Abstract
Coevolutionary dynamics act on both species and their interactions in ways that shape ecological communities. It remains unclear, however, how the structure of communities at larger spatial scales influences or is influenced by local coevolutionary processes, and how mechanisms acting at these different scales feedback onto one another. Here we show that, although species interactions vary substantially over a continental gradient, the coevolutionary significance of individual interactions is maintained across different scales. Notably, this occurs despite the fact that observed community variation at the local scale frequently tends to weaken or remove community-wide coevolutionary signal. When considered in terms of the interplay between community ecology and coevolutionary theory, our results demonstrate that individual interactions are capable and indeed likely to show a consistent signature of past coevolution even when woven into communities that do not.
Ecological interactions often exert important selective pressures on the species involved. For example, the phenologies of lodge-pole pines and red crossbills respond spatially to the presence of squirrels (Benkman et al. 2003). Likewise, palm species undergo changes in seed morphology in response to the extinction of bird dispersing their seeds (Galetti et al. 2013). Interactions can be lost, too, when phenologies of the species involved shift (Rafferty et al. 2015). Kritsky (1991) relates the discovery of the moth Xanthopan morganii, with a proboscis famously over a foot long, which Darwin predicted would exist based on the phenology of local plant Angraecum sesquipedale. In addition, interactions, and the emergent structures they define, are distributed in similar ways across communities at both large or small scales (Jordano et al. 2003). Together, these observations suggest that much ecological structure could be the end result of (co)evolutionary dynamics between species (Eklof et al. 2011; Stouffer et al. 2012). Unfortunately, although the coevolutionary dynamics of pairs of interacting species have been well described at macro-evolutionary (Van Valen 1973) and micro-evolutionary (Gandon et al. 2008) timescales, most attempts to understand how they cascade up to the levels of diversity of both species and interactions found within empirical communities have been inconclusive (Hembry et al. 2014). Notwithstanding, coevolutionary dynamics are often presented as a key driving force behind ecological structure across both time and space (Thompson 1994; Thompson 2005); it is therefore crucial to determine the scale at which they are both relevant and quantifiable.
Historically, the evidence for coevolution in taxonomically diverse communities is quantified as the degree of matching between the phylogenies of two sets of interacting organisms (Legendre et al. 2002). This notion builds on the century-old idea that extant species interact in a way similar to the way their ancestors did (Fahrenholz 1913; Guimarães Jr et al. 2011; Nuismer et al. 2013). Note that testing these assumptions is related to, but markedly more restrictive than, testing for phylogenetic conservatism of species’ interactions (Rezende et al. 2007; Eklof et al. 2011). This is because of additional, higher-order constraints related to the shape of both trees at all depths (Cavender-Bares et al. 2009; Mouquet et al. 2012): ancestral constraints create high phylogenetic inertia which carries forward to extant taxa (Desdevises et al. 2003; Diniz-Filho and Bini 2008; Vale and Little 2010). For this reason, although several systems have been described that exhibit matching phylogenetic structure, many deviate from this assumption for a variety of factors. Detecting matching phylogenies for interacting clades nonetheless indicates that their coevolutionary history is long standing and is therefore suggestive that their extant ecological structure is an outcome of ancestral constraints and/or co-adaptation (Nuismer and Harmon 2014).
Nevertheless, it is important to note that there is more to coevolution than simply observing matching phylogenies or than observing phylogenetic structure of species interactions (Johnson and Stinchcombe 2007). At a large scale (i.e. both temporal, spatial, and organizational), true coevolution addresses both of these dimensions simultaneously: measures of coevolution yield a positive signal when (i) phylogenetic trees are congruent (ii) based on the observation that species at similar positions in both trees interact. This line of thinking does more than building on extent interactions; because of the branching nature of trees, it ensures that the congruence informed by interactions is true at all phylogenetic depths (Nieberding et al. 2010). How different methods to measure coevolution deal with this structure varies, but at least one common thread is that they address macro-evolutionary questions on the basis of macro-evolutionary structures (Price 2003). Although this is somehow different from micro-coevolution (i.e. within and between populations at reduced temporal and spatial scales), this is no less an instance of coevolution. To a certain extent, micro-coevolution (i.e. reciprocal selection over ecologically relevant timescales) proceeds from existing co-phylogenetic structure. For how are species entangled in interactions, if not by their previous evolutionary history? And although matching phylogenies are not expected to result from micro-evolutionary processes (Poisot 2015), there is no valid ground to reject matching phylogenies with matching interactions as proof of a shared evolutionary history, which we will henceforth refer to as coevolution.
