Plant invasion alters trait composition and diversity across habitats

Abstract Increased globalization has accelerated the movement of species around the world. Many of these nonnative species have the potential to profoundly alter ecosystems. The mechanisms underpinning this impact are often poorly understood, and traits are often overlooked when trying to understand and predict the impacts of species invasions on communities. We conducted an observational field experiment in Canada's first National Urban Park, where we collected trait data for seven different functional traits (height, stem width, specific leaf area, leaf percent nitrogen, and leaf percent carbon) across an abundance gradient of the invasive Vincetoxicum rossicum in open meadow and understory habitats. We assessed invasion impacts on communities, and associated mechanisms, by examining three complementary functional trait measures: community‐weighted mean, range of trait values, and species’ distances to the invader in trait space. We found that V. rossicum invasion significantly altered the functional structure of herbaceous plant communities. In both habitats V. rossicum changed the community‐weighted means, causing invaded communities to become increasingly similar in their functional structure. In addition, V. rossicum also reduced the trait ranges for a majority of traits indicating that species are being deterministically excluded in invaded communities. Further, we observed different trends in the meadow and understory habitats: In the understory, resident species that were more similar to V. rossicum in multivariate trait space were excluded more, however this was not the case in the meadow habitat. This suggests that V. rossicum alters communities uniquely in each habitat, in part by creating a filter in which only certain resident species are able to persist. This filtering process causes a nonrandom reduction in species' abundances, which in turn would be expected to alter how the invaded ecosystems function. Using trait‐based frameworks leads to better understanding and prediction of invasion impacts. This novel framework can also be used in restoration practices to understand how invasion impacts communities and to reassemble communities after invasive species management.


