Analysing eco‐evolutionary dynamics—The challenging complexity of the real world

1. The

Despite many advances, most experimental studies involve proofof-principle experiments quantifying the impact of phenotypic variation in highly standardized, simplified environments, using a common gardening design (Matthews et al., 2011). In a common gardening approach, one studies how different phenotypes of focal taxa influence an initially identical environment (the "common garden"; Matthews et al., 2011). Most studies use phenotypically differentiated individuals in their analysis of eco-evolutionary dynamics, and mostly so in artificially constructed communities.
5. Finally, we provide a road map of methods to study eco-evolutionary dynamics in more natural settings. Eco-evolutionary studies involving multiple species are necessarily demanding and might require intensive collaboration among research teams, but are highly needed.

K E Y W O R D S
cryptic eco-evolutionary dynamics, eco-evolutionary dynamics, evolving metacommunities, food web interactions, multi-species guild, nature, regional dynamics F I G U R E 1 Schemes of eco-evolutionary dynamics depicting how evolution of one or multiple species can influence population, community and ecosystem features. Panel (a) shows the traditional focal species perspective, in which evolution of one species is considered (influenced by the environment, indicated by inward-oriented blue arrows), and its impact on population, community and ecosystem features (visualized as outward oriented blue arrows) is quantified. Panel (b) depicts a multi-species perspective, in which evolution of the different interacting species that build up a community can influence their own and the other species' population characteristics, and collectively influence community and ecosystem features. We illustrate the pattern for three interacting species, and each species is represented by three individuals. The schemes also visualize that the phenotype, in part determined by the genotype and in part by the environment (cf. arrows within individuals), is central to eco-evolutionary dynamics, as it is phenotypes that are subject to selection and that can influence population, community and ecosystem characteristics shown or assumed to be genetically determined and isolated from contrasting habitats (e.g., Walsh et al., 2012) or generated through experimental evolution (e.g., Pantel et al., 2015) and quantify effects on population dynamics, species composition or ecosystem features.

| The current view on ecoevolutionary dynamics
In his scheme on how evolutionary trait change can impact ecological processes, Hendry (2017) emphasizes the pivotal role of the phenotype, that is, trait values of a species, that can impact population, community and ecosystem features (Figure 1 left). The phenotypes present in a population result from the genotypes present in the population, the environment's influence (phenotypic plasticity) and their interaction (Govaert, Pantel, & De Meester, 2016;Lynch & Walsh, 1998;Via et al., 1995;Via & Lande, 1985). Both the abiotic environment and interactions with other species can determine selection pressures on the phenotype, and this can result in evolutionary changes in the phenotype if phenotypic variation has a genetic basis (i.e., is heritable). These evolutionary changes in phenotypes can then impact ecological properties of populations, communities and ecosystems, resulting in an eco-evolutionary feedback. The pivotal role of phenotypes in this feedback is in line with the recent upsurge in interest in trait-based ecology (Bolnick et al., 2011) and the relative importance of intra-and interspecific trait variation (Violle et al., 2012). Trait values are the common currency linking evolution to ecology: Unless genetic change influences phenotypes, directly through genetic variants coding for different trait values, or indirectly through for example fitness costs linked to inbreeding, evolution will not impact ecology.
While the phenotype is pivotal to eco-evolutionary dynamics, this does not automatically mean that any phenotypic change impacting ecological processes is an example of eco-evolutionary dynamics. Studies on eco-evolutionary dynamics should quantify to what extent the phenotypic change observed actually reflects evolutionary change or genetic differentiation. Several studies presented in the context of eco-evolutionary dynamics fail to do so. Phenotypic trait change might also be caused by ontogeny or phenotypic plasticity, and such non-genetic trait change can be highly relevant for ecological processes (Bolnick et al., 2011;Lundsgaard-Hansen, Matthews, & Seehausen, 2014;Violle et al., 2012). Yet, any study on eco-evolutionary dynamics should at the least document that evolution is involved and that this evolutionary differentiation impacts ecological processes. Assuming that trait differences as observed in the field reflect genetic differences is unwarranted, as it is well-known that the phenotype of individuals in nature is the combined result of the interaction between genotype and environment. Rarely will the phenotypic differentiation as observed in nature only reflect genetic differences, and assuming so will likely in many cases misrepresent the impact of evolution on ecology . Documenting that phenotypic differences as observed in the field reflect evolutionary differentiation involves much work, and hence, it might be attractive to take the short cut of only focusing on phenotypic variation. Yet, confounding non-genetic phenotypic variation with genetic trait differences interferes with the core of what eco-evolutionary dynamics are about.
The scheme presented by Hendry (2017) is effective in visualizing how evolutionary changes in phenotypes can influence population, community and ecosystem features. It also reflects how most ecoevolutionary dynamic experiments are performed: The phenotypes of a given species are manipulated (e.g., by using individuals from genetically distinct populations that differ in their phenotype), and the impact of this intraspecific phenotypic variation on population, community and ecosystem characteristics is monitored in a common gardening approach (Matthews et al., 2011). The common gardening approach has been powerful in providing proof of principle that evolution can affect ecology and that eco-evolutionary feedbacks need to be taken into account if we want to understand how populations, communities and ecosystems respond to environmental change, including human impact, through the changes in phenotypes of taxa.