The considerations outlined above can be expressed as quantitative predictions. Communities that have assembled by successive divergence events due to coevolution should display phylogenetic congruence, that is (i) have similar phylogenetic trees and (ii) have species at matching positions in the trees that tend to interact (Page 2003). Of course, this matching can be imperfect, as some interactions display substantial variability at ecologically relevant temporal and spatial scales (Poisot et al. 2012; Carstensen et al. 2014; Olito and Fox 2015; Trøjelsgaard et al. 2015), and the same two species can interact in different ways under the effect of local environmental contingencies, spatial mismatch in species phenologies, variations in population abundances, and chance events (Poisot et al. 2015). Variability of interactions, however, does not predict (i) how the coevolutionary signal of pairwise interactions is kept or lost at the scale of the whole community nor (ii) whether or not this variability is related to changes in the amount of coevolutionary signal that can be detected locally.
In this manuscript, we analyze a large dataset of over 300 species of mamallian hosts and their ectoparasites, sampled throughout Eurasia, for which phylogenetic relationships are known. Using a Procrustean approach to quantify the strength of coevolutionary signal (Balbuena et al. 2013), we show that locally sampled communities rarely show strong evidence of coevolution despite the fact that the overall system does at the continental scale. We then provide evidence to support the conclusion that the amount of coevoluationary signal within a local community is predictable based on the importance of interactions for coevolutions in the regional network. We finally show that the contribution of these interactions to coevolution is invariant across scales, and is unrelated to their tendency to vary across space. These results suggest that the key unit at which coevolution ought to be studied is the interaction rather than the complex networks they form, and this is true even at large taxonomical and spatial scales.
Methods
Data source and pre-treatment
We use data on observations of interactions between 121 species of rodents and 205 species of parasitic fleas in 51 locations across Europe (Krasnov et al. 2012b) to build 51 species-species interaction networks. Interactions were measured by combing rodents for fleas, a method that gives high quality data as it has a high power of detection. Previous analyses revealed that this dataset shows significant coevolutionary signal at the continental level (Krasnov et al. 2012a). Importantly, it also provides spatial replication and variability (Canard et al. 2014) at a scale large enough to capture macro-ecological processes. This dataset is uniquely suited for our analysis, as it represents a thorough spatial and taxonomic sampling of a paradigmatic system in which interspecific interactions are thought to be driven by macroevolution and co-speciation events (Combes 2001; Verneau et al. 2009);
The original dataset gives quantitative interaction strengths (expressed as an averaged number of parasites per species per host). In this system, quantitative interaction strengths were shown to be affected to a very high degree by local variations in abundance across sampling locations (Canard et al. 2014), and it therefore seems unlikely that they reflect macro-ecological processes. Therefore, to account for differential sampling effort—which cannot readily be quantified—and across site variations in abundance—which do not pertain to macro-evolutionary proccesses—we only study the networks’ bipartite incidence matrices (that is, presence and absence of infection of hosts by the parasites).
Spatial scales and interaction spatial consistency
Noting that variation of interactions across locations—which can be caused by local ecological mechanisms, as opposed to reflecting evolutionary dynamics—can decrease congruence, we analyze the data at three different levels which we will refer to as continental, regional, and local. Notably, the continental level summarizes the complete dataset whereas both the regional and local levels are location-specific scales.
The first, continental interaction data consists of the aggregated “metanetwork” which includes all documented interactions between species from the regional species pool (Poisot et al. 2012).
The second, regional interaction data accounts for different species composition across sites, specifically by testing whether sampling from the regional species pool affects coevolutionary signal. Within each site, the regional scale is given by the subset of the metanetwork formed by the locally present species (properly speaking, the induced subgraph of the metanetwork induced from the nodes of the local network). Hence the regional networks are always a perfect subset of the continental network, and do not reflect whether species were actually observed to interact locally or not, but whether they can interact at all.
The third, local interaction data also accounts for variation in the interactions between observed species, in addition to encompassing the above. In contrast to the regional scale, the local scale includes only the interactions that were actually observed in the field at a given site. Therefore, the local and regional networks always include the same species, but the local network has only a subset (or, at most, an exact match) of the interactions in the regional network.
We finally define the spatial consistency of every interaction as the number of sites in which the two species involved co-occur, or simply the spatial consistency of an interaction Sij between species i and j is measured by dividing the number of locations in which both are present (Cij) and the number of locations in which they interact (Lij). Because Lij ∈ [0, Cij], this measure takes values in [0, 1]. Larger values reflect high spatial consistency. Note that although they are reported as 0 (i.e. having no interactions), we actually have no information about species pairs that have never co-occured; this is a common, but hard to correct, feature of spatially replicated datasets in which species occurrence varies (Morales-Castilla et al. 2015). Therefore, the values of Sij can only be defined for species that have been observed to co-occur at least once.
Quantifying coevolutionary signal
We quantify the strength of coevolutionary signal in terms of the degree of matching between host and parasite phylogenies, given knowledge of extant species interactions (at varying spatial scales). We do so using the PACo method (Balbuena et al. 2013), which is robust to variations in both number of species and interactions. PACo provides measures of both the network-level congruence (i.e., is the network coevolved?) and the interaction-level signal (i.e., what is the contribution of each interaction to the overall coevolutionary signal?). Strong values of the later metric reflect low contributions to coevolution – interactions that contribute strongly to phylogenetic congruence have low PACo values. Importantly, and by contrast to previous methods such as ParaFit (Legendre et al. 2002), PACo also can be used to meaningfully quantify the contribution of every interaction to the network-level signal even in cases where the entire network shows no significant coevolutionary signal. As required by PACo, the phylogenetic trees for hosts and parasites were rendered ultrametric (i.e., all species are at the same distance from the root).
