A regional assessment of white-tailed deer effects on plant invasion

Analyses deriving insights from 23 research sites across the northeastern and east central US reveals that white-tailed deer lower native plant diversity and increase the fraction of the plant community made up of non-native species. Particularly, deer increase the abundance of the invasive plants garlic mustard and Japanese stiltgrass. Deer are known to avoid eating these species in favour of more palatable ones, thereby indirectly facilitating the success of the non-native invasives. Managing deer abundance has implications for forest plant communities. By maintaining lower deer densities, native plants likely will be better sustained and invasions of unpalatable non-native plants limited.


Introduction
Modern plant communities are anthropogenically altered (Hannah et al. 1994). Habitat loss and forest fragmentation have contributed to acute reductions in biodiversity, species homogenization, and concomitant proliferation of invasive species and some large herbivores (McKinney and Lockwood 1999;Rooney et al. 2004;Alroy 2008). Because large mammalian herbivores can play a prominent role in determining plant community composition (Harper 1977;Crawley 1997;Russell et al. 2001;Côté et al. 2004), understanding their effects on plant species and communities, including plant invasions, is critical for conserving biodiversity.
Large herbivores affect plant communities directly via tissue loss and plant mortality, indirectly through nonconsumptive effects including trampling (Persson et al. 2000;Heckel et al. 2010), accelerating nutrient cycling (Hobbs 1996;Rooney and Waller 2003) and by dispersing plant propagules (Vellend 2002;Myers et al. 2004; Bartuszevige and Endress 2008;Williams et al. 2008;Castellano and Gorchov 2013). Perhaps the most pervasive effect of large mammals on plant communities, however, is their indirect impact of altering interspecific plant competition through selective herbivory and plant response to herbivory (Holt 1977;Bowers 1993;Crawley 1997;Augustine and McNaughton 1998), with large impacts on community assembly and succession (Drake 1990;Hobbs 1996). For example, herbivores can alter successional trajectories when they preferentially consume early or late successional plant species (Hobbs 1996;Crawley 1997;Côté et al. 2004;DiTommaso et al. 2014;Forsyth et al. 2015). Consumption of palatable species can cause unpalatable species to gain an apparent competitive advantage and potentially become dominant or invasive (Leopold et al. 1947;Holt 1977;Augustine and McNaughton 1998;Horsley et al. 2003;Côté et al. 2004;Vavra et al. 2007). For example, pastures and rangeland can become infested with Carduus, Centaurea and Cirsium spp., among others, when grazers consume more palatable species (DiTomaso 2000). Selective herbivory can result in woody plant invasion in savannas, i.e. encroachment, which occurs as grazers reduce herbaceous species, indirectly facilitating establishment of unpalatable woody vegetation (Asner et al. 2004), but the more common result is a reduction of palatable woody plants, which slows succession from field to forest (DiTommaso et al. 2014;Habeck and Schultz 2015). The selective browsing of cervids (e.g. deer, moose, elk) is considered one of the main determinants of forest understory plant species composition and structure (Alverson et al. 1988;Côté et al. 2004;Abrams 2013). Herbivore-mediated shifts in plant communities can limit native plant regeneration, alter the abundance of small mammals, birds and insects, lower ecological stability (e.g. erosion and flood protection), disrupt ecosystem functioning, induce alternative stable states, reduce the economic value of habitats (reviewed in Côté et al. 2004) and trigger or facilitate plant invasions (Stromayer and Warren 1997;Vavra et al. 2007).
