Evolutionary relatedness is typically defined within the context of phylogenetic relatedness or the locations of species relative to one another across a phylogenetic tree that describes the hypothesised evolutionary relationships amongst species. Many studies investigating the effects of phylogenetic relatedness on invasion have done so in the context of the evolutionary relatedness between invasive and native species, i.e. within the context of “Darwin’s naturalisation hypothesis” that species from novel genera may have an advantage when invading new locations because they are less likely to compete with closely related species or share natural enemies with them (e.g. Diez et al. 2008, Lososová et al. 2015). Nevertheless, we are unlikely to know ahead of time the particular species that will invade a given community. For this reason, knowing how the phylogenetic relatedness amongst native species in a given community affects invasion in general is likely of greater utility in understanding spatial heterogeneity in the degree to which native ecosystems resists plant invasions. Ecological communities consisting of species that are more divergent relative to one another across the phylogenetic tree of life can exhibit greater resistance to invasion (Gerhold et al. 2011, Lososová et al. 2015), although the degree of resistance varies considerably across large spatial scales (Iannone et al. 2016). This geographic variability suggests that the effect of evolutionary relatedness of native communities on exotic invasions is itself affected by other factors.

To meet our objective, we first obtained native tree and invasive plant data from 42,626 forested plots from the United States Department of Agriculture’s Forest Inventory and Analysis programme (FIA) located within the two ecological domains (sensuCleland et al. 1997) of the eastern USA (Suppl. material 1, fig. S1.1). The location of these data, as well as all other data used for subsequent analyses in this paper can be found below in the paper’s Data accessibility section. FIA monitors spatiotemporal patterns of forest conditions at the national-level using a fixed grid of permanent plots occurring at an approximate intensity of one plot for every 2,428 ha of forests; all plots are 0.40 ha in size and are defined using a consistent plot design (Bechtold and Patterson 2005). Forests are defined as areas at least 36.6 m wide and 0.40 ha in size that historically or currently have at least 10% tree cover (for trees of any size).

We used four metrics (Table 1) to estimate three characteristics of evolutionary relatedness amongst the native trees in each plot. First, we constructed a phylogenetic supertree of 397 native angiosperm and gymnosperm tree species occurring in the FIA plots of the contiguous 48 states (Potter 2018). This reference phylogeny, with branch lengths measured in millions of years, was constructed based on a review of 189 molecular systematics and paleobotanical studies. Using this phylogenetic tree and the R package Picante (Kembel et al. 2010), we quantified two different aspects of phylogenetic divergence, following Helmus et al. (2007) —phylogenetic species variability (PSV) and phylogenetic species clustering (PSC). PSV quantifies variability amongst species in a hypothetical random (non-selected) trait, thus quantifying how divergent species are relative to one another across an entire phylogenetic tree and the extent to which the phylogenetic arrangement of these species represents a star phylogeny, as illustrated in Table 1. PSC quantifies the degree to which the phylogenetic branch tips for species in a given community are divergent, or clustered, relative to one another. From the same phylogenetic tree, we also estimated the amount of evolutionary history within each plot using the metric phylogenetic diversity (PD) proposed by Faith (1992). PD is estimated by summing the lengths of the phylogenetic branches for the tree species found within a given plot. By definition, PD increases with species richness. The utility of PD in describing the amount of evolutionary history within a given community was also confirmed by Tucker et al. (2017). Lastly, we calculated the phylogenetic species evenness (PSE) metric proposed by Helmus et al. (2007). This metric is an indicator of phylogenetic divergence that accounts for the relative abundances of species within the investigated community. PSE equals PSV when all species are equally abundant in a community and declines as they are decreasingly so.

A benefit of using PSV, PSC and PSE is that these metrics of phylogenetic divergence do not require prior knowledge of the regional species pool from which species could populate a plot (Helmus et al. 2007). An initial correlation analysis revealed PD to be strongly related to and, thus, dependent on, the number of native tree species in each plot (r_PD = 0.82). The three other phylogenetic metrics, however, exhibited weaker relationships with native tree species richness (r_PSV = -0.31, r_PSC = 0.49 and r_PSE = 0.09) and, thus, reflected more independent characteristics.