Results and discussion
Local and regional scale networks show no coevolutionary signal
As host-macroparasite interactions are hypothesized to be ecologically constrained, as a result of their being evolutionary conserved (Combes 2001), the congruence observed at the continental level sets the baseline for what would be expected in local communities. Of course, if ecological mechanisms reduce coevolutionary signal, we should detect coevolution at the continental scale but not locally. Out of 51 sites, 35 show no signal of coevolution at all, 11 show significant coevolutionary signal when using the regional interactions, and 12 show significant co-evolutionary signal using the local interactions (see Supp. Mat. 1 for network-level significance values; Figure 1). These results support the idea that macro-evolutionary processes, such as co-diversification, can have consequences at the macro-ecological level but may not in fact be detectable at finer spatial scales.
Coevolutionary signal is predicted by the contribution of interactions
On the other hand, system-level differences say little about the behavior of individual interactions. Despite the fact most coevolutionary mechanisms act at the interaction level (Thompson 1999), most measures of it are expressed at the community level. We observe here that networks with interactions that are important for coevolution at the continental scale indeed have more coevolutionary signal at the local and regional scales alike (Fig. 2A). Intriguingly, we also find that the distribution of individual interactions’ contributions to coevolution is strongly conserved, regardless of the scale at which the interactions are quantified (Fig. 2B). Because interactions differ in their total contribution to coevolution, this implies that their distribution across networks (i.e. whether the local network is a sampling of strongly contributing, or weakly contributing, interactions) is what actually drives differences in overall coevolutionary signal. Network-level coevolutionary signal emerges directly from the properties of interactions and is not a property of the network itself.
Interactions contributing to coevolution are not more spatially consistent
Beyond their contribution to coevolution, interactions also ultimately differ in how frequently they vary when the species involved co-occur (Carstensen et al. 2014; Olito and Fox 2015; Trøjelsgaard et al. 2015). This can happen, for example, when one of the partner is able to forage for optimal resources (Betts et al. 2015). Once more, the literature on host-parasite interactions assumes that the reason why some interactions are more frequent is because they reflect a significant past history of coevolution (Guimaraes et al. 2007; Morand and Krasnov 2010); that is, the ecological constraints emerge from the evolutionary conservatism. If this were true, we should observe a significant, positive correlation between the probability of observing an interaction and the importance of that interaction for coevolution at the continental scale. Surprisingly, we find that neither is true here since interactions that are important for coevolution are not more spatially consistent (Fig. 3). This implies that the spatial consistency of an interaction do not reflect its evolutionary past, but rather (extant) ecological processes.
The contribution of interactions to coevolution is consistent across scales
Ultimately, coevolutionary signal varies across scale because of the simultaneous variation of species’ interactions and communities’ phylogenetic tree structure. In a system characterised by substantial turnover we would expect the contribution of each separate interaction to differ across scales as well. Instead, we observe here that interactions that contribute strongly to coevolutionary signal at the continental scale also show a significant tendency to contribute strongly at the local (p < 0.05 for positive correlations in 48 out of 51 networks) and regional (in 47 out of 51 networks), and this observation is independent of network-wide coevolutionary signal (Fig. 4). Remarkably, this result implies that the remnants of coevolution are still locally detectable in individual interactions even though coevolution regularly fails to leave its imprint on most local networks.
Conclusions
Overall, the results of our analyses demonstrate that there is a sizeable gap between our current understanding of host-parasite coevolution as the basis of multi-species interactions and its applicability to ecological questions. Local networks show little to no signal of coevolution and the strength of coevolution between two species is a surprisingly poor predictor of how frequently they interact. In contrast to the frequent assumption that phylogenetic structure is a key driver of community structure (Cavender-Bares et al. 2009), these data reveal that this impact is actually minimal at ecologically relevant spatial scales. And yet, despite all the above, individual interactions are somehow able to maintain their coevolutionary signal even when the community they are woven into does not. Thinking more broadly, these discrepancies provide a clear roadmap for bridging the gap between our appreciation of the role of coevolution and its empirically measurable outcomes: network structure is the most parsimonious mechanism by which coevolution proceeds, not the imprint coevolution leaves on ecological communities.
Acknowledgements
We thank Juan Antonio Balbuena for discussions about the PACo method, and members of the Stouffer and Tylianakis groups for comments on an early draft of this manuscript. We are indebted to Matt Hutchinson and Fernando Cagua for contributions to the code of the paco R package. Funding to TP and DBS was provided by a Marsden Fund Fast-Start grant (UOC-1101) and to DBS by a Rutherford Discovery Fellowship, both administered by the Royal Society of New Zealand.