In North America, many large native herbivores, including bison (Bison bison), caribou (Rangifer tarandus), Dall's sheep (Ovis dalli), elk (Cervus elaphus), moose (Alces alces) and pronghorn (Antilocapra americana), have experienced severe range contractions during the past 200 years (Laliberte and Ripple 2004). However, the range and abundance of native white-tailed deer (Odocoileus virginianus; hereafter referred to as deer) increased steadily following steep population declines in the late 1800s (McCabe and McCabe 1997;Laliberte and Ripple 2004). Low predator populations (Laliberte and Ripple 2004) and game laws that restricted hunting, in addition to increasing agricultural, silvicultural and early successional habitat, enhanced deer habitat within the past century, resulting in high deer populations (Alverson et al. 1988;McShea et al. 1997;Waller and Alverson 1997;Côté et al. 2004). Today, deer are the dominant wild ruminant herbivore in east-central and north-eastern USA and, because of their high abundance, are a serious ecological and management concern (McShea et al. 1997;Rooney 2001;McShea 2012). While deer at low abundances can increase floristic diversity (Royo et al. 2010a;Cook-Patton et al. 2014), abundant deer limit diversity and promote floral homogeneity Wiegmann and Waller 2006). At chronically high densities, deer change plant community structure and composition enough to be considered 'ecosystem engineers' or 'keystone herbivores' (Alverson et al. 1988;Waller and Alverson 1997;Côté et al. 2004). In many areas, deer population densities greatly exceed ecosystem carrying capacity (Rooney and Waller 2003), causing long-lasting and potentially irreversible legacy effects (Royo et al. 2010b;Nuttle et al. 2011).
As deer abundance increased during the past century, so did abundance of introduced plants, resulting in often concurrent ecological impacts. Human transport facilitates movement of many species outside their native ranges and, consequently, non-native species are now prominent components of present-day communities (Lockwood et al. 2013). Introduced plant species pose a growing threat to native plant communities, as their presence is associated with altered diversity, community structure and ecosystem function (MacDougall and Turkington 2005;Ehrenfeld 2010;Vilà et al. 2011;Beasley and McCarthy 2011). The fact that populations of deer and introduced plants have expanded concurrently suggests that deer abundance might be linked to introduced plant invasions ( Augustine and McNaughton 1998;Vavra et al. 2007). However, data are lacking on regional effects of deer on native plant communities and plant invasion (Maron and Vilà 2001;Russell et al. 2001;Mosbacher and Williams 2009).
Throughout the past century, numerous experiments using fenced (deer-exclusion) and unfenced (deer-access) plots gauged deer impacts on forest plant communities (e.g. see McShea et al. 1997;Côté et al. 2004;Abrams and Johnson 2012;Habeck and Schultz 2015). Use of paired plots affords valuable insight into effects of large herbivores on floristic composition and on native and introduced plants, yet site-level studies assessing the degree to which deer influence introduced plants have yielded equivocal results. Several paired-plot experiments report deer facilitate certain invasive plants (Knight et al. 2009;Eschtruth and Battles 2009b;Beasley and McCarthy 2011;Kalisz et al. 2014;Dávalos et al. 2015b), others report deer mitigate invasions of different species (Rossell et al. 2007;Shelton et al. 2014) and others find no effect (Bowers 1993;Levine et al. 2012;DiTommaso et al. 2014) or mixed effects (Cadenasso et al. 2002;Webster et al. 2005;Knapp et al. 2008;Shen et al. 2016). Site-level investigations can provide practical insights about local species and conditions, but cannot be extrapolated to regional assessments about deer herbivory and plant invasion. A regional assessment requires data on a range of plant community types across a range of deer densities. Spatially broad investigations can bolster generalizations and forecasts made about ecological processes (Clark et al. 1999(Clark et al. , 2001, such as community assembly and plant invasion (Gill and Beardall 2001;Shen et al. 2016).
Here, we present results of a multisite, regional assessment of white-tailed deer effects on composition, richness and abundance of introduced and native plants in east-central and north-eastern USA. We pool data from 23, paired-plot deer access/exclusion experiments spanning a broad range of invasive plant abundance and deer densities. We predicted that deer access would (i) alter floristic composition and reduce floristic diversity and (ii) increase richness and abundance of introduced plant species and decrease richness and abundance of native plant species.

