Evolutionary ecology of dispersal in biodiverse spatially structured systems: what is old and what is new?

(a) Why disperse?

Comprehensive reviews on dispersal evolution have listed the different selective pressures on dispersal and how their combination results in positive, negative, disruptive or stabilizing selection [3,4,6]. In brief, dispersal is selected in situations in which there is temporal variability in environmental conditions [31], perturbations [32], kin competition [18] and inbreeding depression [33]. In contrast, stable spatial heterogeneity in fitness expectations [34,35] and direct dispersal costs [36], which can result from a variety of factors including trade-offs, fragmentation and landscape resistance, favour non-dispersing individuals.

Beyond these basic selective forces, demography plays a complex role in dispersal evolution. For example, chaotic population dynamics can select for higher dispersal rates [37], while Allee effects can select against dispersal [38]. When populations do not systematically reach carrying capacity, as assumed in occupancy models, increasing population extinction rate does not always select for higher dispersal rates because frequent local extinctions tend to weaken the overall intensity of competition ([39]; but see [40]).

Beyond demography and focusing on more complex life histories, the timing of dispersal has a major impact on the evolution of both dispersal rates and distances [41]. Combining selective pressures on dispersal can lead to complex outcomes, including disruptive selection. For instance, a combination of spatial heterogeneity, kin competition and dispersal costs can lead to the evolutionary branching of dispersal rates [21,22,42].

The evolution of dispersal can interact with the evolution of other traits, like local adaptation [43–46] and self-fertilization [47–49], two traits that are especially sensitive to the spatial scale defined by dispersal (see [50] for a population genetics model that considers dispersal evolution in response to local adaptation and inbreeding depression). The joint evolution of dispersal with local adaptation can result in disruptive selection or alternative evolutionarily stable strategies in temporally variable environments. Indeed, habitat generalists perceive and/or react to this variability to a lesser degree than specialists, hence, associating selective pressures for generalism with selective pressures against dispersal ([43]; see also. [51], considering selfing individuals as ‘pollination generalists’). In spatially heterogeneous habitats, generalists tend to disperse more than specialists ([45]; see also [49] for the selfing evolution equivalent).

The evolution of dispersal can thus be affected by the evolution of other traits through reciprocal selective pressures (i.e. joint evolution, which arises as a response to fitness effects imposed by the evolution of another trait, or conversely, other traits evolve in response to the evolution of dispersal). This naturally generates correlations between dispersal and other traits, i.e. ‘dispersal syndromes’. However, this constitutes just one of the reasons behind the existence of dispersal syndromes [52], because syndromes can also emerge from allocation trade-offs (e.g. foraging versus dispersal), genetic correlations and pleiotropy, structural constraints (e.g. allometric scaling) or shared selective pressures (e.g. inbreeding depression affecting both the evolution of selfing and dispersal when accounting for inbreeding between relatives).

While the evolution of dispersal has traditionally been considered in metapopulations that do not expand in space, dispersal can also evolve during range expansions. In particular, considering the evolutionary dynamics of populations along an invasion front has highlighted the existence of ‘spatial selection’, by which higher dispersal rates are selected at the front of an advancing invasion wave [53–55] for a number of reasons: first, individuals with a higher dispersal rate have a higher probability of ending up at the front of the wave (ecological filtering; [56]). As a consequence, passive assortative mating in sexually reproducing species can further enhance this pattern. Second, dispersal can evolve via trade-offs between competitiveness and dispersal [57] owing to gradients of local competition strength, since denser populations far from the front exhibit more intense competition. Alternatively, a general trade-off between foraging and dispersal can also generate the same selection for higher dispersal rates at the front [58]. Finally, enhanced kin competition owing to repeated ‘founder effects’ at the invasion front also promotes increased dispersal [59]. Spatial selection is interesting as it can lead to accelerating range expansions, which may explain the rapid recolonization of trees in temperate latitudes since the last glacial period (Reid’s paradox; see [60]). The evolution of higher dispersal rates at the invasion front is also reminiscent of the evolution of lower dispersal in ageing demes [61]; in both cases, newly colonized patches harbour individuals with higher dispersal rates than older ones.

