Volume 110, Issue 2 p. 443-456
RESEARCH ARTICLE
Free Access

The invasion paradox dissolves when using phylogenetic and temporal perspectives

Adrienne R. Ernst

Corresponding Author

Adrienne R. Ernst

Plant Biology and Conservation, Northwestern University, Evanston, IL, USA

Negaunee Institute for Plant Conservation Science and Action, Chicago Botanic Garden, Glencoe, IL, USA

Correspondence

Adrienne R. Ernst

Email: [email protected]

Search for more papers by this author
Rebecca S. Barak

Rebecca S. Barak

Negaunee Institute for Plant Conservation Science and Action, Chicago Botanic Garden, Glencoe, IL, USA

Search for more papers by this author
Andrew L. Hipp

Andrew L. Hipp

Center for Tree Science, The Morton Arboretum, Lisle, IL, USA

Search for more papers by this author
Andrea T. Kramer

Andrea T. Kramer

Negaunee Institute for Plant Conservation Science and Action, Chicago Botanic Garden, Glencoe, IL, USA

Search for more papers by this author
Hannah E. Marx

Hannah E. Marx

Deparment of Biology, University of New Mexico, Albuquerque, NM, USA

Search for more papers by this author
Daniel J. Larkin

Daniel J. Larkin

Department of Fisheries, Wildlife, and Conservation Biology, University of Minnesota, St. Paul, MN, USA

Search for more papers by this author
First published: 10 November 2021
Citations: 7

Handling Editor: Ayub Oduor

Abstract

  1. The prediction that higher biodiversity leads to denser niche packing and thus higher community resistance to invasion has long been studied, with species richness as the predominant measure of diversity. However, few studies have explored how phylogenetic and functional diversity, which should represent niche space more faithfully than taxonomic diversity, influence community invasibility, especially across longer time frames and over larger spatial extents.
  2. We used a 15-year, 150-site grassland dataset to assess relationships between invasive plant abundance and phylogenetic, functional and taxonomic diversity of recipient native plant communities. We analysed the dataset both pooled across all surveys and longitudinally, leveraging time-series data to compare observed patterns in invasion with those predicted by two community assembly processes: biotic resistance and competitive exclusion. We expected more phylogenetically and functionally diverse communities to exhibit greater resistance to invasion.
  3. With the pooled dataset, we found support for the long-standing observation that communities with more native species have lower abundance of invasive species, and a more novel finding that more phylogenetically diverse communities had higher abundance of invasive species. We found no influence of aggregate (multivariate) functional diversity on invasion, but assemblages with taller plants, lower variability in plant height and lower seed mass were less invaded. Viewed longitudinally, the phylogenetic diversity relationship was reversed: the most phylogenetically diverse communities were most resistant to invasion. This apparent discrepancy suggests invasion dynamics are influenced by both site attributes and biotic resistance and emphasizes the value in studying invasion across time.
  4. Synthesis. Our results provide insight into the nuances of the diversity–invasibility relationship: invasion dynamics differed for different dimensions of diversity and depending on whether the relationship was evaluated longitudinally. Our findings highlight the limitations of using single time-point ‘snapshots’ of community composition to infer invasion mechanisms.

1 INTRODUCTION

Invasive species threaten biodiversity, ecosystem health and restoration efforts (Simberloff et al., 2013; Vilà et al., 2011). Due to the severity of these threats, and the opportunities that invasions pose for addressing fundamental ecological questions, a rich subfield of ecology has developed around invasive species. Invasion biology owes much of its legacy to Charles Elton's pioneering work, The Ecology of Invasions by Animals and Plants (Elton, 1958). One of the most influential ideas Elton proposed is that invasive species will be less likely to establish in communities with more native species. This prediction—that diversity confers invasion resistance—has important management implications: if true, promoting native plant diversity, already a priority for conservation and restoration, will have ancillary benefits of reducing invasive plant dominance.

The relationship between plant community diversity and invasibility has been described as paradoxical: while some studies have shown that more diverse communities have fewer invaders (Beaury et al., 2020; Kennedy et al., 2002; Naeem et al., 2000), others have shown the exact opposite relationship, with more species-rich communities being more invaded (Davies et al., 2005; Peng et al., 2019; Smith & Côté, 2019; Stohlgren et al., 2003). Studies have drawn heavily on Elton's original idea to explain negative diversity–invasibility results (Richardson & Pyšek, 2008). Under the biotic resistance hypothesis, the number of native species correlates with the amount of niche space occupied, such that maximizing native richness minimizes niche space available for invasive species (Elton, 1958; Fridley et al., 2007).