Data description
We compiled data sets in which herbaceous and woody floristic composition and abundance were quantified in replicated deer-exclusion and deer-access plot experiments across 23 sites, resulting in 446 experimental units (223 plot pairs) (Table 1). We acquired data sets by directly contacting investigators of previously published (15 sites) and unpublished data (6 sites) and collecting additional data from established plots (2 sites, Long Run and Marienville) ( Table 1). Sites were located in temperate deciduous or mixed deciduous forests across east-central and north-eastern USA (Table 1; Fig. 1). Sites were initially established to answer a range of research questions, not solely the effects of deer on introduced plants (Table 1). Overstory species typically included oak (Quercus spp.), maple (Acer spp.), beech Table 1. Descriptions of 23 experimental sites and data used in pooled analyses testing the effect of white-tailed deer on introduced and native plants in east-central and north-eastern USA. Floristic composition data were collected from deer-access (unfenced) and  Gleason and Cronquist (1991) Acorns and oak regeneration  (Fagus grandifolia), tulip-poplar (Liriodendron tulipifera) and black cherry (Prunus serotina) (Table 1). Deer density estimates varied across sites from 4 to 107 deer km −2 (Table 1; for estimation methods used, see Supporting Information- Table S1). The timing and duration of deer exclusion varied across experiments (Table 1). Six sites were established in the late 1980s/early 1990s, and the remaining 17 sites in the 2000s. At 15 sites, deer exclusion was imposed for 6 years or less, while at the other eight sites it ranged from 8 to 17 years. During the summer growing season, abundance data of herbaceous and woody species up to 2 m in height was recorded. Sampling intensity, plot area and replication varied across sites (Table 1). A recent meta-analysis showed no relationship between plot area and plant community responses to deer (Habeck and Schultz 2015). Fence heights used to exclude deer were a minimum of 2 m. Fence mesh size varied across experiments; therefore, deer may not have been the only mammalian herbivore excluded (e.g. see Bowers 1993).
We acknowledge that the paired-plot approach has limitations, including fence-line effects, fences providing artificial support for vines and concentrated perch areas for birds, and an unrealistic total absence of deer in fenced plots (Russell et al. 2001;White 2012). Deer also presumably exerted an influence prior to experimentation (Russell et al. 2001), leaving behind legacy effects even after culling (Royo et al. 2010b;Nuttle et al. 2011), such as altered seedbank composition (DiTommaso et al. 2014), which could limit vegetation response to deer exclusion. Beyond the scope of this work, drivers of invasion could vary between areas with deer access vs. deer exclusion. Despite these limitations, herbivoreexclusion experiments remain among the most straightforward of ways to test the effects of herbivores on plant invasions (e.g. Parker et al. 2006).

Data set pooling
The pooling approach taken here has the benefit of increasing statistical power and reducing type II error rates (i.e. false negatives; Blettner et al. 1999). We processed the most recent floristic assessment from each experiment to analyse vegetation patterns at single points in time. However, we recognize that plant communities and deer densities vary temporally. Thus, analysing the temporal aspect of plant community assembly could improve conclusions about the interaction between deer and plant invasion since legacy effects play out over decadal time frames (Royo et al. 2010b;Nuttle et al. 2011;White 2012). Nonetheless, Figure 1. Locations of 23 deer research sites in east-central and north-eastern USA included in pooled analyses. Symbol size indicates sampling intensity across sites, which are labelled with two-letter codes (see Table 1 for additional site information). because sampling was spread across a wide range of sites and years, we expect observed patterns to be robust. We analysed equal numbers of deer-exclusion and deer-access plots from each data set (Table 1) and weighted plots equally.
We analysed plant species presence/absence and two abundance metrics, cover and stem density. Across experiments, plant abundance was quantified in several ways, including stem density (14 sites), per cent cover (8 sites), cover classes (i.e. ranges of per cent cover; 15 sites) and/or density classes (i.e. ranges of population density; 7 sites) ( Table 1; for ranges of cover classes used and for treatment of density classes, see Supporting Information- Fig. S1, Text S1). We converted cover data to cover classes (for detailed processing methods, see Supporting Information-Text S1) and then used midpoints of cover classes (e.g. the midpoint of a 5-25 % cover class is 15 %) in analyses (hereafter referred to as cover).