(b) Dispersal plasticity: context- and state-dependencies

Dispersal need not be a fixed trait but can respond plastically to internal and external cues [2]. The biotic and abiotic contexts can affect both emigration and immigration [62]. The extrinsic context typically includes patch-quality proxies (or simply patch types, e.g. [63]), such as patch size [64], patch age [65], population growth rate, population density [66–68] and carrying capacity [69], as well as available information about such proxies (number of mated or reproducing conspecifics, e.g. [70]). Information about the habitat matrix can also modulate context-dependent dispersal decisions [71,72]. The reliability of cues is an important parameter affecting the evolution of dispersal plasticity [73].

Intrinsic individual-level conditions affecting dispersal include sex [74–76], age [77], developmental stage [78], body size or condition [79,80] or any proxy of competitiveness, for example. Intrinsic and extrinsic cues can of course be used at the same time and interact to determine dispersal decisions [81].

(c) Consequences of dispersal

In a single-species context, dispersal is known to have multiple consequences, both ecologically and evolutionarily [3,82]. From an evolutionary point of view, dispersal promotes genetic mixing: immigrants bring new alleles into populations and thus increase local genetic diversity. Conversely, a (very) low dispersal rate between populations increases differentiation. The mixing effect of dispersal thereby constrains local adaptation and may lead to a ‘migration load’ known from classic population genetics models. This is especially striking in continuous space models of adaptation on environmental gradients, where high dispersal rates tend to impede local adaptation [83–86]. The stage at which dispersal takes place is also critical to the evolution of local adaptation, so that higher dispersal rates might actually not affect local adaptation in certain cases [87–90].

From an ecological point of view, dispersal is also an important parameter controlling the dynamics of metapopulations in at least two different ways. First, dispersal drives the synchrony of local population dynamics, but in a nonlinear manner that also depends on other factors such as perturbation frequency, the relative speed of population dynamics of the focal species versus its natural enemies or resources or the spatial extent of dispersal [91–96]. Second, dispersal determines the resilience of metapopulations to perturbations as higher dispersal rates can lead to more rapid recolonization after local extinction.

At the metapopulation level, an important effect of higher dispersal rates is higher occupancy, as long as dispersal translates into higher colonization rates [97–99]. In other words, species occupy more patches when they disperse more. However, this general rule can be counteracted in two important ways. First, when the costs of dispersal are substantial, dispersal rates that are too high can induce a decrease in colonization rate since emigrating effectively leads to a decrease in local birth rate [99]. Put differently, if dispersal is costly, high emigration rates lead to a reduction of total metapopulation size. Second, when local populations experience an Allee effect, dispersal hinders the maintenance of population sizes above the Allee threshold [38].

Overall, it is important to note that these general effects are often based on random, unconditional dispersal. They may, therefore, be counteracted or even inverted by conditional emigration and habitat selection [46,100,101].

(d) Lessons from existing theory

In order to better define future research avenues, we have discussed above what is known from the metapopulation literature. If we summarize our understanding of single-species systems to a very abstract level, we can reduce dispersal theory to a single fundamental proposition and two fundamental principles: the proposition is that dispersal and its evolution are driven by the contrast between local (inclusive) fitness expectations and the fitness expectation elsewhere [69,102]. Most studies looking at more specific effects can be traced back to two general principles: dispersal will be more frequent in systems where (i) the temporal variance in fitness expectations is large and (ii) the spatial correlation in fitness expectations is low [32,35,63], i.e. where the spatio-temporal variance in fitness expectations is large.