| Understanding complexity in real ecosystems
If we want to understand eco-evolutionary dynamics in nature, however, we need to take into account that in reality, species co-occur in diverse communities (Figure 1, right). Each species has its own population features and dynamics, and collectively, these species comprise a community. The phenotypes of each species might not only influence population dynamics and other population characteristics of the species itself, but also influence those of other interacting species (Figure 1; McPeek, 2017b). These changes in the population abundances and other population features (e.g., size distribution) of all species collectively determine community characteristics and ecosystem features. While each species can influence population, community and ecosystem features, they do so when interacting: The population features of each species potentially influence or are influenced by the evolution of any member of the community, and it is the joint evolutionary change and its impact on populations that determines community and ecosystem characteristics. While "the phenotype" is still pivotal, it is the phenotypes of all interacting species that is crucial and which captures the more realistic complexities that arise when species can influence each other. This makes a crucial difference in mindset: If we want to understand how ecoevolutionary dynamics play out in nature and go beyond proof-ofprinciple, we need to move towards understanding evolutionary responses and their ecological impacts in multiple interacting species simultaneously. To our knowledge, no study so far has done this in nonmicrobial systems. Figure 1 remains an oversimplification, as the community that is depicted only represents one subset of interacting species. In reality, local systems will be characterized by food webs involving predators (Gravel, Massol, Canard, Mouillot, & Mouquet, 2011;Poisot, Stouffer, & Gravel, 2015), parasites (Lafferty, Dobson, Kuris, & Tilman, 2006;Thompson, 2005) and mutualists (Koskella, Hall, & Metcalf, 2017;Macke, Tasiemski, Massol, Callens, & Decaestecker, 2017) (Figure 2). A body of knowledge exists for predator-prey, hostparasite and host-mutualist co-evolution and how this evolution impacts interaction strengths and population dynamics (Penczykowski, Laine, & Koskella, 2016). Again, most of these studies quantify evolution either in the host or in the parasite and study the impact of this evolution on the other species, with only a limited number of studies incorporating evolution of both partners or three interacting species and how they influence the other partners (Benkman, Holimon, & Smith, 2001;Brunner et al., 2017;Decaestecker, Gersem, Michalakis, & Raeymaekers, 2013;Ford, Kao, Williams, & King, 2016;Frickel, Sieber, & Becks, 2016;Frickel, Theodosiou, & Becks, 2017;Hiltunen & Becks, 2014;Masri et al., 2015;Thompson, 2005). There are, to our knowledge, so far no empirical studies reporting on how evolution in multiple species of predators and prey, hosts and parasites, or mutualists, simultaneously impacts the dynamics of the populations of each individual species, their community composition or resulting ecosystem functions.
Even, Figure 2 is still an oversimplification, as local food webs interact with food webs in other patches in the landscape ( Figure 3). In this spatial context, local selection and evolution in response to abiotic environments and species interactions are linked by regional dispersal and gene flow. Thus, the same tension exists in ecology and evolution between the sorting of species and genotypes locally and the regional remixing of species and genotypes. Regional mixing is often maladaptive because a poorly adapted species or genotype lands in a suboptimal habitat, but it also can provide a match to optimal habitats or create novel, more fit genotypes or more diverse communities of species that provide diversity for selection to act upon. The dynamics of these so-called evolving metacommunities (Urban et al., 2008;Urban & Skelly, 2006) are poorly studied. Exceptions include the elegant proof-of-principle experiment with "communities" represented by different Pseudomonas fluorescens strains evolving in a landscape of food sources in multiwell plates (Venail et al., 2008), the evolution-mediated priority effects experiments of   Low-Décarie et al. (2015). While these excellent experiments grasp key features of evolving metacommunities, they all involve highly artificial settings. Given that theory shows that the levels of regional dispersal can strongly influence how evolution affects community assembly (Loeuille & Leibold, 2008;Urban & De Meester, 2009;Urban et al., 2012;, there is a need to develop methods and approaches to study the features of evolving metacommunities in sufficiently realistic settings, through experimentation and field surveys. Obviously, we do not suggest that studies on eco-evolutionary dynamics that fail to address the full complexity depicted in predator adapts in response to the novel setting that results from the evolution of competing prey. In another example, the capacity to evolve might be hampered or increased in a landscape setting because of gene flow, codetermined by metapopulation and metacommunity structure. Within a landscape, local eco-evolutionary feedbacks might spill over to influence ecological processes (e.g., community assembly), evolutionary trajectories, and eco-evolutionary dynamics in nearby patches by modulating the numbers and types of dispersing individuals.
These suggestions correspond with an ecologist's perspective on eco-evolutionary dynamics. Evolutionary biologists, in contrast, are often biased towards studying species in isolation or strictly coevolving species pairs. We need to move beyond demonstrating that the evolution of just a single focal species impacts ecological F I G U R E 3 In an evolving metacommunity, the interacting local communities forming local food webs inhabiting specific habitat patches can interact through the dispersal (green arrows) of genotypes/species with other such local communities. This results in a regional group of interacting communities comprised of evolving or coevolving species (Thompson, 2005;Urban et al., 2008). The size of the habitats and the corresponding populations might vary widely in real landscapes, contributing to asymmetries in dispersal rates processes. We argue that it is time to incorporate more ecology in eco-evolutionary dynamics, not only as response variables, but also by incorporating an explicit multi-species context both in the eco-to-evo and in the evo-to-eco side of the feedback.