In scenarios that do not support the biotic resistance hypothesis, environmental factors are commonly invoked to explain invasion patterns (Fridley et al., 2007; Levine & D'Antonio, 1999). Explanations for this discrepancy have been attributed to different processes operating between: (a) large and small spatial scales (Davies et al., 2005; Tomasetto et al., 2019), (b) experimental and observational studies (Levine & D’Antonio, 1999; Peng et al., 2019) and (c) abundance-based and presence/absence-based measures of invasion (Cleland et al., 2004; Smith & Côté, 2019). Generally, biotic resistance is better supported at smaller spatial scales, in experimental settings and in explaining invader abundance rather than richness. While recent advances have helped disentangle this invasion paradox, relatively few studies have moved beyond species richness as the key measure of native diversity (but see work related to Darwin's naturalization conundrum; e.g. Ma et al., 2016; Marx et al., 2016; Pinto-Ledezma et al., 2020). Over-reliance on richness is a potential limitation of diversity–invasibility studies evaluating biotic resistance, as the number of native species present in a community may be a poor proxy for available niche space.

Specifically, while species richness is a critical component of diversity, it captures neither the ecological differences between species nor the evolutionary dynamics—selection and drift—that shape those differences (Cadotte et al., 2013). Using metrics of diversity that potentially better reflect niche differences and/or their origins may better test the biotic resistance hypothesis. In this paper, we test two such metrics: functional and phylogenetic diversity of the native community. Functional diversity, as a measure of trait distribution, is thought to capture some aspects of the niche directly and can reflect the breadth of strategies used by the organisms that make up a community (Díaz & Cabido, 2001; McGill et al., 2006). However, the utility of functional diversity as a proxy for the niche hinges on selecting ecologically relevant traits, a notoriously difficult task (Funk et al., 2017; Lefcheck et al., 2015). By integrating over the evolutionary history of a species and the lineage from which it arises, phylogenetic diversity is hypothesized to capture further aspects of niche differentiation (Cavender-Bares et al., 2009; Webb et al., 2002). Phylogenetic diversity has been used as a proxy for functional trait diversity or as a way to capture the evolutionary history of traits that cannot be easily measured or are not currently recognized as influential for ecosystem processes (Larkin et al., 2015; Pearse & Hipp, 2009; Tucker et al., 2018). However, phylogenetic diversity will only reflect ecological differences between species at phylogenetic scales at which ecologically important traits are conserved or exhibit phylogenetic signal (Cavender-Bares et al., 2004; Mayfield & Levine, 2010; Tucker et al., 2018). Given the potential and limits of each of these components of diversity, it is important to consider multiple facets and evaluate the degree to which they provide similar versus distinct insights into community assembly generally, and invasion dynamics more specifically.

A shared limitation of both phylogenetic diversity and multivariate functional diversity measures is that summarizing multiple niche axes as a single number can mask important ecological dynamics, particularly opposing trends in niche axes (Roscher et al., 2012; Trisos et al., 2014). By breaking multivariate functional diversity into component traits, ecological trade-offs may become apparent (Pellissier et al., 2018; Spasojevic & Suding, 2012). Thus, the analyses of individual functional traits can elucidate the community assembly processes shaping invasion dynamics and allow identification of particularly influential traits (Butterfield & Suding, 2013; Carboni et al., 2016; Funk et al., 2017). Paired with phylogenetic diversity and functional diversity, single trait indices can help provide mechanistic insights into invasion.

Observational tests of the diversity–invasibility hypothesis are frequently limited by the availability of data at multiple time points. For example, some recent large-scale observational studies have found that communities with more native species had fewer non-native species, and concluded that biotic resistance drives this relationship (Beaury et al., 2020; Iannone et al., 2016). However, these studies lacked time-series data enabling changes in the relationship between native and non-native species richness to be evaluated over time, making it difficult to distinguish between two possible drivers: (a) increased native species exhibiting higher resistance to invasive species (biotic resistance), or (b) invasive species out-competing native species, resulting in fewer native species (competitive exclusion). A challenge in synthesizing the diversity–invasibility relationship is that comparing observational and experimental findings is complicated by the relative paucity of observational studies that include temporal trends (Gallien & Carboni, 2017).

We sought to evaluate whether niche space occupied by native communities would be represented more faithfully by metrics that incorporate species differences. To test how plant invasions are influenced by phylogenetic, functional and taxonomic diversity, we used a dataset encompassing 150 grassland sites located across a 10 million hectare area. The dataset includes comprehensive vegetation surveys with repeated surveys spanning a decade for most sites. These repeated surveys enabled us to track changes in diversity patterns over time and compare them to the patterns expected under different community assembly processes. With these data, we investigated whether: (a) more diverse communities were less invaded, (b) native communities exhibited biotic resistance and (c) invasive species competitively excluded native species.