(a) Horizontal interactions: metacommunities

In metacommunities [104,105] with horizontal interactions (i.e. non-trophic interactions), the evolution of dispersal has been considered in the context of competition–colonization trade-offs [106,107] and evolution-mediated priority effects [108]. The parallels to single-species work and thinking are unmistakable (see figure 1 for an overview) and these parallels also become obvious when considering dispersal plasticity, especially density-dependent dispersal in a metacommunity context, where single-species theory can directly be transposed to the multispecies context taking into account differential strengths of competition (see [109] and figure 2).

• Download figure
• Open in new tab
• Download PowerPoint

• Download figure
• Open in new tab
• Download PowerPoint

While the parallels between single- and multi-species drivers of evolution are clear when assuming asexual reproduction since genotypes are then indistinguishable from species, sexual reproduction and kinship may a priori prevent us from extrapolating from single to multispecies meta-systems. However, this added complexity may rather be a matter of quantitative effects than qualitative differences.

For instance, recombination of genetic material may happen at the interspecific level via hybridization, which is an important eco-evolutionary mechanism [110] and an analogue to sexual reproduction. Inbreeding depression affects single-species dispersal evolution, but it has no clear multi-species equivalent, except maybe if we assume the unlikely case of hybrid vigour playing a role that is similar to overdominance in single-species models. Similarly, kin competition seems irrelevant at the interspecific level as an ultimate determinant of dispersal, since the probability of sharing the same allele between individuals from different species should, a priori, be very low. However, this may not be the case owing to interspecific gene flow from hybridization, as discussed above, or owing to incomplete lineage sorting (ancestral polymorphism between species; e.g. [111,112]), for example. When considering proximate determinants of dispersal, phylogenetic proximity of interacting species might represent an analogous factor to relatedness in single-species cases, because phylogenetically close species might be more likely to have similar niches, and hence compete more strongly for resources (see [113] for a more detailed discussion on the link between phylogenetic proximity and traits).

While these examples are speculative and call for more formal investigations, they highlight that, at least for horizontal communities, the classic mechanisms discussed in single-species models, especially the proposition and two fundamental principles derived above, mostly hold (figure 1), but probably with altered relative importances.

(b) Vertical interactions: meta-food webs

Does the above conclusion hold when moving to vertical interactions such as predation and parasitism? Existing theory can inform us here. For instance, predation affects the evolution of dispersal in predators and prey alike since it intrinsically controls both species dynamics and thus the spatio-temporal dynamics in (expected) inclusive fitness [114–119]. This patterning of spatial fitness expectations, which here relies on predator–prey cycles, is different from what can be expected in a simple horizontal competitive setting or even under mutualism [114] and represents an important driver of dispersal evolution. Note that, as for horizontal communities, the fundamental forces driving dispersal evolution do not differ from those in single species discussed above [120,121]. This example shows how, in Hutchinson’s [122] words, interspecific interactions can form the ecological theatre in which the evolutionary play of the focal species unfolds [14]. Here specifically, one species’ demography is another species’ temporal variation of the environment when these species interact (figure 3). This biotic determination of selection pressures is interesting since it may actually reduce ‘parameter space’: while theoretically any spatio-temporal autocorrelation of fitness expectations can be assumed, biotic interactions, by generating characteristic population dynamics in interacting species, here determine, and constrain, the nature of this autocorrelation. Importantly, we have models that allow us to capture the essence of these patterns (see [41] for a discrete-time discrete-fitness expectation model).

• Download figure
• Open in new tab
• Download PowerPoint

Going one step further, the study of the joint evolution of interaction strength and dispersal has almost never been undertaken. This is of interest because the strength of interspecific interactions determines the extent to which interacting partner densities impact each other; hence, the drivers of dispersal evolution [114,115]. Furthermore, via its feedback on the demography of interacting species, dispersal can impact the evolution of interaction strength. Thus, the joint evolution of dispersal and interaction strength is likely to further constrain the population dynamics of interacting partners and dispersal evolution. Calcagno et al. [123] modelled the evolution of colonization rates in a multi-trophic occupancy setting in which colonization rate trades off with competitive ability, but they did not incorporate changes in top-down or bottom-up controls as consequences of colonization evolution (see also [124] for the two-level, predator–prey case). In the case of symbiotic interactions, Ledru et al. [125] tackled the question of dispersal evolution together with that of the evolution of mutualism. They conclude that localized competition favours the evolution of mutualism and that mutualistic types tend to be associated with lower dispersal rates, a finding that matches with studies on the joint evolution of intraspecific altruism and dispersal [126,127].