| INTRODUC TI ON
Many of the world's most important and valuable ecosystems have been drastically altered by invasive species (Simberloff & Rejmanek, 2011). To understand the ecological dynamics associated with these alterations, it is vital to examine both the differences between the dominant invaders and resident communities and the impacts that invaders have on them (Cadotte, Campbell, Li, Sodhi, & Mandrak, 2018;MacDougall, Gilbert, & Levine, 2009). Trait-based analyses can provide insights into both of these aspects (Drenovsky et al., 2012).
Trait-based analyses are increasingly being used in invasion studies because they can illustrate how invasive and native species differ in fitness and niche requirements (Funk, Standish, Stock, & Valladares, 2016). Invasive species are thought to be successful in their introduced environments because they occupy novel or empty ecological niches, and/or they possess fitness differences that drive competitive dominance over resident communities (MacDougall et al., 2009). However, measuring both niche and fitness differences is notoriously difficult, especially when considering interactions with many species. Thus, we require surrogate measures for species' niche and fitness differences, and functional traits provide this opportunity (Cadotte, 2017;Kraft, Godoy, & Levine, 2015;Laughlin, 2014). Violle and Jiang (2009) argue that the differences between species functional traits (i.e., morphological, physiological, biological characteristics) can be used to understand niche differences. A wellstudied example of this assumption is exhibited by the invasive plant Centaurea solstitialis, in California grasslands, where its deep root system allows it to access water deep in the soil that other species in the recipient community cannot (Hierro, Maron, & Callaway, 2005).
Yet, the flipside of the empty niche hypothesis is successful invaders will impact those resident species with similar resource requirements ("niche overlap hypothesis," MacDougall et al., 2009;Gallen & Carboni, 2017). In this case, as the invader increases in abundance, we should observe decreasing abundance of species with similar traits, eventually resulting in a complete exclusion of similar species, while dissimilar species appear less impacted (MacDougall et al., 2009). In addition, invasive species have been found to cause the functional homogenization of communities across the landscape, ostensibly because they reduce diversity and eliminate certain species nonrandomly (Qian & Guo, 2010;Villéger, Grenouillet, & Brosse, 2014). These nonrandom changes in community functional diversity would be expected to alter how ecosystems function (Cadotte et al., 2018). Therefore, by examining both the functional traits of species and the manner in which functional diversity is altered during the process of invasion we gain insights not only into the possible mechanisms governing the success of invasive species, but also into the impacts of invasion. However, no single functional diversity F I G U R E 1 Proposed effect of an invader on resident species using three trait measures: community-weighted means (CWM), range of trait values (RTV), and distance to invader (DTI), on resident species occurrence and relative abundance. Invaded community 1: The invader has high trait overlap with resident species. RTV of the entire community remains constant, with CWMs shifting only moderately toward the invader's own trait values, while species with low DTI are excluded. Invaded community 2: The invader occupies new niche space. RTV of the entire community increases, CWM shifts toward the trait values of the invader, and very few species have low DTI and are thus impacted through niche overlap. Invaded community 3: The invader occupies new niche space. RTV of the entire community decreases as certain species are excluded, CWM shifts toward the trait values of the invader, and very few species have low DTI and are thus impacted through niche overlap. The resulting shifts of CWM and RTV along an invasion gradient are shown on the right using height as an example for a trait measure captures all the relevant information to assess mechanisms for invader success and impact (e.g., Pavoine, Bonsall, Dupaix, Jacob, & Ricotta, 2017). Therefore, complementary functional diversity measures are required to fully assess invader success and impact.
Here we compare three measures along an axis of increasing invader abundance, each covering specific aspects of community functional structure, relative to the invader ( Figure 1): Community-weighted mean (CWM), range of trait values (RTV), and species' distances to the invader in trait space (DTI).
First, CWM values provide information about the contribution of dominant species to ecological processes and ecosystem function (Cadotte, 2017;Grime, 1998;Muscarella & Uriarte, 2016). Hence, on the one hand, if invaded communities show large shifts in CWM values with increasing invader abundance (converging toward the traits of the invader), this likely indicates that the invader is substantially different than the dominant species of the un-invaded community and that the invader has occupied previously unused niche space (e.g., Invaded Community 2, in Figure 1). But, on the other hand, shifts in CWM values can also result from invasion-driven changes in the abundances of the resident species Table S3. For example, observing an increasing CWM value for plant height across a gradient of invasion can indicate that co-occurring species in invaded communities tend to be taller because shorter species were extirpated through shading from the competitive invader (e.g., Invaded Community 3, in Figure 1).
By examining how the range in trait values (RTV) varies across an invasion gradient, we gain insight into the influence of the invader on the total community trait space (Ordonez, Wright, & Olff, 2010).
An increase in the overall RTV for a specific trait in a community following the arrival of the invasive species suggests that the invader has occupied novel trait space, lending support to the "empty niche hypothesis" (Elton, 1958;e.g., Invaded Community 2, in Figure 1). On the other hand, a decrease in RTV with increasing invasion would instead point to the creation of a strong selective biotic filter that reduces the diversity of trait values (e.g., Invaded Community 3, in Figure 1). For example, a dominant plant invader that lowers and homogenizes light availability would eliminate traits associated with shade intolerant species. This decrease in RTV is consistent with an invader acting as a selective filter that might result in community shifts not predicted by the niche overlap hypothesis.
Finally, the functional structure of an invaded community might be altered by invasion even in the absence of strong CWM or RTV shifts (e.g., Invaded Community 1, in Figure 1). This might be the case if community impact is driven mainly by niche overlap between the invader and the resident community, rather than by the introduction of novel traits (e.g., Invaded Community 2, in Figure 1) or by the creation of a selective filter (e.g., Invaded Community 3, in Figure 1). By assessing the effect of DTI on the abundance of individual species, we can test whether invasion results in the specific exclusion of species that share similar functional characteristics to the invader, and seemingly greater niche overlap. If the dominant invader is competitively superior, then we should expect that low DTI species should be most adversely affected by increasing invader abundance (MacDougall et al., 2009). However, if the invader modifies the environment in such a way that filters against the traits of certain species, then perhaps there will be no strong relationship with DTI.
We apply this framework to study how the invasive Eurasian vine Vincetoxicum rossicum (locally known as "dog-strangling vine") affects herbaceous communities in Rouge National Urban Park in Toronto, Canada. V. rossicum has been spreading through the Park for approximately 60 years (Moore, 1959) and is now the most dominant herbaceous plant in the Park (Livingstone, 2018). Due to its prolific rate of spread and the fact that it forms dense stands, it poses a significant threat to native biodiversity (DiTommaso, Lawlor, & Darbyshire, 2005). In addition, V. rossicum occurs across a number of different habitat types and at different densities in the Park. It is therefore a perfect model system to investigate how invasion alters the functional structure of plant communities in different habitats.
Specifically, here we use seven plant functional traits to examine the mechanisms driving V. rossicum invasion and impact in two habitat types: open meadows and forest understory.
We predict that V. rossicum will alter invaded communities, both by dominating the functional structure of the community and by se- Based on this, we expect that increasing V. rossicum abundance will be positively correlated with CWM values for height, stem width, SLA, LNC, and LCC, and negatively correlated with number of leaves and LDMC. Secondly, we predict that increasing abundance of V. rossicum will lead to a decrease in the RTV in the community, which would indicate that the invader nonrandomly excludes certain resident species by altering local environmental conditions. However, there can be CWM shifts without increases to RTV if the invader replaces species with similar traits (e.g., Invaded Community 1, in Figure 1), or it occupies unique space while excluding dissimilar species (e.g., Invaded Community 3, in Figure 1). In order to tease apart this effect, we contrasted two RTV measures for invaded communities; one including V. rossicum's values and the other excluding them.
Finally, if V. rossicum competes most with similar species, we expect that resident species that are further away from the invader in functional trait space (greater DTI values) will be unaffected in invaded communities, while resident species with traits closer to the invader will be more likely to be outcompeted.