| A NOTE ON DEFINITI ON S: ECO -E VOLUTIONARY DYNAMIC S AND ECO -E VOLUTIONARY FEEDBACK S
Given the explicit multi-species perspective that we here advocate, we need to revisit the definition of an eco-evolutionary feedback. Researchers differ in what they mean by an "eco-evolutionary feedback." The strict definition is that the feedback has to involve the same species and traits (cf. the narrow definition listed by Hendry, 2017): Evolution in a focal species impacts ecological processes, which subsequently impact the further evolution of the focal species. There is empirical evidence for feedbacks that involve the same actors (Becks et al., 2012;Brunner et al., 2017;Bull, Millstein, Orcutt, & Wichman, 2006;Chitty, 1967). In a multi-species context, however, it is difficult to maintain this strict definition, because this would imply that we either have to develop a new term for all other feedbacks between ecological and evolutionary processes, or miss most of the relevant interactions between ecological and evolutionary dynamics. We argue that it is preferable to define an eco-evolutionary feedback as any feedback between an ecological and an evolutionary process (cf. the broad definition listed by Hendry, 2017). Key to documenting such a feedback is that one needs insights into both the impact of ecology on evolution and the impact of evolution on ecology in the same (multi-species) system. To empirically illustrate an ecoevolutionary feedback in a given system (i.e., a local population, local community or local ecosystem, or a metacommunity or metaecosystem), a researcher would want to show that ecological processes impact evolution and that this evolutionary change impacts one or more ecological processes or their resulting patterns. The feedback does not necessarily need to involve the same species or traits and if it does, might do this through multiple intermediate steps ( Figure 4). For example, imagine that climate change causes trait evolution in a predator, and that this evolution affects prey community composition. This situation would conform to an eco-evolutionary feedback even though the ecology of the prey species may then perhaps not affect the evolution of the predator. The key point is that ecology affects evolution and evolution affects ecology, regardless if the same species are involved. If the feedback does result in additional evolutionary changes of the species that evolved in the first place, this can be specified and would be an example of a feedback loop in the narrow sense, resulting in a broad-sense and narrow-sense definition of eco-evolutionary feedbacks ( Figure 4; following Hendry, 2017).
A key reason to adhere to a broad definition of eco-evolutionary feedbacks is that many of the impacts of evolution on ecology that matter for our understanding and prediction of ecological processes do not involve feedbacks in the narrow sense. Some of the more important influences of evolutionary on ecological processes might involve feedbacks on very different members of the food web or even on communities in different localities. The importance of eco-evolutionary feedbacks should be quantified by the effect size of evolutionary change on ecological processes, not by the subject of its influence. Moreover, multiple steps in a cascade of effects might eventually result in a feedback in the narrow sense, even though it was not studied originally because too many steps were involved. For instance, evolution in response to climate change in a lizard might influence community composition of the guild of prey species, which might result in a change in vegetation structure, so that the sparser vegetation imposes selection on body colour in the predator linked to thermoregulation. This example involves only four steps, but it is likely that such cascades might involve many more steps. In some cases, it might at any of the intermediate steps even lead to the local extinction of species or to qualitatively entirely different ecological trajectories.
Especially in a multi-species context, such complexities are likely to be common and need to be considered. By adhering to a narrow definition only, we would limit eco-evolutionary feedbacks to the simplest examples of it.
Our bias to more simple contexts is driven by the ease of its modelling and of experimentation. We argue that there is no intrinsic problem in constructing simplified models and experimental settings, but we need to be careful that the resulting low-dimensional settings do not bias our definitions of what eco-evolutionary feedbacks are and our appreciation of their impact. Much of the impact of eco-evolutionary dynamics for the fields of ecology and evolutionary biology lies in their potential for indirect, multiplicative effects. It is important that we always attempt to understand how far we can simplify a complex system while still retaining its characteristic dynamics and structure. To do that, however, one also needs to study the more complex settings or nature itself.