2 MATERIALS AND METHODS

2.1 Dataset

Our dataset encompasses 150 grassland sites scattered across 10 million hectares in the state of Illinois, USA. The sites were surveyed under the Illinois Natural History Survey's Critical Trends Assessment Program (CTAP) between 1997 and 2016. The goal of the CTAP program is to evaluate trends in Illinois grasslands broadly. In order to assess changes in a representative set of grasslands, sites were randomly selected throughout the state and encompass a range of grassland subhabitats (Carroll et al., 2002). Within each site, twenty 0.25-m2 plots were placed at 2-m intervals along a 41-m transect. Plots were placed 1-m away from the transect on alternating sides. GPS coordinates and permanent metal markers were used to resample the same locations each year. Ground cover was estimated for each species rooted in each plot using a modified Daubenmire method. In some cases, plants could only be identified to genus. For more detailed methods, see Carroll et al. (2002). When possible, sites were resurveyed every 5 years. Seventy-three per cent of sites were surveyed at least twice, 59% were surveyed at least three times and 28% were surveyed four times. Any sites noted as having been managed with herbicide or weeding were excluded from our analysis for that year.

To prepare the dataset for analysis, we standardized species names using the Taxonomic Name Resolution Service (Boyle et al., 2013) and manually assigned species' origins (native or non-native) based on Taft et al. (1997) and invasive status following the Midwest Invasive Plant Network (MIPN, 2019).

2.2 Characterization of phylogenetic diversity

To construct a phylogeny for all 528 native species observed in the vegetation surveys, we followed the methods described in Barak et al. (2017). Briefly, we began with a published mega-phylogeny of 32,223 plants (Zanne et al., 2014). Using the weldTaxa and makeMat functions from the Morton R Project (https://github.com/andrew-hipp/morton), we grafted in taxa that were present in the survey data but absent from the tree (167 identified to species and 30 identified to genus) and pruned species that were part of the original tree but absent from the survey data. Missing species were grafted onto the base (crown) of the genus except for two species which lacked congeners in the Zanne et al. (2014) tree, we selected sister taxa for these two species. The missing species accounted for 19% of the total abundance across the dataset. The two species lacking congeners were uncommon, with one occurring in seven sites and the other occurring in only one plot in a single site. While our approach of adding missing species as genus-level polytomies reduced phylogenetic resolution, community phylogenetic metrics calculated from trees lacking even any resolution below the genus level have been shown to be strongly correlated with those calculated from species-level trees (Qian & Jin, 2020).

For each plot within each site and year, we calculated abundance-weighted mean nearest taxon distance (MNTD) and mean pairwise distance (MPD) to characterize phylogenetic diversity of the native community. MNTD is calculated as the mean phylogenetic distance between each species and its closest co-occurring relative and MPD is calculated as the mean phylogenetic distance between all pairs of co-occurring species (Webb, 2000). MNTD and MPD were selected because they quantify phylogenetic divergence, are hypothesized to capture niche differentiation and are commonly used in community phylogenetic analysis, facilitating comparison among studies (Tucker et al., 2017). As MPD is calculated by averaging relatedness between all co-occurring species, it spans the full depth of the community phylogeny, reflecting deeper divergences and older evolutionary relationships. In contrast, MNTD is calculated between closest relatives, reflecting more recent evolutionary relationships. Both MNTD and MPD require the presence of multiple co-occurring species; plots were excluded from analysis if they had fewer than three native species.

Mean nearest taxon distance can be sensitive to differences in species richness (Webb, 2000), confounding taxonomic and phylogenetic diversity. To remove the effect of species richness, we calculated standardized effect sizes of MNTD ([observed value – expected value]/standard deviation of the expected value). For comparability between phylogenetic diversity metrics, we also calculated standardized effect sizes for MPD. Expected values were calculated under an ‘independentswap’ null model with 999 permutations using the picante package v.1.8.2 (Kembel et al., 2010). This null model randomizes species co-occurrences while maintaining species richness and species occurrence frequency, and has been shown to detect niche-based assembly more reliably than other commonly used null models (Kembel, 2009). Standardized effect sizes of MNTD (SES MNTD) and MPD (SES MPD) >0 indicate that co-occurring species are more distantly related than expected by chance, while values <0 indicate closer relatedness than expected by chance. In other words, higher positive values of SES MNTD and SES MPD indicate higher phylogenetic diversity.

2.3 Characterization of functional diversity

To characterize the functional composition of the plots over time at each site, we sourced trait data from the TRY database (Kattge et al., 2020) and used the bien package v.1.2.4 (Maitner et al., 2018) to access the BIEN database (Enquist et al., 2016). Eleven candidate traits were selected based on their availability and demonstrated importance in functional ecology literature (Appendix S1). All available trait data for the native species found in vegetation surveys were downloaded from TRY and BIEN. Next, we developed a two-step procedure for trait selection. First, following Pakeman (2014), we narrowed the candidate traits to those that were available for >80% of the total abundance across all species at the site level. Second, we filtered candidate traits to minimize trait correlation while providing coverage across key axes of functional trait strategies—resource acquisition, competitive ability and reproductive strategy (Lefcheck et al., 2015). Multiple traits that met the coverage thresholds were related to the leaf economics spectrum, which are often highly correlated and potentially functionally redundant (Laughlin, 2014). From these leaf economics spectrum traits, we selected SLA, as it had the highest coverage and is well-supported as a proxy for resource acquisition (Violle et al., 2009; Wright et al., 2004). Second, we selected vegetative height, which is associated with competitive ability (Fréville et al., 2007; Leach & Givnish, 1996). The third trait selected was seed mass, which is linked to reproductive strategy (Grime & Jeffrey, 1965; Turnbull et al., 1999). To characterize the functional composition of the resident native community in each plot, we calculated both multivariate functional diversity and single functional trait indices. For multivariate functional diversity, we used functional dispersion (FDis) because it reflects functional divergence—arguably the most important aspect of functional diversity under the biotic resistance hypothesis (Laliberte & Legendre, 2010). FDis is the weighted mean distance in multidimensional trait space of individual species to the abundance-weighted centroid of all species. We calculated FDis using the FD package v.1.0.12 (Laliberte et al., 2014).