| Site
The observational field study was conducted in Rouge National Urban Park, located in Toronto, Ontario, Canada's largest city. This study was conducted in the summer of 2016, across 23 sites in two distinct habitats: meadow (open, sunny areas) and forest understory (shaded areas). We set 13, 50 by 50 m sites in meadow habitat and 10, 30 by 30 m sites in understory habitat. Each site was stratified into an equidistant grid of 25 plots, totalling 575 plots for the full study. Sites were chosen based on a varying degree of V. rossicum abundance (i.e., our invasion gradient). To quantify species abundances, two trained observers estimated the two-dimensional area occupied by each species' in each of the study plots to attain a value of percent cover. The meadow habitat included a total of 31 resident species for which we were able to obtain trait data, out of which 15 species are exotic. The understory habitat included a total of five resident species for which we were able to obtain trait data, out of which three species are exotic.

| Field sampling and laboratory processing
In summer 2016, from early June until mid-October, data for seven traits were collected from the 23 sites from 36 species across the two habitats: height, stem width, number of leaves, SLA, LDMC, LNC, and LCC. Each of these traits has been shown to predict plant strategy and competitive ability (Table 1) Leaf samples were frozen at −20°C for at least 24 hr (Kleyer et al., 2008). After leaves were thawed in deionized water, the fresh weight was measured, and leaves were scanned to determine leaf area (Kleyer et al., 2008). The leaves were then dried for a minimum of 48-hr, in a VWR standing oven 70°C and then reweighed for dry weight. All the weighing took place using a Mettler Toledo ML Series precision balance. Specific leaf area (SLA) was calculated as area of a leaf in millimeters squared (mm 2 ) divided by the dry weight of the same leaf in milligrams (mg). Leaf dry matter content (LDMC) was calculated as the dry weight of a leaf in mg divided by the fresh weight of the same leaf in grams.
Leaf nitrogen content and leaf carbon content were determined using the LECO 628 series elemental analyzer. Composite samples were made for species with extremely small leaves using leaves collected from the same plots, as the minimum weight that the elemental analyzer can detect is 0.1 g.  As a preliminary step, in order to assess whether the variance in CWM and RTV across plots was constant along the invasion gradient, we grouped plots into four groups corresponding to increasing abundances of V. rossicum (e.g., 0-0.25 and 0.25-0.5) and then ran