| THE NEED FOR A MULTI -S PECIE S CONTE X T
Above, we argued that the complexities of multi-species contexts and evolving metacommunities should be more often included in studies on eco-evolutionary dynamics. Otherwise, we risk introducing biases by focusing on the impacts of a single, focal species. We see three main risks. First, species are expected to strongly differ in how their evolution affects ecological processes. This may be the result of (a) differences in evolutionary potential, (b) differences in abundances, (c) differences in dispersal capacity and (d) differences in interaction strength, linked to variation in effect traits. Eco-evolutionary studies focusing on a single focal species are likely biased towards abundant species, strong interactors and species known to evolve. Indeed, many eco-evolutionary feedback studies use species that are abundant in the system studied (e.g., Bassar et al., 2010;Farkas et al., 2013;Urban, 2013), use keystone species (Matthews et al., 2016) or use species that are known to be strong competitors (Pantel et al., 2015;Walsh et al., 2012). There is nothing intrinsically wrong with these choices if one wants to explore how strong eco-evolutionary feedbacks can be. The resulting proof-of-principle studies should, however, not be used to infer how strong eco-evolutionary feedbacks are on average. For instance, in zooplankton communities, Daphnia are generally assumed to be strong competitors and keystone grazers. All else being equal, one might thus expect that eco-evolutionary feedbacks on community structure and on top-down control of algae might be smaller for the other community members, yet no such comparative data exist. Conversely, while our focal species approach might in many cases result in a stronger than average impact of evolution on ecological processes, we might also miss cases of strong impacts because understudied species might show a particularly strong evolutionary response or evolve in unexpected ways that elicit strong ecological responses. If we want to move towards predicting eco-evolutionary effects, we will need to know for which species they are important and for which ones they are not. Therefore, we will require comparative data on species in the same guilds, including those with limited impact. Third, one also needs to consider the community context of the evolutionary response itself. Evolutionary trajectories can be very different depending on whether a species is kept in isolation compared F I G U R E 4 A schematic representation of different types of eco-evolutionary feedbacks. Evolutionary change is depicted as G influencing P in green boxes (right part of each panel), whereas ecological responses are depicted as Eco1/Eco2/Eco3 (left part of each panel). In each scheme, there is at least one arrow from ecological to evolutionary processes and from evolutionary to ecological processes, representing the feedback. (a) Narrowsense eco-evolutionary feedback loop: An environmental change leads to evolutionary trait change, which feeds back to influence further evolutionary change in the focal species; (b-f) broad-sense eco-evolutionary feedbacks, representing eco-evolutionary feedbacks that not necessarily feedback to the original actor: (b) Environmental change leads to evolution, which impacts an ecological process; (c) same but this ecological change influences a second ecological process; (d) same as (b) but the ecological change influences evolution in another species; (e) evolution in the first species leads to evolution in a second species, which influences an ecological process; (f) as in (c) but the second ecological process influences the evolution of the first species. Case (f) depicts the lizard example in the main text: Evolution of thermal tolerance in response to climate change in a lizard influences community composition of its prey, which results in a change in vegetation structure, which influences thermal adaptation in the lizard. The red (dotted line) and blue (full line) boxes emphasize the narrow-and broad-sense definition of eco-evolutionary feedbacks, respectively Eco3

Narrow-sense Eco-evo feedback
Broad-sense Eco-evo feedback to when it is exposed to selection pressures in the presence of a community (Barraclough, 2015). Simulations showed that evolution of response traits can be reduced in the presence of other species (de Mazancourt et al., 2008), while in other settings, the presence of other species can enhance evolutionary responses, especially for traits important in species interactions (Osmond & de Mazancourt, 2013). Community context can thus lead to different evolutionary trajectories compared to the ones predicted from single-species studies, with important ecological consequences (Lawrence et al., 2012). Barraclough (2015) emphasized that evolutionary biologists should carefully take the community context into consideration if they want to obtain insights into how species will evolve in response to specific selection pressures, and this will obviously also impact the ecological consequences of evolution. Community context may even determine whether evolution will occur. If the community harbours species that are preadapted to the environmental change that is imposed, changes in species composition through species sorting might occur so fast that a given species is driven to local extinction before it has the time to evolve in response to the environmental change Vanoverbeke et al., 2016). This might be common, even for species that have a high evolutionary potential.
If common, then studies that use experimental evolution on species in isolation to first generate populations that are adapted to different conditions (e.g., Pantel et al., 2015) risk over-or underestimating the impact of eco-evolutionary dynamics, as the species would have shown less or more evolution if embedded in the natural community.
In this context, one might argue that genetic constraints for evolution in the first place might be important in a community context, as they alter the rate of adaptation and may thus increase or decrease the likelihood that species sorting dominates and impedes evolution .
Finally, a key reason to include a multi-species perspective in a landscape context is that the exchange of species and genotypes among food webs in a landscape ( Figure 3) will affect local evolutionary responses, either through the immigration of preadapted genotypes or species Vanoverbeke et al., 2016) or through maladaptive gene flow (Richardson, Urban, Bolnick, & Skelly, 2014). This stresses the importance of taking spatial and evolutionary processes as well as species interactions into account (Thompson, 2005;Urban et al., 2016), and points to the evolving metacommunity concept provides as an integrative framework for carrying out eco-evolutionary dynamics research Urban et al., 2008Urban et al., 2016).