2.4 Single functional trait metrics

We also calculated three indices for individual traits in the native community: (a) weighted mean absolute deviation (MAD), the sum of the relative abundance-weighted difference between each species and the average trait value for the assemblage—a single trait analogue of FDis (Laliberte & Legendre, 2010); (b) range, the difference between the maximum mean trait value and the minimum mean trait value in a community; and (c) community-weighted mean (CWM), the abundance-weighted mean trait value for a community (Violle et al., 2007). We calculated these indices using standardized mean trait values (each trait was scaled to mean = 0, standard deviation = 1) to facilitate comparison between traits and for comparability with multivariate functional diversity.

2.5 Statistical analysis

Our overall approach was to analyse how invasive species cover responded to the characteristics of native communities. As our analyses occurred at the plot level, we did not analyse invasive species richness—most plots had few invasive species, making invader richness approximate a binary response. We implemented models to address the following questions: (a) How does native community diversity correlate with invasive species abundance when viewed without a temporal perspective? (pooled model); (b) Do native communities exhibit biotic resistance to invasion? (longitudinal biotic resistance model); and (c) Do invasive species competitively exclude (displace) native species following invasion? (longitudinal competitive exclusion model). For each of these models, we tested the following 11 native community attributes as fixed predictors: SES MNTD, SES MPD, FDis, species richness, CWM seed mass, CWM height, seed mass range, height range, MAD SLA, MAD seed mass and MAD height. We additionally considered CWM and range for SLA; however, they were eliminated because the three SLA metrics were highly correlated (Pearson's correlation coefficient >0.7). We tested for correlation among all of the fixed predictors considered using the corrplot package v. 0.84 (Wei & Simko, 2017) and no other pairs of predictors had correlations >0.7 (Figure S1). For model selection, we implemented both forward and backward selection based on log-likelihood ratio tests following Zuur et al. (2009). Analyses were carried out in R version 3.6.1 (R Core Team, 2019). Models were implemented using the glmmTMB v.1.0.2.1 package (Brooks et al., 2017) and visualized using the ggplot2 package v.3.3.3 (Wickham, 2016).

2.6 Pooled model: Does native diversity correlate with invader abundance?

For the first model, we analysed how invasive species cover was related to the diversity of native communities in all plots for all years, regardless of when the data were collected (invasive species abundance ~ native diversity). The level of invasion was calculated as the summed cover of all invasive species in each plot. We estimated generalized linear mixed effects models (GLMMs) with a negative binomial distribution to account for overdispersion. Fixed predictors considered were the 11 native community attributes described above in addition to year. We included year as a fixed effect to test for directional changes over time. To control for the spatially nested structure of the data and repeated sampling of the same plots, both site and plot within site were included as random effects.

2.7 Longitudinal biotic resistance model: Are more diverse communities less likely to become invaded?

To disentangle different mechanisms by which native and invasive species interact, we examined invasion over time following Muthukrishnan et al. (2018; invasive abundance at subsequent visit ~ initial native diversity prior to invasion). To test if native diversity confers resistance to invasion, we compared the native diversity metrics at each plot's initial survey to the invader abundance at the subsequent survey. For this model, we excluded all plots with invasive species present at the first survey to remove the effects of established invaders, for example, an ‘invasional meltdown’ (Simberloff & Von Holle, 1999) wherein the presence of one invasive species facilitates further invasion. We also excluded plots that were only sampled once and had no subsequent measure of invasive species abundance. As there were relatively few plots that fit these criteria and had more than two visits, we focused our analysis on two time points, assessing the relationship between diversity during the first visit and abundance of invasive species during the second visit. We used linear mixed effect models for this analysis, the fixed predictors considered were the same suite of native community attributes as above and site was the sole random effect. As the model did not include more than one subsequent visit per plot, it was not necessary to include plot as a random effect.

2.8 Longitudinal competitive exclusion models: Do invasive species out-compete native species?

We investigated whether invasive species competitively excluded native species, and how the characteristics of the native community shifted following the invasion (change in diversity ~ invasive species abundance). We used linear regressions to model the change in each community attribute over time in those plots that had at least two surveys. We estimated separate linear regressions for each community attribute, where the response was the community attribute, and the explanatory variable was year (community attribute ~ year). The rates of change reflect the average annual change in each native community attribute and were calculated using the model coefficients. The rate of change in each native community attribute was then compared with invasive species cover. We compared plots that were uninvaded throughout all surveys to those that were invaded.

