechanisms of these covariations. As such, further mechanistic studies testing the processes that explain the correlations are now warranted. Nevertheless, we may expect the evolution of longer dispersal distances to be slow for species with currently low dispersal ability, simply because these also tend to have low demographic turnover.
The fact that the relationships between dispersal and other life-history traits were highly variable among the dispersal measurements considered also suggests that the selective pressures acting on each of those components potentially may have decoupled effects on other traits. Noticeably, the dynamics of specialisation and effective dispersal will probably interact in populations facing changed spatial pressures, with consequences for community composition and functioning. Devictor et al. (2008) already showed that human-driven environmental changes result in biotic homogenisation. The link between the ability of a species to maintain gene flow and its specialisation will probably reinforce this homogenisation, as the consequence of an increased pressure for higher effective dispersal rates (as imposed by fragmentation and climate change). This would favour generalist species over specialists. Likewise, the presumed absence of an oogenesis-flight trade-off has important consequences for the evolutionary potential at invasion fronts, as both dispersal and demography may jointly evolve towards increased invasiveness.
Finally, our analyses showed evidence of low phylogenetic constraints acting on dispersal in butterflies, consistent with the observation of Pavoine et al. (unpublished) who quantified the relative importance of common ancestry and habitat filtering in shaping the evolution of butterfly traits (including the expert score we used) within a metacommunity, and who showed that habitat filtering has the dominant effect, whereas phylogenetic constraints were much lower. The strong association of dispersal and demography reinforces this idea that both demography and mobility are evolutionary labile traits, which have converged in distant clades subjected to common environmental constraints, for instance, habitat instability. There is, however, a possibility that the small sample sizes available for some mobility measurements did not reveal the phylogenetic constraint on the corresponding mobility trait.
Consequences for species functioning under changed environmental conditions
The positive relationship between demography and dispersal ability has major implications for both species invasiveness and species persistence. Low rates of displacement through landscapes disfavour the persistence of species facing climate change or habitat fragmentation (Henle et al. 2004; Ockinger et al. 2010), although some authors argue that butterfly species with intermediate dispersal levels would decline the most (Thomas 2000). Given that dispersal correlated with demography, the challenges imposed by habitat fragmentation or by climate change should disproportionately impact species with low demographic turnovers, as these species also proved to have low dispersal rates. Specialist butterflies should suffer more from these environmental changes than generalists, because they have developed low dispersal abilities. However, both a low turnover and a high specialisation per se predispose species to extinction (Henle et al. 2004; Barbaro & van Halder 2009). Together, the correlations among traits would thus globally increase the discrepancy between species at risk and species less at risk in face of global environmental changes.
The existence of a colonisation syndrome has been questioned in theoretical studies (Ronce et al. 2000). Our study provides evidence of such a syndrome in butterflies where the turnover of individuals within populations was positively correlated with dispersal ability. In the same vein, high growth rates predispose plants to invasiveness (van Kleunen et al. 2010). In butterflies, strong dispersal tends to be related to fast turnovers, which will reinforce the invasiveness of those species that have high values for both. The Large White butterfly (Pieris brassicae) has, for example, very high dispersal power and a very fast turnover. These traits probably worked together to produce its invasive success (Feltwell 1982).
Consequences for the choice of substitute species
Accounting for species’ dispersal ability is of primary importance to develop efficient conservation strategies under global environmental changes (Brook et al. 2000). The lack of dispersal has been recognised as a main limitation of models for predicting biodiversity patterns (Guisan & Thuiller 2005; Engler & Guisan 2009). If the relevant dispersal data are unavailable, modellers either assume that there is no dispersal, or, on the contrary they assume that dispersal is unlimited. The addition to these models of dispersal data, even if imprecise, will help reduce the uncertainty of their predictions (Engler & Guisan 2009). Therefore, an attractive solution should be using a substitute species, i.e. a species used on the assumption that it shows how the species of conservation concern might respond to a given environmental disturbance (Caro et al. 2005). The critical element in the choice of this substitute is therefore its similar response to the focal processes. Our results give some insights on how substitutes for dispersal ability should be chosen.
An intuitive idea has been choosing the most closely related species for which the information is available (as did Schtickzelle et al. 2005b). The low impact of phylogenetic relationships on butterfly dispersal questions whether that is a valid approach. Rather, we suggest that a species with comparable demography is in most cases the best choice. Then, if one wishes to gain insight into gene flow, the proximity in species specialisation and particularly in thermal tolerance and adult habitat range can be considered alternatively.
Wing length has also often been used as a proxy for species mobility (e.g. in Ockinger et al. 2010), which was justified by the widespread correlation between flight ability and body size (Paradis et al. 1998; Sutherland et al. 2000; Komonen et al. 2004). We may, however, wonder whether body size is a valuable proxy for dispersal traits. Dispersal might correlate directly with wing length. However, its value as a dispersal proxy might be artificially inflated by the allometry of other traits related to dispersal. We show here that the potential advantages of summarising dispersal by body size (wing length) is reduced given that (1) wing length relationships are low for most dispersal measurements, (2) allometric traits may have either no relationship or an inverse relationship with dispersal and (3) several non-allometric traits correlated better with dispersal (noticeably the specialisation or the egg-laying strategy) (Table 5). Using wing length alone will probably be insufficient to accurately predict a species’ dispersal ability. Hence, although inferring dispersal ability from wing length may be the ‘least bad’ solution for species for which information on other traits is unavailable, the precision of this prediction will be rather low for most components of dispersal. Likewise, Sutherland et al. (2000), using a positive relationship between mammal body size and dispersal distance, have tried to apply this correlation to predict the expected median or maximum dispersal distance for species of given body sizes. The predictive capacity of their correlation was rather low, indicating that body size is a poor predictor of mammal dispersal abilities. In butterflies, species of similar wing length, however, may be preferred as a substitute in cases where the frequency of long-distance movements is an issue.
Generality of the patterns
It is difficult to generalise the patterns we observed in butterflies (i.e. a strong association of dispersal with demographic traits, variable effects of body size and low phylogenetic constraints) across different taxonomic groups. Several studies have examined the relationships between dispersal and other traits at the species level. However, both the dispersal measurements and the traits considered varied widely among these studies. To our knowledge, the relationship with demographic traits was only investigated in plants (Thomson et al. 2010), but only the dispersal mechanism was considered, whereas the frequency, the distance or the efficiency of dispersal were all ignored.
The allometry in dispersal distances was observed in several taxa: marine fishes, mammals and birds with larger adult size dispersing larger distances (Sutherland et al. 2000; Bradbury et al. 2008); and the dispersal distance is positively correlated with propagule size across a wide variety of actively dispersing organisms (Jenkins et al. 2007). These reviews ignored other traits (in particular, the demographic traits co-varying with dispersal in butterflies), and their results cannot be interpreted as evidence for a direct effect of body size on dispersal. Our study showed that this effect may exist for the frequency of long-distance dispersal, but not for gene flow. As not all other studies separately addressed these two components of dispersal, we cannot generalise at this stage about the pattern of dispersal allometry.
It seems that the phylogenetic dependency of dispersal has not been assessed per se before. Rather, in some comparative studies cited here, the correction for phylogenetic dependency was applied a priori, with a variety of methods (PGLS, family added as either a categorical or a random variable and phylogenetic independent contrasts), whereas in other studies phylogenetic dependence was not considered at all. Investigating how phylogeny constrains dispersal across taxa certainly deserves further attention as this comparison would help us better understand the patterns of dispersal evolution.