| KE Y S TUDIE S ON ECO -E VOLUTIONARY DYNAMI C S IN A COMMUNIT Y CONTE X T
To evaluate to what extent empirical studies so far addressed the multi-species context in which eco-evolutionary dynamics operate in nature, we provide an overview of eco-evolutionary dynamic studies that focus on feedbacks at the community level in Supporting Information Table S1. We selected studies documenting how evolution, genetic differentiation or genetic diversity impacts community assembly and community features, because they do incorporate an explicit multi-species context at least in their endpoints. The table illustrates the degree to which these studies fulfil three criteria that are linked to the above highlighted multi-species context, and lists whether (a) the study involved evolutionary change embedded in a natural or multi-species context (i.e., evolution as it occurred in nature or in a multi-species experimental evolution setup), (b) the study involved an analysis of evolution in multiple species (rather than just one focal species) and (c) the endpoint community involved co-evolved populations (i.e., whether the species of the responding community were allowed to co-evolve with the evolving focal species in the experimental evolution trial or came from the same habitats as from which the genetic variants in the focal species were isolated). In defining communities as an endpoint, we refer to communities of guild members, that is, species that can potentially compete with each other. The endpoint community can be guild members of the same trophic level as the evolving focal species or can be a community of guild members that is affected by evolution of species that interact with the community. An example of the latter is the study of terHorst et al. (2014)  Overall, the studies that come closest to grasping the key dimensions of the multiple species context of eco-evolutionary dynamics involve microbial systems. These studies, however, most often are carried out under very artificial conditions, as they involve experimental evolution of multiple species embedded in reconstructed communities in artificial laboratory settings. Some very elegant studies that make inferences on community features are not listed in the table because they involve different strains of the bacterium P. fluorescens rather than multiple species and thus in essence involve only one species (Fukami et al., 2007;Venail et al., 2008). By excluding these or any other studies from the table, we do not want to suggest that their results are irrelevant to the theory of community assembly, but rather that they do not meet the criteria of this particular literature review.

| HOW TO EMB R ACE A MULTI -S PECIE S PER S PEC TIVE IN ECO -E VOLUTIONARY DYNAMIC S
As mentioned, it is often impractical to incorporate the full complexity of natural systems in eco-evolutionary feedback studies. Yet, there is a need to explicitly study and quantify the emergent features that result from including more complexity and exploring to what extent a multi-species context changes our predictions on the strength and direction of eco-evolutionary impacts on ecological processes.
A lot of progress could be achieved if there would be a strong effort towards incorporating specific aspects of the complexity sketched in Figures 1-3 in eco-evolutionary research. While at times challenging, important progress can be obtained through collaborative work with different research teams focusing on different organism groups in the same systems. Alternatively, eco-evolutionary insights for a given system can be built up over time. For instance, a researcher might begin with understanding the evolution of one species and its effects on ecology and then add insights into the effects of evolution in other species of the same system. We here outline a few approaches on how specific dimensions of the complexity resulting from a multi-species and evolving metacommunity framework could be tackled in empirical work.