3 RESULTS

3.1 Pooled model

The pooled model was a ‘snapshot’ that estimated the relationship between native diversity and invasive cover at all plots across all sites and all years. It included 1,606 unique plots from 125 sites with one to four surveys each, of which 737 plots from 69 sites included repeated visits; pooled across surveys, it encompassed 2,826 plot-years. The average numbers of native and invasive species per plot per visit were 5.9 ± 3.0 (S.D.) and 1.9 ± 1.4 respectively. Our final model included phylogenetic diversity (SES MNTD), species richness and individual traits (CWM height, CWM seed mass and height range) as predictors of invasive species abundance (Figure 1).

Details are in the caption following the image
Partial effects of all native community predictors included in the final pooled model of invasive species abundance (per cent cover). Each plot represents the predicted partial effect of the predictor when all other predictors are held constant. The lines and ribbons indicate the mean and 95% confidence intervals predicted by a generalized linear mixed effects model respectively

We observed a negative diversity–invasibility relationship when native diversity was measured as species richness (χ2 = 32.415, p < 0.001), indicating that plots with more native species were less likely to be invaded. However, we found a positive relationship between native phylogenetic diversity and invasive species abundance (χ2 = 7.194, p < 0.001); on average, the least-invaded plots had native plant species that were more closely related to each other. We found a negative relationship between native CWM height and invader abundance (χ2 = 4.334, p = 0.037) and positive relationships between both native height range (χ2 = 8.141, p = 0.004) and native CWM seed mass (χ2 = 5.897, p = 0.015) and abundance of invasive species, that is, less-invaded plots had native plant assemblages that had lighter seeds, were taller and had lower variability in height.

3.2 Longitudinal biotic resistance model

Our analyses supported native phylogenetic diversity (SES MPD) conferring biotic resistance against invaders (χ2 = 4.66, p = 0.031; Figure 2). That is, plots with higher native SES MPD had lower invasive species abundance at the next survey. The biotic resistance model only included plots that were uninvaded in the first survey, comprising 329 plots from 52 sites. There was no relationship between initial native diversity and subsequent invasive species abundance for other native diversity metrics.

Details are in the caption following the image
Relationship between native phylogenetic diversity and invasive species cover. Values to the left of the red dashed line indicate native communities with lower MPD than expected by chance (phylogenetically clustered), while values to the right indicate native communities with higher than expected MPD (phylogenetically overdispersed). The black line and grey ribbon indicate the mean and 95% confidence intervals predicted by a generalized linear mixed effects model respectively

3.3 Longitudinal competitive exclusion models

We did not observe any significant relationships between invasive species abundance and change in native diversity over time. For several attributes of the native community, change over time varied widely in plots that were never invaded (Figure 3). In invaded plots, diversity increased over time to a similar extent as the uninvaded plots (Figure S2). For phylogenetic diversity, 50.7% of invaded plots increased in diversity over time, compared with 47.6% of uninvaded plots. For functional diversity, 51.6% of the invaded plots became more functionally diverse over time, as did 52.3% of the uninvaded plots. Native species richness increased over time in 63.9% of invaded plots and 62.2% of uninvaded plots.

Details are in the caption following the image
Relationship between change in native community attributes and invasive species abundance. These three attributes of the native community, native species richness (a), phylogenetic diversity (b) and height MAD (c), demonstrate high variability in uninvaded plots. Green dots indicate plots that were uninvaded and remained uninvaded throughout all surveys. Purple dots indicate plots that had become invaded. The y-axis indicates the observed change in each native community attribute at each plot over the surveys

4 DISCUSSION

Our results show that, while phylogenetic diversity and functional composition of native species were linked to invasive species abundance, the dynamics were more complicated than theory suggests. Based on the fine spatial resolution of our dataset (0.25-m2 plots) and our use of an invasion metric accounting for abundance, we anticipated a negative diversity–invasibility relationship. We tested for effects of both phylogenetic and functional diversity, hypothesizing that they would have similar relationships with invasibility, and potentially stronger effects than species richness given their greater potential to reflect species' niches. Our results are only partially consistent with this hypothesis. We did find that communities with higher initial phylogenetic diversity (SES MPD) had lower abundances of invasive species during subsequent surveys. This is a phylogenetic analogue to the long-standing observation that, at fine spatial resolution, communities with more native species are less susceptible to dominance by invasive species (Kennedy et al., 2002; Naeem et al., 2000). However, when pooled across surveys, rather than using a time series, communities with higher phylogenetic diversity (SES MNTD) had higher levels of invasion, suggesting that sites that support high diversity also support high invasion. This pattern has often been reported in observational studies, which, like our pooled model, typically lack temporally explicit data (Marcantonio et al., 2014; Stohlgren et al., 2003). We further found that two traits (native community plant height and seed weight) predicted abundance of invasive species, such that communities with taller plants and lighter seeds had lower cover of invasive species. However, we found no evidence relating multivariate functional diversity or SLA to invasive species abundance.