| Trait-based analyses
Levene's homogeneity of variance test across the groups. With this analysis we also aimed to screen for potential homogenization (i.e., lower variance across plots) in the functional structure of communities at high V. rossicum abundance.
We then used linear mixed effect models to test whether CWMs and RTV were affected by V. rossicum abundance. Both trait-based measures were used as dependent variables in the models with V. rossicum relative abundance as the independent variable and site as a random factor. We used the percent cover data to estimate V. rossicum relative abundance in each plot. Separate models were fit for the understory and meadow sites. All statistical analyses were carried out using R statistical software (R Core Team, 2015). The dbFD package was used to calculate measures of community-weighted means (Laliberté & Legendre, 2010). LME4 package was used to fit the linear mixed effect models (Bates, Maechler, Bolker, & Walker, 2015). The car package was used to calculate Levene's test (Fox & Weisberg, 2011).
Finally, to assess the effect of DTI on co-occurring species abundances, we followed two steps: (a) estimating the effect of the invader on each species and (b) relating this effect to the functional distance to the invader (DTI). In the first step, we used the lme4 package (Bates et al., 2015) to fit, for each species, linear mixed effects models, with site set as a random factor, to examine the relationship between the relative abundances of all other species in each community and the

| Meadow habitat
In the meadow habitat, number of leaves, SLA, LDMC, and LNC were all significantly related to V. rossicum relative abundance in the community ( Figure 2). Specifically, in accordance with our predictions, number of leaves and LDMC were negatively correlated, while SLA and LNC were positively correlated with V. rossicum abundance (Table 2).
In contrast to our predictions, we found no significant relationship of height with V. rossicum abundance in this habitat. Levene's test showed that variances for all traits, except # of Leaves an SLA, were unequal across invader abundance groupings, and specifically decreased with increasing V. rossicum abundance (p-value < 0.05, Table S1).

| Understory habitat
In the understory habitat, CWM values for all traits, except stem width, were significantly explained by V. rossicum relative abundance ( Figure 2). Models for height, LDMC, and LNC showed positive correlations to invader abundance (Table 2). However, models for number of leaves, SLA, and LNC showed negative correlations to V. rossicum, which was partially in contrast to our a priori predictions (Table 2).
Again, Levene's test showed that variances for all traits were unequal across invasion abundance groupings and specifically increased with increasing V. rossicum abundance (p-value < 0.05; Table S1).

| Meadow Habitat including invader trait values
For the analyses conducted in the meadow that included V. rossicum trait values, height, stem width, and number of leaves were negatively correlated with invader relative abundance while SLA was positively correlated with invader relative abundance ( Figure 3). However, these relationships were statistically significant only for stem width and SLA ( Figure 3). Levene's test showed variance for all traits execept SLA and LNC were unequal across V.rossicum abundance.

| Meadow Habitat excluding invader trait values
When we excluded V. rossicum trait values in the RTV calculation, all trait ranges decreased with increasing V. rossicum abundance, though none of these relationships were statistically significant (only marginally significant for height, # leaves and LCC; Table 3). In addition, comparing the two RTV values (Figure 4)

| Understory Habitat
In the understory, only models of stem width and number of leaves exhibited statistically significant relationships of RTVs with V. rossicum abundance (Figure 3). However, for all traits except LCC, there were negative correlations to V. rossicum relative abundance, which indicates that as invader relative abundance increases trait ranges generally decrease (Table 3). Levene's test showed that variance for stem width across the gradient in V. rossicum abundance was unequal, and specifically, variance showed a bell shape distribution (p-value < 0.05; Table S2).

| Effect of distances to invader on species relative abundances
In the meadow habitat, we found a nonsignificant negative relationship between DTI and the coefficients of species relative abundance versus invader abundance ( Figure 5). On the contrary, in the understory we found a marginally significant positive relationship between DTI and how species relative abundance was affected by V. rossicum ( Figure 5). This indicates that the abundance of species further away from V. rossicum in functional trait space tends to be unaffected in highly invaded communities, while species with traits similar to the invader tend to decrease in abundance. Our results are similar to Hejda andDe Bello, (2013), andKnapp andKühn, (2012), which have shown that biological invaders were functionally dissimilar to resident species co-occurring in a habitat and that this is a mechanism through which invasions can alter communities.