| Experiments
In experimental studies, there are a number of complexities linked to the multi-species features highlighted in Supporting Information Table S1 that can relatively easily be incorporated, and Supporting Information Table S1 lists some studies that have partially done so.
While these complexities might result in more challenging experiments, they can still be feasible when designed in a reasoned way and would have strong added value. The following is not intended to be an exhaustive list of possibilities on how to implement aspects of a multi-species perspective in eco-evolutionary experiments, but rather is intended to serve as a source of inspiration for the many ways in which a multiple species perspective can be approached: 1. Community context of evolution. In studies that use experimental evolution, quantifying evolution in the presence of other community members should be more often considered (Barraclough, 2015). Studies that compare eco-evolutionary feedbacks upon evolution in isolation versus in a community context might reveal how important this aspect can be, and studies on microbial systems hint on its potential importance (Fiegna, Scheuerl, Moreno-Letelier, Bell, & Barraclough, 2015;Lawrence et al., 2012). Studies based on genetic differences as observed in nature can often be assumed to incorporate this relevant multispecies evolutionary context. However, even in these cases, researchers should be careful to understand how communities covary with selective factors. If community context varies with selection, then the chance exists that evolution is modified not just by the focal selective factor, but by coexisting species as well. Experiments that directly use samples from natural communities to initiate the gardens for a common gardening approach will naturally incorporate these interactions. However, many such experiments create communities from whatever local community is easiest to sample rather than the appropriately co-evolved community. One interesting way to explore the importance of these interactions is a double transplant, where genetically differentiated populations of a focal species are transplanted in two types of common gardens, that is, seeded with the communities of the two source habitats of the focal species. Again, studies comparing eco-evolutionary impacts in the presence and absence of co-adapted species might provide us with crucial insight into the importance of such co-adaptation. In a context of host-parasite (Brunner et al., 2017;Decaestecker et al., 2013;Frickel et al., 2016;Masri et al., 2015) and mutualistic (Ford et al., 2016;Macke et al., 2017) interactions, there is already a quite large number of studies that take co-evolution into account in their analysis of joint dynamics of the two interacting species. Recently, co-evolution was also shown to impact invasion success of exotic species (Faillace & Morin, 2016). Yet, we know of no empirical studies that compared the strength and direction of eco-evolutionary feedbacks on community members that were either allowed or not to co-evolve in response to the same experimental conditions together with the focal species. Such experiments might reveal strong cases of cryptic eco-evolutionary dynamics.
3. Resurrecting communities for reciprocal transplants. The resurrection ecology approach (Decaestecker et al., 2007;Franks, Hamann, & Weis, 2018;Stoks, Govaert, Pauwels, Jansen, & Meester, 2016;Sultan, Horgan-Kobelski, Nichols, Riggs, & Waples, 2013) applied to multiple species simultaneously might provide a powerful way to obtain insight into the impact of evolution as it occurred in nature on ecological processes. Here, again one could test the impact of evolution of every species separately and in combination, and carry out "transplants" over time (Houwenhuyse, Macke, Reyserhove, Bulteel, & Decaestecker, 2018;Penczykowski et al., 2015), replacing evolved populations by representatives of their ancestors either for the whole community or for each of the member species, and quantify its feedback on ecological processes. Resurrection ecology can be applied on layered archives of dormant stages (mainly in aquatic systems, e.g., Stoks et al., 2016) or when dormant stages have been collected at different moments of a population's history (Franks et al., 2018).

Field transplants with adapted/nonadapted species sets. In field
transplant experiments, an interesting avenue might be to carry out "community transplants," in which entire co-evolved communities are reciprocally transplanted (Alexander, Diez, & Levine, 2015) and compared to treatments in which community composition is maintained but in which the populations of all dominant species or a selection of key species is replaced by members of the same species but using genotypes isolated from the other habitat.
The latter approach is only possible if community composition is not too divergent among habitats, but would provide an integrated analysis of the impact of the joint evolution of all species.
Ultimately, we want to understand natural systems. Therefore, a critical, but often missing, step is comparing eco-evolutionary impacts between experiments and wild systems. For example, a common gardening experiment might reveal a strong feedback from the evolution of increased grazing on primary productivity. The next step should be to evaluate whether primary productivity is lower in natural systems in which higher grazing rates have evolved (Walsh et al., 2012). If not, then some critical ecological or evolutionary feature is likely missing from the common gardening experiments, including co-evolved species that could moderate the effects. For instance, the evolution of increased foraging rates of an intermediate consumer in response to selection from an apex predator increased prey diversity and decreased density in mesocosms (Urban, 2013). Across ponds differing in the evolution of the focal species, prey diversity and density differed in the same ways as the experiments, suggesting an important eco-evolutionary dynamic had been captured by the simplistic experiments.
While such comparisons do not provide solid proof, because similar patterns might in principle also be generated by ecological processes, their results can be suggestive of the potential importance of the ecoevolutionary feedback in nature. Overall, we should not lose sight of the main reason to study eco-evolutionary interactions: to understand natural dynamics and patterns.