4.1 Phylogenetic pattern of biotic resistance

In our biotic resistance model, the relationship between phylogenetic diversity and invasion aligned with our expectations: plots with higher initial phylogenetic diversity had lower abundances of invasive species in subsequent surveys. To our knowledge, few studies have demonstrated that native phylogenetic diversity confers invasion resistance as measured by decreased abundance of invasive species over time (Galland et al., 2019; Qin et al., 2020). And our study is the first to demonstrate this trend outside the experimental communities through a large-scale observational study. Although previous studies have found that more phylogenetically diverse communities are less invaded, they either did not account for change over time or only measured invasion as presence/absence (Gerhold et al., 2011; Iannone et al., 2016; Loiola et al., 2018; Lososová et al., 2015; Whitfeld et al., 2014; Yessoufou et al., 2019). As native species richness did not confer invasion resistance in our study, this supports the idea that phylogenetic diversity may be a more faithful representation of species' niches than species richness.

There have also been few studies testing whether phylogenetic diversity affects the likelihood (rather than relative abundances) of invasions over time. Those that have been performed have varied in their findings. An experimental study demonstrated that phylogenetic diversity conferred biotic resistance within wetland microcosms and was a better predictor of invasion resistance than species richness (Qin et al., 2020). Another experimental study found that phylogenetic diversity decreased colonization by unplanted species (Galland et al., 2019). One study found a negative relationship between phylogenetic diversity and invasive species reproduction, but no effects on survival or biomass (Feng et al., 2019). Other studies have shown a positive effect of phylogenetic diversity on experimental invaders (El-Barougy et al., 2020), and that invasive species drove such a relationship by decreasing native phylogenetic diversity (Bennett et al., 2014). Within our dataset, we found that the relationship between phylogenetic diversity and invasion changed direction when the data were pooled but change over time was not explicitly considered.

4.2 Biotic interactions key to invasion, but no clear community assembly process

While we found partial evidence that biotic resistance mitigates invasion in these grassland communities, we found several lines of evidence that could be consistent with biotic resistance or competitive exclusion and were not able to disentangle the two. In our pooled model, plots with the most native species had the lowest invasive species abundance. In the context of Elton's invasibility hypothesis, some recent observational studies concluded that more species-rich communities confer biotic resistance based on a snapshot of invasion (Beaury et al., 2020; Iannone et al., 2016). Other non-longitudinal observational studies have concluded that the negative relationship between species richness and invasion is due to invasive species driving decline in native species via competitive exclusion (Hejda et al., 2009; Michelan et al., 2010). The differences in these studies' conclusions are difficult to resolve, as observational studies that do not examine changes over time cannot differentiate causation from correlation, or the relative roles of biotic resistance by native species versus competitive exclusion by invasive species (Warren et al., 2017). In our study, the effect of native species diversity was demonstrated only by measuring its effect on changes over time in abundance of invasive species. Thus, our results highlight the importance of longitudinal data and the limits of using snapshots to infer invasion dynamics.

Communities with taller native species underwent less invasion, which could be consistent with either competitive exclusion or biotic resistance. Height is generally thought to confer a competitive advantage because it directly correlates with a species' ability to intercept light (Keddy & Shipley, 1989). As height was not retained in the biotic resistance or competitive exclusion model, we cannot say whether taller native communities conferred invasion resistance or if invasive species drove the outcome by displacing shorter native species. Previous work supports the idea that taller invasive species are more successful and have a greater negative impact on native communities (Divíšek et al., 2018; Hejda & de Bello, 2013).

4.3 No evidence for competitive exclusion

We found no direct evidence that competitive exclusion by invasive species decreased native diversity. Competitive exclusion can be a slow-acting process, so the time frame we studied (up to 15 years) may not have been sufficient to observe this phenomenon (Yackulic, 2017). There was also high variability in diversity metrics in the uninvaded sites, which challenged the detection of differences within the invaded sites. Furthermore, we had reduced statistical power following the removal of sites that did not meet our analytical criteria. In other systems, competitive exclusion has been shown to shape the invasion process (Jucker et al., 2013; Muthukrishnan et al., 2018). Based on our analyses, it appeared that biotic resistance played a stronger role than competitive exclusion. However, we did not find significant effects of species richness on biotic resistance despite finding a negative relationship between species richness and invasion in the pooled model. Biotic resistance and competitive exclusion are the primary community assembly processes thought to drive negative diversity–invasion relationships. Given that we did not find evidence consistent with either process, it is likely that our models did not identify signals of all major community assembly processes.

