Tests of alternative evolutionary models are needed to enhance our understanding of biological invasions

I. Introduction

Biological invasions, whereby species are introduced outside their native range and spread rapidly (Richardson et al., 2000), are a global, cross-continent phenomenon (van Kleunen et al., 2016), and an inherently macroecological process. The notion that biological invasions might interact with macroevolution has a long history, and includes ideas proposed by Darwin (1859). ‘Darwin's naturalization conundrum’ (sensu Diez et al., 2008) identifies hypotheses for an intersection between ecological patterns of invasion and macroevolutionary history, as reflected in phylogenetic patterns. In a hypothesis that has been called ‘Darwin's naturalization hypothesis’ (sensu Daehler, 2001), Darwin (1859) proposed that introduced species might be more successful in communities without close relatives. For example, this might occur if competition is stronger between closely related species, leading to competitive release in communities with more distant relatives (but see Cahill et al., 2008; e.g. Burns & Strauss, 2011), or if close relatives share predators, leading to predation release in communities with distant relatives (Hill & Kotanen, 2009). Darwin also proposed the ‘pre-adaptation hypothesis’ (sensu Ricciardi & Mottiar, 2006), in which he suggested that the presence of close relatives might signal shared ‘pre-adapted’ traits, leading to greater invader success (Darwin, 1859). Evidence for both hypotheses, as well as no pattern in some studies, leaves open many possible interpretations (Duncan & Williams, 2002; Ricciardi & Mottiar, 2006; Strauss et al., 2006; reviewed in Procheş et al., 2008; Cadotte et al., 2018). Thus, Darwin's naturalization conundrum remains largely unresolved (reviewed in Cadotte et al., 2018).

An assumption underlying Darwin's hypotheses is that close relatives are generally ecologically similar, perhaps having similar functional traits that influence competition for resources, interactions with natural enemies or adaptation to the environment. The assumption that phylogenetic distance reflects functional distance has often gone untested, and the studies that have tested this assumption often do not find that close relatives are functionally similar, or at least not in a simple, linear fashion (Cadotte et al., 2009; Flynn et al., 2011; Zheng et al., 2018). Although testing for a linear correlation between phylogenetic and functional distance is a reasonable first step, it does not take advantage of the power of models of trait evolution, which generate testable predictions and may lead to new insights into the interactions between ecology and evolution (Cadotte et al., 2018).

II. Why we need an explicitly evolutionary perspective

Recent progress in the modeling of trait evolution can lead to testable alternative hypotheses for a role of macroevolution in biological invasions (Cadotte et al., 2018). Evolutionary models can be used to simulate trait evolution on a phylogeny or test alternative models of trait evolution for empirical trait data. For example, the Brownian motion (BM) model of evolution is a single-parameter model that incorporates an evolutionary rate, σ². The BM model of evolution models drift, or selection that shifts randomly over time, and is often referred to as ‘random walk’. As lineages evolve, they become progressively more divergent, without bound. Thus, trait variance across lineages becomes progressively larger over evolutionary time. The Ornstein–Uhlenbeck (OU) model of trait evolution explicitly incorporates both drift, like the BM model, and selection (Hansen, 1997; Butler & King, 2004). The OU model includes σ², or the rate of drift, α, the strength of selection, and θ, an optimum trait value. For OU models of evolution, trait variance increases across lineages over evolutionary time, within bounds described by α, the strength of selection. For OU models of trait evolution, phylogenetic distance is not expected to scale linearly with functional distance (Cadotte et al., 2018).

The incorporation of evolutionary models leads to specific predictions about the relationship between phylogenetic and functional distance (Cadotte et al., 2018; Tucker et al., 2018). This approach goes beyond simple tests for the presence of phylogenetic signal (e.g. using Blomberg's K, which assumes BM evolution (Blomberg et al., 2003)). For a single trait evolving under BM, Cadotte et al. (2018) predicted that the variance in functional distance should increase at greater phylogenetic distances (Table 1). When multiple traits evolve via BM, the relationship is more linear, with greater functional distance predicted at greater phylogenetic distances (Cadotte et al., 2018). These two models make different predictions about invasions. For example, for multiple traits under BM, more distant relatives should be more functionally distinct and thus might compete less strongly. For a single trait under BM, more distant relatives should be more variable in their strength of competition. For an OU model of evolution, one might predict distant relatives could be less functionally distinct because they are constrained by selection on key traits, possibly leading to strong competition between distant relatives (Table 1). Thus, evolutionary models suggest alternative hypotheses about how macroevolutionary processes might shape invasions.

III. A case study invasion experiment

Observational studies of Darwin's hypotheses typically consider the number of taxa that invade a community, and whether the community members are closely or distantly related to the invaders. In addition to the probability of successful invasion, Darwin's hypotheses also suggest that the performance of a single invader might differ in communities of close or distant relatives (Li et al., 2015). For example, if close relatives compete strongly, then invader biomass might be highest in communities of distant relatives (i.e. the invader might experience competitive release when growing with more distant relatives) (Li et al., 2015).