| Field data
Ideally, we would like to be able to quantify the impact of evolution on ecological processes directly in the field, as this would capture natural complexity in a real metacommunity setting instead of in isolated, artificial settings typically constructed in the laboratory. The key problem with field surveys is, however, that they yield patterns from which processes need to be inferred and that similar patterns might result from many processes. This imposes an important limitation on the degree to which mechanisms can be disentangled, including making the distinction between evolutionary and ecological processes as drivers of a particular pattern or change. To make strong inferences, it will often be necessary to tap into additional sources of information. Partnering field surveys with experiments is a particularly valuable approach. Below, we suggest some approaches that can provide insight into the role of evolution in structuring natural metacommunities, each with their own limitations: 1. Contribution of evolution to community trait values in a metacommunity. If one has data on a spatial survey in which not only the species composition but also genetically determined trait values of local populations of key species have been measured, one can use either local or regional averages of trait values of one or more of those species to quantify how local trait variation affects community trait values across the metacommunity (Brans et al., 2017). This can be done using regressionbased methods (Lajoie & Vellend, 2015) or using methods that partition ecological and evolutionary contributions to community trait change (as in Govaert et al., 2016, but adapted to spatial data; L. Govaert et al. unpublished). In order to be able to apply eco-evolutionary partitioning metrics or to apply the regression-based method on data reflecting evolution, one needs genotypic trait values for local populations of the (multiple) focal species (Brans et al., 2017). (species sorting, natural selection) and spatial (dispersal) drivers (Cottenie, 2005;Leibold et al., 2004). Again, in order to be able to link these patterns to evolutionary trait change, the intraspecific trait variation needs to be quantified in common garden experiments, isolating the genetic component of trait variation. In the resulting dataset, one can, in addition to linking variation in species composition, community trait distribution and intraspecific trait variation in the focal species to local (environmental) and regional (spatial) drivers, also explore to what extent intraspecific trait variation in the focal species is associated with the deviations of the expected community trait variation. In these analyses, one cannot only quantify the structure of the evolving metacommunity (Urban et al., 2008), but also obtain insight in the likelihood that evolutionary trait change contributed to metacommunity structure and community trait distribution.
3. Population genomics in an evolving metacommunity context. In the future, population genomics (Rudman et al., 2018) will likely be key to reduce the amount of work while still being able to link genetic differences to ecological responses (e.g., community composition and population density). For this approach to be powerful, we will need further insights in how genomic data link to trait values in target species. Given that these links might be complex and not necessarily strongly repeatable (e.g., Becks et al., 2012;Gompert et al., 2014;Nosil et al., 2018), there is still a long way to go for population genomics to translate in reliable estimates of trait variation, except for some well-studied cases that involve major effect loci or gene clusters (Jones & Gomulkiewicz, 2012;Lamichhaney et al., 2016).
As the link between genomic variation and trait values becomes more reliable, however, it will open tremendous opportunities for studying eco-evolutionary dynamics in the field (Rodríguez-Verdugo, Buckley, & Stapley, 2017;Rudman et al., 2018).
A key limitation of all the analyses highlighted in the previous paragraphs when based on field surveys is that they only reveal the current pattern, implying that important transient changes that might have been caused by evolution but leave no trace in the current metacommunity cannot be quantified. This is very different from how we quantify eco-evolutionary dynamics in an experimental approach such as common gardening experiments. In a common gardening set-up, one starts with standardized "gardens" to which different genotypes of one or more species are added to then monitor how the garden changes as a function of the manipulated genotype (Matthews et al., 2011). In such settings, one quantifies the impact of evolution as the differentiation of the garden among treatments developing from a common starting point. In most field surveys, however, one only has a snapshot view of the resulting pattern. There is no guarantee that initial environmental conditions or initial community composition of the different sites was identical, and it is actually very unlikely that they were. So samples. Yet, linking this disappearance to evolution might not be straightforward. Conducting repeated common garden and common gardening experiments in these systems over time might be powerful to reveal evolution and its impact on ecology. Along the same lines, combining survey data with resurrection ecology (Decaestecker et al., 2007;Franks et al., 2018;Houwenhuyse et al., 2018;Stoks et al., 2016;Sultan et al., 2013) might prove very powerful. As having temporal series of data is important, we here make a plea to develop research programs on eco-evolutionary dynamics that involve long-term ecological research (LTER) sites (Knapp & Smith, 2001) and/or sites amenable to resurrection ecology approaches (Franks et al., 2018). If one has very consistent and detailed time series, one can apply process-oriented approaches such as, in the case of population dynamics, integral projection modelling (IPM; Coulson, Tuljapurkar, & Childs, 2010;Easterling et al., 2000;Smallegange & Coulson, 2013). Several of IPM-inspired analyses have been published on a number of unique datasets on trait values and population dynamics of mammals to show that evolutionary trait change impacts population dynamics in the field (Ozgul et al., 2010;Pelletier, Clutton-Brock, Pemberton, Tuljapurkar, & Coulson, 2007;Pelletier, Garant, & Hendry, 2009).
Extending this approach to include multiple species would open important opportunities. The availability of the necessary data will, however, be an important obstacle for such an approach, especially because most LTER sites collect valuable ecological data, while neglecting evolutionary data. The most powerful approaches to multispecies eco-evolutionary dynamics studies would be the combination of well-designed field studies that identify cases of particular interest, which are then combined with the appropriate experimental approaches (such as field transplants, common garden experiments, resurrection ecology or experimental evolution) to test particular hypotheses. To our knowledge, no study so far engaged in such an endeavour in a truly multiple species context. While this will no doubt be a huge effort, it would be worth it, as it would allow quantifying the importance of eco-evolutionary dynamics in ecologically relevant settings and would likely reveal several cases of cryptic eco-evolutionary dynamics (Kinnison et al., 2015).