| D ISCUSS I ON
Second, we found that in the meadow, species richness is reduced by six species in the maximally invaded plots relative to un-invaded plots. Similarly, in the understory species richness is reduced by four species in the fully invaded plots.  Figure 6).
Indeed, by comparing RTV of the invaded communities between the analyses that included and excluded the invader, it is clear that the invader makes a novel contribution to SLA trait space in the meadow habitat (i.e., empty niche hypothesis). This result highlights that V. rossicum possesses a much higher SLA than most species of un-invaded meadow communities, which, in turn, suggests that V. rossicum has a higher relative growth rate (RGR) compared to resident species (Perez-Harguindeguy et al., 2013). Further, this result is in line with studies by Grotkopp, Rejmánek, and Rost (2002) and Grotkopp and Rejmánek, (2007)

Understory
Functional distance to invader Response to invader important mechanism of invasion impact in the understory than in the meadow. As a consequence, in the understory, species that overlap too much in their traits with V. rossicum tend to be excluded in highly invaded communities, while species that are functionally different are less affected (e.g., Invaded Community 1, in Figure 1).
Nevertheless, on a whole, V. rossicum reduced the abundance of every other species in the understory, which is consistent with the idea that the invader has a higher fitness compared to most resident species (MacDougall et al., 2009).

| Broader implications
The results of this study demonstrate how an invasive species can change communities through alterations to community functional structure. In addition, to communities being increasingly dominated by the traits of the invader, certain species and their traits, are persisting with apparently little negative impact, while other species are excluded from communities. As a consequence, the variance in functional structure (regardless of the exact measure) across communities decreases across the invasion gradient. The net result of these functional alterations is that invaded communities become increasingly spatially homogeneous in their traits and less functionally diverse. This systematic functional alteration will likely result in changes to ecosystems function (Cadotte, Carscadden, & Mirotchnick, 2011). Differing trends across habitats also highlights the need to consider both environmental and biotic filters as factors in the process of invasion-driven community change.
Our finding that an invader can act as a filter, causing trait shifts in invaded communities mirrors similar trends being observed for other invasive species in different habitats (Gallien & Carboni, 2017). Furthermore, the clear fitness advantage of V. rossicum over resident communities also points to a typical mechanism through which invasive species are more successful than resident communities.
However, it also needs to be noted that some of these trends could be the results of differential site histories and characteristics. This is because, though we did include site as random factor in our models, the effect of site histories was not included explicitly in the analyses and it is known to be a potential confounding factor in biological invasions (Ehrenfeld, 2010).
Finally, we also found that V. rossicum is more prevalent in the understory, which could be because of lower species richness in this habitat. Species richness has been shown to reduce the effects of biological invasions in previous studies by reducing the rate of establishment and in certain cases repelling invasion all together (Fargoine & Tilman, 2005). This trend indicates support of the diversity-resistance hypothesis, which states that in more diverse communities there is increased competition for niche space, and this acts as a barrier to potential invaders (Levine, Antonio, Levine, & Antonio, 2010).
Conducting experiments on biological invaders in two distinct habitats can provide insights into how invasive species dynamics and impacts can vary across distinct landscapes. Trait-based assessments to quantify impacts of invasive species are seldom used in invasion ecology, however they provide valuable insights into underlying mechanisms and impacts that aren't entirely observable using other methods. Further, by using trait-based analyses to characterize how communities are impacted during invasion, we gain insight into which traits, and their diversity, need to be considered when restoring invaded ecosystems (Laughlin, 2014;Ostertag, 2015). Overall using trait-based frameworks such as the one used in this study allows for a more complete and nuanced understanding of how invasive species impact communities.

AUTH O R CO NTR I B UTI O N S
MWC and DSS conceived the ideas and designed methodology; DSS and SWL collected the data; DSS and SWL analyzed the data; F I G U R E 6 Vincetoxicum rossicum at an understory study site in the Rouge National Urban Park in Toronto, Ontario, Canada. Photograph: m. Cadotte DSS and MC led the writing of the manuscript. All authors contributed critically to the drafts and gave final approval for publication.

DATA ACCE SS I B I LIT Y
Data from this study has been submitted to DRYAD.