Here, we present new analyses of our previous experimental data (Zheng et al., 2018), incorporating evolutionary models. In a common garden experiment manipulating the composition of resident plant communities, we (Zheng et al., 2018) found that communities with functionally distant close relatives had the greatest invader biomass. This runs counter to the classic assumption that close relatives should be functionally similar (Darwin, 1859). There was greater divergence between the invader and some Asteraceae than between the invader and more distant relatives. Thus, the relationship between phylogenetic distance and functional distance is not simply linear and positive (Fig. 1). Rather, the relationship between phylogenetic distance and functional distance is qualitatively more consistent with the predictions of an OU model of evolution (Fig. 1; compare with Fig. 5c in Cadotte et al., 2018).

Although analyses with 21 species might have low statistical power (Blomberg et al., 2003), model comparisons on the three functional traits measured in this experiment address this possibility. An OU model (−log likelihood = −73.47) was a better fit than a BM model (−log likelihood = −75.59) of evolution (P = 0.039) to the functional trait relative growth rate (RGR) (Fig. 2). Mexican Asteraceae were especially fast growing, and Chinese species from the introduced range and non-Asteraceae were typically slower growing (Fig. 2). The OU model (−log likelihood = −120.77) fit specific leaf area (SLA) slightly, although not significantly (P = 0.063), better than a BM model (−log likelihood = −122.49), and OU model (−log likelihood = −108.50) was a significantly (P = 0.00046) better fit than a BM model (−log likelihood = −114.63) for leaf area ratio (LAR). Overall, these analyses are consistent with a strong role for selection in the evolution of these functional traits. This leads to a new question: what selective factors have led to divergence among close relatives (Table 1)? For example, has competition in the past caused co-occurring close relatives to diverge? Are close relatives more likely to have diverged in allopatry, leading to divergent close relatives (Cadotte et al., 2017)? Do sympatric and allopatric speciation lead to different patterns for invaders?

Although this experiment suggests some intriguing patterns, it was not designed primarily to test macroevolutionary hypotheses about biological invasions. For example, we might expect alternative patterns if the common garden experiment, and traits such as RGR, were measured in the native range (Mexico), rather than the introduced range (China). In other words, the Asteraceae from Mexico might grow more slowly in their native range. Also, datasets that include sister taxa would help to distinguish between environmental filtering and evolution and are still needed. Next, we identify possible study design considerations that take advantage of the power of evolutionary models (Cadotte et al., 2018).

IV. The way forward

We suggest several features for consideration in future study designs (see also Cadotte et al., 2018). We especially focus on experimental designs, because previous experiments have suggested that invaders can both respond to and influence community phylogenetic composition (Bennett et al., 2014; Li et al., 2015). Thus, experiments are needed to avoid confounding cause and effect (Burns, 2014), because invaders can be both ‘drivers’ of community change as well as ‘passengers’ responding to community context (MacDougall & Turkington, 2005), limiting the power of observational tests. Here, we argue that experimental designs should: (1) consider the identification of taxon samples that lead to a continuous range of phylogenetic distances; (2) replicate across multiple invaders and within clades; (3) consider belowground functional trait diversity, as well as the more commonly measured aboveground functional traits; and (4) replicate across different environmental conditions.

First, some consideration should be given to measures of phylogenetic distance in initial study design. It should be noted that, in our case study (Fig. 1), we observe a substantial gap in phylogenetic distances, because this sample includes a large proportion of Asteraceae followed by more distant relatives, such as Malvaceae, Fabaceae and Poaceae (Fig. 2). The focal invader, Chromolaena odorata, is an Asteraceae, leading to a cluster of close phylogenetic distances to the invader (Figs 1, 2). When the primary consideration in study design is the co-occurring species in a community (such as, e.g. Zheng et al., 2018), such features of the data might be inevitable. However, an alternative goal might be to sample evenly across the phylogeny, creating a stronger test for phylogenetic and functional distance relationships (Fig. 3). For example, we propose a balanced phylogeny, sampling invaders from each clade (Fig. 3). A study might achieve such balance by sampling evenly from monocotyledons and dicotyledons, for example.

Second, replication at the levels of invader and clade would also create a stronger test of macroevolutionary hypotheses (Fig. 3). We present a case study with a single invader (Zheng et al., 2018); however, replicating across multiple invaders will be key to the identification of general trends. A possible experimental design could take advantage of multiple clades of close relatives (e.g. congeners) to ensure replication at various phylogenetic distances (Fig. 3) and sufficient sampling amongst close relatives (Cadotte et al., 2017). Further, this design leads to reduced nonindependence among internal phylogenetic branches. Nonindependence results from resampling the same branch many times, because deeper splits are more frequently sampled for unbalanced phylogenies (Cadotte et al., 2017). It should be noted that, in our example, we proposed a balanced sampling across distantly related clades (Fig. 3).