All these examples highlight that knowledge from single-species settings seems applicable, even to meta-food webs. Our proposition and the two fundamental principles introduced above certainly hold. In addition, the multi-species context also introduces the possibility that a given species’ abundance or occupancy dynamics and spatial distribution become the ‘environment’ in another species’ ultimate and proximate drivers of dispersal (figure 3).

(a) Biotic proximate determinants of dispersal: co-dispersal and mobile links

As discussed above, proximate biotic drivers of dispersal that lead to context-dependency are clearly present at the interspecific level [129–133], but they can overall be explained relying on theory developed for single-species systems (figure 2).

However, an additional process that is not addressed by classic studies is the occurrence of ‘co-dispersal’ (figures 1 and 2), i.e. cases where the two interacting partners disperse together (figure 4). This echoes the idea of mobile links [134]: organisms connecting habitats via the transport of resources or processes (think of hippopotami linking terrestrial and aquatic ecosystems via transfer of matter; for many more examples, see [30,135] and [136]). Here, these mobile links transport other organisms and their genes.

• Download figure
• Open in new tab
• Download PowerPoint

For many symbionts, but also for animal-dispersed plant seeds or phoretically dispersed organisms (organisms travelling on other organisms, e.g. [14]), the movements of the host or vector may be unrelated to dispersal (being, instead, related to foraging or seasonal migration) and yet they result in dispersal for the symbiont species. For example, foraging pollinators may disperse parasites among plant species [137] or female mosquitoes, while blood feeding, may vector different parasites (e.g. viruses or malaria). In such a situation, the usual dispersal drivers do not apply, and are replaced by the drivers of movement for the host or vector [138,139]. This mismatch in spatial scale may lead to conflicting selection pressures.

Except in the case of obligate mutualisms, this dependence on dispersal is almost always asymmetric, e.g. plants need animals for seed dispersal but animals can disperse without seeds; symbionts often need their host to disperse, but symbiont-free hosts can still disperse. In these cases, the dispersal of the host organism may be controlled by both its own genes and those of the interacting species, through its extended phenotype. For instance, symbionts might manipulate the dispersal of their hosts so that infected individuals disperse more or less than uninfected ones [140–142]. Conversely, infection could trigger an adaptive state-dependent dispersal strategy in the host [121,143–146]. Specifically, endosymbionts and, in particular intracellular endosymbionts, on the one hand, will co-disperse with the host unless they are disposed of by its immune system. There is evidence that endosymbionts might induce dispersal, select for dispersal in their host, bias immigration success or affect the dispersal distance of their host (e.g. Wolbachia, Rickettsia: [147] and [140]; microbiome and metacommunity: [148]). Ectosymbionts, on the other hand, will not exhibit co-dispersal fidelity with certainty but can induce dispersal (and co-dispersal). However, the processes and dynamics are different in ectosymbionts since they may be affected by the environment surrounding their host and hosts may attempt to modify symbiotic load by relocating to a different habitat [149].

While manipulation by the symbiont represents one way of addressing the above-mentioned conflict over selection pressures, adaptive dispersal strategies for the host may also lead to curing (i.e. extinction of the parasite; [147]). To make the issue more complex, hosts can be co-infected by multiple parasites where competition between parasites can impact transmission (dispersal) as discussed by Godinho et al. [150]. Finally, parasites may be co-transmitted [151,152], whereby one relies on another for transmission to new hosts. These complexities are beyond the scope of the current article.