| US ING COMPLE XIT Y TO MAK E THING S S IMPLE AG AIN: EMERG ENT PAT TERN S AND RU LE S
A key problem with increasing study complexity is that the results are often contingent on starting conditions and that apparent minor differences can lead to strongly deviating patterns and trajectories (Fukami, 2015;Losos, Jackman, Larson, Queiroz, & Rodriguez-Schettino, 1998;Scheffer, Carpenter, Foley, Folke, & Walker, 2001). This contingency results in reduced power to derive generic laws (Lawton, 1999) or to make predictions on responses to future change  We thus need to obtain insight in the importance and the resulting patterns of eco-evolutionary dynamics in a multi-species context for two reasons. First, ignoring complexity may lead to erroneous judgements on the importance or even the direction of eco-evolutionary dynamics, so that relying on proof-of-principle experiments in highly simplified settings is risky, for instance when making predictions on responses to human-induced change .
Adding a community or metacommunity context might change dynamics in a spectacular way (Kinnison et al., 2015), but has not been systematically explored. Second, while the results of studies that add a multi-species context might depend on the systems, species and environmental conditions studied, investment in a large number of eco-evolutionary dynamics analyses in a community context or in nature might reveal patterns that can be translated into "rules" that we can apply to estimate the likely effect of evolution on ecological processes given certain conditions. Potential rules are for example, the suggestion made by Bailey et al. (2009) that the impact of evolution might fade as one moves from populations to communities to ecosystems, because of an increasing number of potential sources of variation as complexity increases. While this seems a logical rule, it might be challenged in the case of evolution in traits that directly might impact ecosystem functioning. Another potential rule is that the amount of eco-evolutionary dynamics might scale with body size, because of shorter generation times enabling higher rates of evolutionary change. Many processes scale with body size (Peters, 1986), and there is evidence for such a scaling in metacommunity ecology (De Bie et al., 2012).
Moreover, for microbial organisms it has indeed been shown that a few days of difference in time of establishment is sufficient for evolutionary diversification to impact ecological dynamics (Fukami et al., 2007;Venail et al., 2008;Zee & Fukami, 2018). So there are indications that eco-evolutionary feedbacks might be more prevalent in organisms with short generation times , even though it has not been systematically analysed and clear cases of eco-evolutionary feedbacks have been reported for many long-lived species too (Hendry, 2017). We also note that when ecoevolutionary feedbacks are studied across community types (e.g., gut microbiota or microbial communities that are eaten upon), timescales differ among interacting partners, enhancing the potential of organisms with short generation time to have a strong impact on ecological processes through trait evolution. This has been well-established in studies on host-parasite interactions (Gandon, Buckling, Decaesteckerà, & Day, 2008) but also applies in a broader context for all co-evolutionary and eco-evolutionary dynamics (Dercole, Ferrière, Gragnani, & Rinaldi, 2006). Other potential rules might be that evolutionary dynamics might be more important in species-poor (Urban et al., 2008) and more isolated systems Vanoverbeke et al., 2016). Developing insights in such rules might involve the combined effort of empirical work and targeted modelling. Theoretical modelling can help to identify instances where eco-evolutionary feedbacks could potentially be important (Govaert et al., 2018), or to test the strength of these feedbacks in more complex systems (de Andreazzi, Guimarães, & Melián, 2018).
Further development of theory will be crucial in order to be able to incorporate eco-evolutionary dynamics in predictions  and management beyond the specific settings that have been empirically documented.

| CON CLUS IONS
In the past, ecologists ignored evolution, whereas evolutionary biologists ignored ecological interactions in multi-species communities. The field of eco-evolutionary dynamics is starting to alert ecologists to the substantial impacts of evolutionary change in their models and empirical work. We hope that this opinion paper together with other recent reviews and synthesis papers (e.g., Barraclough, 2015;Kinnison et al., 2015) will encourage researchers working on eco-evolutionary feedbacks to not focus on single species in isolation but to consider realistic ecological contexts in their work. While integrating evolution into ecology can revolutionize ecology, we predict that having a multi-species ecological context built into our eco-evolutionary models and empirical framework may be equally important in revolutionizing the field of eco-evolutionary dynamics. At this stage, we do not know to what extent the multi-species context hinders rapid evolution or promotes it, and how evolutionary change impacts ecological features that are the result of the interaction of multiple species beyond the simplest of modules (one single predator and prey species, or one host and one parasite species). Simple species interaction modules for the most abundant species and strongest interactors might perhaps capture dominant eco-evolutionary processes sufficiently to understand key features of some natural communities. However, in other cases, indirect effects and cryptic eco-evolutionary dynamics will necessitate a more sophisticated understanding. There is a need for a research framework to tackle eco-evolutionary dynamics in a truly multispecies context. Obviously, responses will vary widely, but the question is whether we can detect emergent patterns that allow us to predict how the dynamics will be influenced as we build in more realism. Combining this realism with the rigour in testing hypotheses in eco-evolutionary dynamics might be the key challenge in the field of eco-evolutionary dynamics in the coming decade.

ACK N OWLED G EM ENTS
We acknowledge financial support from the KU Leuven Research Fund (project C/16/17/002) and from National Fund for Scientific Research-Flanders (FWO) projects G0C3818 and G0B9818. M.C.U.
received support from NSF grants DEB-1555876 and PLR-1417754.
We thank two reviewers, associate editor Martijn Egas and senior editor Charles Fox for their feedback on a previous version of the manuscript.

CO N FLI C T O F I NTE R E S T
The authors declare no competing interests.