To illustrate replication across clades, imagine that invaders ‘d’, ‘f’ and ‘i’ are experimentally introduced to resident communities composed of close, intermediate or distant relatives (Fig. 3a). If a single functional trait determines invader success and evolves by BM (Cadotte et al., 2018), the variance in invader performance might be greater at greater phylogenetic distances (Table 1). By replicating across multiple invaders, this prediction about the variance in invader response could be tested (e.g. Fig. 3b). This design also minimizes problems with uneven sampling across phylogenetic distances, where sampling more across deep branches in the phylogeny leads to more data at greater phylogenetic distances, violating homoscedasticity assumptions (Cadotte et al., 2017). Three or more close relatives within clades would be ideal for this, and taxonomy could be used as a quick design tool; for example, one could sample groups of multiple congeners.

Third, future studies should consider the measurement of root and stem traits in addition to leaf traits (reviewed in Funk et al., 2017). Several recent studies have suggested that root traits sometimes describe an independent axis of trait differentiation, compared with commonly measured leaf traits, such as SLA (Fortunel et al., 2012; Medeiros et al., 2017). For example, Fortunel et al. (2012) found that root and leaf traits were orthogonal across 758 woody tropical species. This suggests that the measurement of leaf traits alone is insufficient to address the full range of functional trait variation in plants. Root traits might be particularly important for some hypotheses about biological invasions (Funk et al., 2017). For example, escape from competition with more distant relatives (Darwin's naturalization hypothesis) could include escape from belowground competition for soil resources. Thus, studies that consider both belowground and aboveground functional traits are needed, especially if multiple functional traits correlate with invasion (Table 1).

Fourth, environment often influences functional traits, and thus the relationship between functional distance and phylogenetic distance might be context dependent (Burns & Strauss, 2012). Environmental condition is an important driver on functional traits. The functional trait value partly reflects the adaptation of plants to the environment (McIntyre et al., 1995; Bernhardt-Römermann et al., 2008), and trait–environment correlations can also be the result of plastic responses to the environment rather than representing genotypic differences (Reich et al., 2003). Giordani et al. (2012) found that functional traits (growth form, reproductive strategy and photobiont type) of epiphytic lichens were strongly related to environmental conditions. Cornwell & Ackerly (2009) also found that functional trait values can be influenced by water, nitrogen and light conditions. Therefore, when testing the relationship between functional distance and phylogenetic distance, experimental designs should, at a minimum, control for the environment in which functional traits are measured (i.e. traits should be measured in a common garden). Additional design considerations might include multiple environmental treatments (e.g. drought, nitrogen, light). For example, we predict context dependence when competition governs invasion, because traits such as SLA are frequently correlated with growth rate and competitive ability and are also frequently plastic (Burns & Strauss, 2012).

Although these proposed features add complexity to the experimental design, some features of previous studies could be simplified, in the interest of focusing on macroevolutionary hypotheses. For example, compared with our study (Zheng et al., 2018), which sampled resident taxa from both the native and introduced ranges, one could design an alternative experiment with species from a single range. Our experiment (Zheng et al., 2018) was also conducted across a species richness gradient, but a future experiment could choose a single number of species for resident communities, further simplifying the experimental design (Fig. 3).

V. Conclusions

The consideration of evolutionary models in studies of biological invasion will lead to novel research questions and insights (Table 1; Gerhold et al., 2015), and is an important frontier in invasion biology (Cadotte et al., 2018). For example, patterns of divergence between close relatives suggest interesting future research directions (Zheng et al., 2018). Whether the communities have evolved in situ, or whether close relatives diverged in allopatry and assembled later, might have important implications for evolutionary and ecological interactions (Anacker & Strauss, 2014; Nuismer & Harmon, 2015). For example, divergence between close relatives could be shaped by a history of competition and divergent selection amongst co-occurring species (i.e. ‘ecological character displacement’, reviewed in Stuart & Losos, 2013). Alternatively, selection in allopatry might more closely resemble BM evolution. An explicitly evolutionary approach, including the testing of the predictions of alternative evolutionary models (Table 1), is an important step forward in this research program, and more experimental studies are especially needed that consider issues such as replication across clades, balanced phylogenetic sampling and multivariate trait evolution. This might allow us to move beyond a simplistic view of Darwin's naturalization conundrum as two opposing hypotheses, towards a more synthetic and mechanistic view of the role of macroevolution in contemporary biological invasions.

Author contributions

Y-LZ designed and conducted the experiment. JHB conducted the analyses. All authors contributed to the writing.