4.4 Linking individual traits to invader abundance

Some of our findings could be consistent with biotic and/or environmental processes, which we were unable to distinguish due to limitations of our data. For example, in the pooled model, we found that plots with greater height variability among native species had higher abundance of invasive species. If small differences in height range were due to native communities with consistently taller species, this could reflect a competitive advantage of tall-statured species impeding invader success (Fréville et al., 2007; Leach & Givnish, 1996). However, this is unlikely in our study as there was little correlation between range in plant height and mean plant height (r = 0.13). These results could also be explained by disturbance, which drives trait divergence and can promote invasive species (Grime, 2006; Hierro et al., 2006; Jauni et al., 2015).

Native communities with heavier seeds underwent greater invasion. Seed mass is usually linked with reproductive output (Leishman & Murray, 2001; Moles, 2018) through a trade-off between seed size and number of seeds produced for a given amount of energy—though this does not necessarily reflect differences in lifetime fitness (Moles et al., 2004; Moles & Westoby, 2006). Large seeds are more likely to survive, especially in stressful conditions including high competition and low light, but reproduction of large-seeded species may be limited by seed numbers (Catford et al., 2019; Grime & Jeffrey, 1965; Turnbull et al., 1999). In contrast, smaller seeded species are generally more prolific, have increased dispersal, remain viable in the seed bank longer and grow out of their vulnerable juvenile stage faster (Moles & Westoby, 2006; Rejmánek & Richardson, 1996). Additionally, multiple studies have shown that native communities with heavier seeds are more heavily invaded (Carboni et al., 2016; Catford et al., 2019; Fried et al., 2019). While the exact mechanism remains unclear, this pattern has been attributed to hierarchical differences in competitive ability driven by seed size, increased ability to withstand invasion pressure at the seedling stage and increased microsite availability for smaller seeded invasive species.

4.5 The rich get richer?

When we analysed the data pooled across surveys, more phylogenetically diverse assemblages were more invaded. This trend persisted when we re-implemented the pooled model restricted to plots included in the longitudinal biotic resistance model. Positive relationships between diversity and invasibility are usually found when the spatial resolution of a study is large (Cleland et al., 2004; Levine & D'Antonio, 1999). This is generally attributed to greater environmental heterogeneity or other factors that covary with spatial scale and increase diversity in general, that is, across native and invasive species alike (Davies et al., 2005; Fridley et al., 2007). While it remains somewhat unclear what drove positive diversity–invasibility relationships at neighbourhood scales in our study, it is likely that aspects of the sites or their management facilitated both high phylogenetic diversity and high invasion abundance. For instance, moderate grazing has been shown to increase native diversity by suppressing dominant species (Hallett et al., 2017), but can also increase opportunities for establishment and seed dispersal by invasive species (Jauni et al., 2015).

The unexpected positive relationship between phylogenetic diversity and invasion in the pooled model casts doubt on the underlying assumptions of our hypotheses, specifically that more closely related species are more ecologically similar and biotic interactions are the primary drivers of community assembly at small scales. Indeed, there are reasons to expect inconsistency in these patterns (Mayfield & Levine, 2010). For example, these results could be explained by trait diversity being concentrated within a small number of families. Most of the dominant species in North American grasslands represent three families (Asteraceae, Fabaceae and Poaceae; Towne, 2002), and previous work has shown that communities assembled from few lineages are more likely to have overdispersed traits (Prinzing et al., 2008). It is also possible that assemblages composed of distantly related species have lower niche overlap/packing and thus more empty or underutilized niche space for invaders to exploit. A recent experimental study found that colonization of more functionally diverse communities was greater and came to a similar conclusion: that higher functional diversity may increase unsaturated niche space (Galland et al., 2019). Likewise, biotic resistance is not always the primary factor determining invasive species establishment, even at the local scale (El-Barougy et al., 2020; Gallien & Carboni, 2017). Other factors such as disturbance (e.g. human habitat modification; Tomasetto et al., 2013), changes in resource availability and environmental stress can facilitate invasion and alter diversity–invasibility relationships (Clark & Johnston, 2011; Davis & Pelsor, 2001; Fridley et al., 2007).

While other observational studies have found higher native diversity in invaded communities, they are similarly limited in their ability to determine the processes driving this pattern (Loiola et al., 2018; Marcantonio et al., 2014). Previous observational studies have frequently treated invasion as binary (presence/absence) rather than comparing differences in invader abundance or species dominance (see Tomasetto et al., 2013). Such differences in invasion metrics have been shown to change the direction of the diversity–invasibility relationship (Cleland et al., 2004; Smith & Côté, 2019).

The contrasting effects of phylogenetic diversity on invasion in our pooled and biotic resistance models were unexpected. A possible explanation is that phylogenetic diversity could have different effects at different stages of invasion, consistent with the ecological strategies that make for a successful invasive species being fluid across invasion stages (Gallien & Carboni, 2017; Ma et al., 2016). It could be that the functional traits of the recipient community that were key to resisting early stage invasions were more phylogenetically conserved than traits that confer resistance at later invasion stages. Most plots in our sites had already been invaded and invasive species may facilitate one another via invasional meltdown (Simberloff & Von Holle, 1999). The effects of invasional meltdown were controlled for in the biotic resistance model by only including plots with no initial invasion, but this was not the case for the pooled model. Additionally, the lack of invasion in the previously uninvaded plots that we restricted our analysis to may indicate something unique about them. That is, they may not constitute a representative sample, with potentially important underlying differences between plots (beyond native species composition) that differentiated whether they were invaded or uninvaded in the first place.