The consequences of co-dispersal can be manifold. From an evolutionary point of view, dispersing with or without a symbiont is known to be different, and can lead to different dispersal and trait evolutionary trajectories [153,154]. These effects can occur under both mutualistic and parasitic symbioses [155], but parasitic symbioses are also prone to evolve towards manipulation of host dispersal [156,157]. Unfortunately, similarly detailed analyses involving more realistic meta-food webs are missing. In conclusion, co-dispersal clearly represents an additional driver at the inter-specific level that cannot occur in single-species systems.

(b) Increased trait dimensionality in metacommunities and meta-food webs

In general, moving from a single- to a multi-species context should increase system complexity and thereby also the dimensionality of the relevant trait space (figure 1). Débarre et al. [158] show that increased trait dimensionality tends to destabilize evolutionary equilibria. This in turn increases opportunities for evolutionary branching and can therefore favour diversification, which could lead to polymorphisms of dispersal or other traits [159] with (quasi-)equilibrium reached when the systems settle—across different species—in an ideal-free fitness distribution [160]. By contrast, a more complex biotic environment can be understood as inducing a ‘cost of complexity’ [161], i.e. adaptation is slower when it is required on more dimensions, or if mutations are more pleiotropic (and thus quicker), phenotypic variation can decrease with the number of dimensions (the ‘preservation of perfection’ effect [162]).

Taken together, from Débarre et al. [158] we may predict that diversity in species interactions begets diversity (at least in trait distributions; see also [163] outside the context of dispersal evolution). At the same time, the cost of complexity [161] and the preservation of perfection [162] introduced above may imply that evolution will be slower in more complex systems, intrinsically, owing both to increased dimensionality and to decreased standing variation in the case of pleiotropy. These impacts on the speed of evolution have interesting consequences for eco-evolutionary feedbacks: Eco-evolution is particularly likely to lead to novel, emergent phenomena when ecological and evolutionary timescales are similar [14]. If complexity tends to reduce the speed of evolution as outlined above, it may, therefore, decouple ecological and evolutionary timescales and prevent eco-evolutionary feedbacks [14]. The ultimate consequence of this interplay may be that dispersal evolution is overall slower and more ‘diffuse’ in multispecies meta-systems. Potentially, this could reduce divergence in dispersal strategies across species or, conversely, lead to neutral coexistence of diversity. One can speculate that this may consequently reduce the probability of dispersal limitation being overcome by evolution, which implies that current models that are largely single-species models would drastically overestimate the potential for adaptation and range shifts.

(a) A proposition and general principles

In order to attain this synthesis of dispersal evolutionary ecology and biodiversity research, we here take a first step to understand the drivers of dispersal in biodiverse systems. At the same level of abstraction as the proposition and the two general principles for dispersal evolution in single-species systems laid out above, we argue that nothing changes when we consider multi-species systems. Individuals should still emigrate whenever (inclusive) fitness expectations elsewhere are larger than those at the current site of residence and dispersal should occur more often whenever the spatio-temporal variation in fitness expectations is large. Whatever forms the ecological theatre for the evolutionary dispersal play, intra- or inter-specific competition, predation, parasitism or mutualism, does not matter in detail. At a lower level of abstraction, drivers may change in relative importance and additional drivers, such as co-dispersal, may directly impact costs and benefits of dispersal (figures 1 and 2).

(b) Directions for future research

Since many points discussed above are largely speculative, they call for more investigation, both theoretically and empirically. In the following, we provide a non-comprehensive list of potentially relevant questions and topics.

Going back to the most abstract level mentioned above, it is relevant to determine what are, in general the most important drivers of fitness expectations for each species (abiotic conditions, intra-specific or inter-specific interactions; e.g. [183]). These define the ecological theatre (selection pressures) for the evolutionary play. Moreover, it is relevant to ask which of these drivers are responsible for the largest temporal variance in fitness expectations and what the characteristic spatial scale of autocorrelation in this variance is. Furthermore, with an increasing number of interactions, is spatio-temporal variance in fitness-expectations likely to increase or decrease? Are fitness expectations at all predictable in a world of highly complex and dynamic interactions?