We assessed overall plant invasion in each plot by compiling data on invasive plant richness and cover data, following Iannone et al. (2016). Invasive richness is the number of invasive plant species found and invasive cover is the summed total cover of these species (which can exceed 100% of the area of a plot because of overlap by individual plants). Invasive richness and cover serve as indicators of invasive species establishment and dominance, respectively. FIA defines invasive plants in accordance with USA Executive Order 13112 as non-native plant species likely to cause economic or environmental harm (Ries et al. 2004). The list of monitored invasive plants in the eastern USA includes approximately 76 different species; some species were merged into single groups to accommodate field identification. These species include 18 forbs, 9 grasses, 21 shrubs, 14 trees and 14 vines (Suppl. material 2, table S2.1). Although invasive plant sampling protocols, including the specific species monitored, vary between the Northern and Southern FIA Administrative Regions (described in Suppl. material 2, text S2.1), this variability has little effect on macroscale invasion patterns (Iannone et al. 2015). Furthermore, most of the invasive plants monitored are of concern to large portions of each and, in some cases both, of the administrative regions, alleviating concerns about spatial variability in invasive plant species pools.

Using a mixed-effects modelling framework developed by Dixon Hamil et al. (2016), we modelled invasive richness and cover in response to the four phylogenetic metrics (PSC, PSV, PSE and PD), allowing for independent (i.e. random) intercept and slope estimates for each of the 91 ecological sections (sensuCleland et al. 1997) nested within our study region (Suppl. material 1, fig. S1.2). Phylogenetic metrics were standardised [x – mean(x)/SD(x)], allowing for cross-variable comparison (Schielzeth 2010). Inspection of variance inflation factors (Legendre and Legendre 2012) revealed no evidence of complications due to collinearity. We assessed model fit using the coefficient of determination for generalised mixed-effect models, i.e. R²_GLMM (Johnson 2014). To assess statistical significance, we resampled section-level slope estimates (i.e. conditional plus fixed estimates) 10,000 times allowing for replacement and calculated the bias corrected, accelerated 95% confidence interval (CI) of each bootstrapped distribution (Efron and Tibshirani 1986). We assumed the phylogenetic metric to be a statistically significant indicator of biotic resistance across our study region when its 95% CI encompassed values less than zero. Likewise, we assumed a phylogenetic metric to be an indicator of invasion success when its 95% CI encompassed values greater than zero. We relied on bootstrapped rather than actual distributions to estimate 95% CIs because actual distributions were non-normal (Suppl. material 3, fig. S3.1).

We compiled data from the FIA database on plot-level maximum tree height (m) and relative tree density. Maximum tree height was used as an inverse indicator of environmental harshness. Marks et al. (2016) found maximum tree height to be negatively correlated to multiple environmental indicators of environmental harshness across forests of both the eastern and western USA. Relative tree density was calculated as described by Potter and Woodall (2014). This standardised metric quantifies the number of trees per unit area currently in a given FIA plot and divides that value by the maximum number of trees that could grow in that plot. We then calculated means and SD for each of these metrics at the ecological section level (Suppl. material 1, fig. S1.2; Suppl. material 1, table S1.1). Means estimated overall section-level environmental harshness and relative tree density, while SD of these metrics served as section-level indicators of the environmental (abiotic and biotic) variability experienced by invading plants. As an additional, indirect metric of environmental variability, we calculated the mean Jaccard’s distance (based on absence/presences of native tree species) amongst plots within each of the 91 ecological sections. Ecological sections having greater mean Jaccard’s distances in native tree species composition were assumed to be more environmentally variable.

We conducted preliminary regression analyses to assess the degree to which maximum tree height, relative tree density and SD in these variables predicted forest age. We did so to rule out the possibility of our findings reflecting nothing more than forest successional stages, i.e. variation between young and old forests in the degree of invasion. Both maximum tree height and relative tree density explained relatively small proportions of variability in FIA estimates of forest stand age at both the individual plot and ecological section levels (range in R² = 0.04 to 0.24). SD of maximum tree height and relative density also explained relatively small proportions of forest stand age (R²= 0.25 and 0.04, respectively). These low R² values revealed that mean and SD of maximum tree height and relative density were largely indicative of environmental conditions other than forest successional stages.

We modelled slope estimates for relationships between phylogenetic metrics indicative of biotic resistance and invasive richness and cover in each of the 91 ecological sections (determined as described above) in response to the section-level mean and SD of maximum tree height and relative tree density, as well as section-level estimates of mean Jaccard’s distance (model shown in Table 2). We used these statistical models to assess whether environmental harshness, relative tree density and/or environmental variability drive spatial heterogeneity in the degree to which evolutionary relatedness contributes to invasion resistance. We assessed model fit using the conditional estimate for the coefficient of determination for generalised mixed-effect models, i.e. R²_GLMM (Johnson 2014).