Differences in phylogenetic diversity metrics may also explain the opposing effects we found in our pooled and biotic resistance models. The phylogenetic diversity metric selected in the biotic resistance model reflects deeper relationships within the tree (MPD), whereas the metric selected in the pooled model (MNTD) reflects more recent divergences. These differences in phylogenetic depth make them sensitive to different processes (Mazel et al., 2016). Studies on relatedness between invader and recipient communities have found that invaders tend to occur in communities when they have at least one close relative (measured via shallow metrics like MNTD), but are more distantly related to the community as a whole (measured via deep metrics like MPD; Marx et al., 2016; Qian & Sandel, 2017). It could be that deeper divergences better reflect aspects of niche differentiation important to invasive species establishment.

4.6 The absence of a relationship between multivariate functional diversity and invasion

We expected native assemblages with higher multivariate functional diversity (FDis) to be less invadable, but it is not entirely surprising that this was not the case. Functional diversity is only meaningful if the traits included are linked to ecological processes relevant to invasion (Funk et al., 2017; Lefcheck et al., 2015). The traits included in this study may not adequately capture key niche axes involved in the invasion process; notably, we had no below-ground traits, which are important in competition (Broadbent et al., 2018; Foxx & Fort, 2019). We were also unable to obtain functional traits for all species in the dataset. While we attempted to minimize the effects of missing trait data, the lack of complete trait coverage could have biased our estimates of functional trait diversity (Májeková et al., 2016; Pakeman, 2014).

Previous observational studies have found that uninvaded native communities have higher functional diversity than invaded communities, which was interpreted as evidence of invasive species decreasing functional diversity through competitive displacement (Jucker et al., 2013; Michelan et al., 2010). Of the few invasion studies to manipulate functional diversity experimentally, two found that invader performance decreased in more functionally diverse native communities, providing support for the biotic resistance hypothesis (Byun et al., 2020; Feng et al., 2019). However, another experiment found that colonizers were more successful in more functionally diverse plots and attributed this to unoccupied niche space increasing with functional diversity (Galland et al., 2019).

5 CONCLUSIONS

This study demonstrates that our understanding of invasion dynamics is highly sensitive to treatment of time and selection of diversity metrics. Without the inclusion of phylogenetic diversity, our findings would primarily confirm the well-documented phenomenon that, at a given point in time, more species-rich communities are less invaded. The patterns we observed with phylogenetic diversity illustrate that temporal context is critical to elucidating invasion dynamics. If only snapshots in time were considered, the positive correlation between native diversity and abundance of invasive species would suggest that site attributes drove both. In contrast, based on time-series data, it appears that higher diversity enabled native communities to resist invasion. Both may be true. Thus the ‘invasion paradox’ dissolves into a discrepancy in perspectives. And it suggests that managing grasslands for invasion resistance will require attention to both site conditions and native community diversity. This conclusion, which will align with the field observations of many land managers and community ecologists, is drawn from our synthesis of pooled and longitudinal studies. Diversity–invasibility relationships are likely too complex to study from one standpoint alone.

ACKNOWLEDGEMENTS

The authors thank the Illinois Department of Natural Resources for funding the CTAP program, the botanists of the Illinois Natural History Survey who collected the data and David Zaya for his assistance in accessing and interpreting the CTAP dataset. This material is based on the work supported by the National Science Foundation Graduate Research Fellowship under Grant No. DGE-1842165. This manuscript was greatly improved with the input of three anonymous reviewers.

    CONFLICT OF INTEREST

    The authors declare no conflict of interest.

    AUTHORS' CONTRIBUTIONS

    A.R.E. and D.J.L. conceived the ideas and the methodology; A.R.E. analysed the data with input from D.J.L., A.T.K. and A.L.H.; A.R.E. drafted the manuscript; A.R.E., R.S.B., A.L.H., A.T.K., H.E.M. and D.J.L. all contributed critically to the draft and gave final approval for submission. Except for first and last authors, authors are listed alphabetically, not in the order of contribution.

    PEER REVIEW

    The peer review history for this article is available at https://publons.com/publon/10.1111/1365-2745.13812.

    DATA AVAILABILITY STATEMENT

    Data available from Dryad Digital Repository https://doi.org/10.5061/dryad.qz612jmgp (Ernst et al., 2021). Trait data can be accessed through the TRY database: https://www.try-db.org (Kattge et al., 2020) and BIEN database: https://bien.nceas.ucsb.edu/bien/ (Enquist et al., 2016).