From a theoretical point of view, we have noted that co-dispersal is largely understudied. While some work exists at the interface between evolutionary epidemiology and dispersal evolution, as recently reviewed by Zilio et al. [128], a comprehensive theoretical treatment is lacking.

The same holds true for the joint evolution of dispersal with interaction traits, as well as considering higher-order interactions involved in dispersal (e.g. dispersal of parasites with complex life cycles and multiple successive host species). Future work could identify which ‘parameters’ of interactions are important for dispersal (interaction signs, immediacy, effect on abundances, correlations with other traits, to name but a few). Dispersal syndromes with or without co-dispersers (importance of symbiosis for dispersal plasticity and correlation with other traits) may have to be more clearly included into theory. We would also like to highlight that the step from communities and food webs towards ecosystems has to be taken, by taking into account fluxes of matter and studying constraints on dispersal at the ecosystem level, for example [10]. Feedbacks will be interesting to understand, namely feedbacks between the evolution of dispersal in multiple species, but also feedbacks on local and regional community dynamics and realized interaction strengths, which can in turn modify dispersal evolution.

At the opposite end of the continuum of biological organization levels, the genome, Débarre et al. [158] show that epistasis and trait correlations may to some degree modulate responses, and the ‘cost of complexity’ model suggests that modularity in genetic architecture may be a solution to simplify the problem of adaptation. Overall, this calls for putting the genetic architecture of traits, including dispersal, at the centre of theoretical and empirical investigations [14,184–186]. Using a multilevel network modelling approach [187] represents a promising way forward.

From an empirical point of view, dispersal plasticity certainly requires more investigation. One could imagine, for instance, that a symbiont’s dispersal strategy may be less plastic, being more constrained by the host dispersal plasticity, or could it be that symbiont dispersal plasticity is induced by host dispersal plasticity (when the host’s environment changes, the same is true for the symbiont)? More generally, it would be relevant to understand what reliable cues, triggers or correlates of fitness-relevant drivers exist at the meta-food web level and how accurately such interspecific information can be perceived. Particularly, increasing the numbers and diversity of interactions may make every single one less fitness-relevant. If dispersal decisions are plastic, external cues may thus become less relevant and reliable as triggers of dispersal in comparison with internal states like body mass/size, hunger level or health status.

(c) Summary and conclusion

We here discussed to which degree theory built for single-species systems also holds at the multi-species level and advances our understanding of dispersal evolution in meta-systems. Are effects exacerbated or averaged out? Do they change in relative importance?

We highlight that, at the highest level of abstraction, dispersal theory consists of a single fundamental proposition and two fundamental principles that presumably hold within and across species. At a lower level of abstraction, biotic interactions form the ecological theatre for the evolutionary dispersal play: biotic interactions like trophic interactions mediate patterns of fitness expectations in space and time, thereby reducing the realm of possible patterns and providing us with a modelling tool. While this implies a simple transposition of already-known drivers to a multispecies context, some drivers may require upscaling or more complex transpositions (figures 1 and 2). The latter provides interesting possibilities for linking micro- and macro-evolutionary dynamics. In this article, we have identified an important quantitative modulator of dispersal evolution: increased trait dimensionality, which may impact the speed of evolution. Finally, an additional factor to consider is co-dispersal, which may lead to scale and selection pressure mismatches between hosts and symbionts. A large number of open questions and important potential consequences in both ecological and evolutionary terms call for more investigation. Taken together, we hope that these efforts will allow researchers to fully merge dispersal evolutionary ecology with the ongoing push to develop evolutionary metacommunity ecology [164,165]. This merger would enable better investigation of spatial eco-evolutionary feedback loops.