Prior to analysis, we standardised explanatory variables [x – mean(x)/SD(x)], allowing us to assess the relative contribution of each to this spatial heterogeneity (Schielzeth 2010). To account for possible spatial autocorrelation amongst more closely located ecological sections, we incorporated the ecological province (Suppl. material 1, fig. S1.3), i.e. the larger ecoregion scale in which each ecological section was nested, into statistical models as a random effect. The distribution of slope estimates for relationships between invasion and phylogenetic metrics were left-skewed (Suppl. material 3, fig. S3.1). To address this, we transposed slope estimates to positive values greater than 1 and transformed the data using a power transformation [(x + abs(min(x)) + 1)⁷] to help meet the assumption of normality. Inspection of variance inflation factors (Legendre and Legendre 2012) revealed no evidence of complications due to collinearity.

Statistical analyses were conducted in R v 3.0.1 (R Core Team 2014) using the nlme package (Pinheiro et al. 2017) to construct linear mixed-effects models, the MuMIn package (Bartoń 2014) to estimate coefficients of determination and the bootstrap package (Tibshirani and Leisch 2013) to conduct bootstrap analyses.

Mapping the section level slope estimates determined via mixed-effects modelling confirmed spatial variability in relationships between phylogenetic (PSC, PSV, PD and PSE) and invasion (richness and cover) metrics (Fig. 1). Slope values for PSC and PSV were largely negative across ecological sections, while those for PD and PSE were largely positive. The 95% confidence intervals (CI) of the bootstrapped distributions of section-level slope estimates revealed these trends to be statistically significant, i.e. 95% CIs did not encompass zero (Suppl. material 3, fig. S3.2). Slope values for PSC and PSV also exhibited greater spatial variability and magnitudes in their relationships with invasive richness and cover than did PD and PSE (Fig. 1; Suppl. material 3, fig. S3.2). (See Suppl. material 3, table S3.1 for values of section level slope estimates.) Given that PD and PSE were not indicative of biotic resistance to invasion, we did not investigate the drivers of spatial heterogeneity in relationships between these metrics and invasion metrics.

Spatial variability in the degree to which both PSC and PSV were related to invasive richness was largely explained by mean and SD maximum tree height (Table 2; R²_GLMM of statistical models = 0.55 and 0.47, respectively). Slopes of all other model terms were not statistically significant from zero (p-values = 0.15 to 0.57). Slopes for mean maximum tree height were negative (Table 2), revealing both PSC and PSV to be more negatively related to invasive richness in ecological sections having greater maximum tree heights (Figs 2a, c). In contrast, slopes for SD maximum tree height were positive (Table 2). This reveals both PSC and PSV to be less negatively related to invasive richness in ecological sections having more variable maximum tree heights (Figs 2b, d). Standardised slope estimates revealed mean maximum tree height to have more than twice the effect as SD on this spatial variability for PSC and a 27% greater effect than SD for PSV (Table 2). Slopes for mean maximum tree height had smaller p-values (Table 2) and smaller 95% CIs relative to their slopes (Fig. 2), than those for SD maximum tree height, suggesting a more consistent impact on spatial heterogeneity in relationships between invasion richness and both PSC and PSV.

Spatial variability in the degree to which both PSC and PSV were related to invasive cover was explained solely by mean maximum tree height (Table 2; R²_GLMM of statistical models = 0.63 and 0.49, respectively). Slopes of all other model terms were not statistically significant from zero (p-values = 0.26 to 0.95). Slopes for mean maximu m tree height were negative (Table 2), revealing both PSC and PSV to be more negatively related to invasive cover in ecological sections having greater overall maximum tree heights (Figs 3a, b).

Follow-up mapping of section-level estimates of statistically signification explanatory variables shown in Table 2, i.e. mean and SD of maximum tree height, revealed spatial patterns contributing to our current findings. The greatest values for mean maximum tree height occurred in and around the Appalachian Mountains (Fig. 4a) where both PSC and PSV exhibited stronger negative associations with invasive richness and cover (Fig. 1). This pattern contributed to the strong negative slope values exhibited by maximum tree height in all statistical models (Table 2). The greatest values for SD of maximum tree height occurred in the southern portion of our study region (Fig. 4b) where PSC and PSV both tended to exhibit weaker and sometimes positive relationships with invasive richness (Fig. 1). This pattern contributed to the positive slope values for SD maximum tree height in statistical models for relationships between invasive richness and both PSC and PSV (Table 2).

Iannone BV III, Potter KM, Guo Q, Jo I, Oswalt CM, Fei S (2018) Data on native tree diversity (species richness and phylogenetic), biomass, relative density, tree height and invasive plants in forests of the eastern USA. Purdue University Research Repository.

https://purr.purdue.edu/publications/3072/1 doi: 10.4231/R7GX48TW.