Some plant species in almost every data set were unidentified and marked as unknown at genus, family or growth habit level (e.g. forb, fern, graminoid, woody seedling). We excluded these from analyses that required knowledge of native/introduced status, but otherwise included them in indicator species analyses and when determining total plot species richness and abundance. We statistically tested effects of deer access/exclusion and deer density on richness and abundance of unknown species [see Supporting Information- Table S2]. We standardized species taxonomy and native/introduced status according to the United States Department of Agriculture Plants Database (USDA NRCS 2012). Taxa with status code of 'Native and Introduced' (i.e. some infra-taxa are native and some are introduced) were classified as native. We define introduced plants as invasive according to the USDA Forest Service (1998) and the USDA NRCS (2012).

Statistical analysis
We used mixed effects linear regression to test for effects of deer on relationships between native and introduced species richness and abundance. Introduced vegetation (i.e. species richness, cover or density) was the response variable, native vegetation and deer access/exclusion or deer population density were the fixed effects, and plot pair and site were the random effects. For the native vegetation effect in these models, native cover was used in the introduced cover analysis, native stem density in the introduced stem density analysis, and native species richness in the introduced species richness analysis. Deer density analyses were utilized only for unfenced, deer-access plot data here and below.

Deer effects on floristic composition, diversity and community-level abundance.
We used the multiresponse permutation procedure (MRPP) (Mielke and Berry 2007) to test for community-level differences in floristic composition between deer-access and deerexclusion plots using the Sørensen (Bray-Curtis) distance measure, which is not likely to exaggerate the influence of outliers in heterogeneous data, with PC-ORD software (McCune and Grace 2002). We conducted separate MRPPs for presence/absence and each abundance metric, cover and density. We calculated Shannon diversity (H′) (Shannon 1948), a combined measure of species richness and relative abundance (Hill 1973;McCune and Grace 2002), for each plot where at least one species was present. We calculated H′ for each abundance metric to determine floristic diversity using the equation: where S is the total number of species measured according to each abundance metric and p i is the proportional abundance of species i in the plot.
We used linear mixed effects analysis of variance (ANOVA) to test for effects of deer access/exclusion and linear mixed effects regression to test for effects of deer population density on plant richness and absolute abundance of (i) native species, (ii) introduced species and (iii) total species (native plus introduced plus unknown species) and on Shannon diversity. Deer access/exclusion (fencing treatment) or deer population density were fixed effects and plot pair and site were random effects. We evaluated both absolute and proportion of introduced plant abundance (i.e. relative abundance) and plant species richness because they represent different indices of plant invasion; the former represents actual introduced plant abundance/species richness, while the latter represents the portion of plant community abundance/species richness composed of introduced plants. Absolute introduced plant abundance/species richness was evaluated based on the main effect of deer and proportion introduced plant abundance/species richness was evaluated based on the interaction of the deer effect with total vegetation. A significant interaction indicates that the ratio of introduced to total plant abundance/ species richness (i.e. proportion introduced) varies with the deer effect. The ratio of introduced to native vegetation provides another index of plant invasion and was evaluated by testing the interaction of the deer effect with native vegetation. A significant interaction would indicate that the ratio of introduced to native vegetation varies with the deer effect. Total cover could exceed 100 % due to overlapping leaves of different species. We excluded sites lacking introduced plants from community-level mixed model analyses with introduced plants in the response variable.
We acknowledge that accurate deer density estimation is particularly difficult in forests (Putman et al. 2011). Total population counts can underestimate the actual number of deer by a factor of four or more (Andersen 1961). While distance sampling (Buckland et al. 1993(Buckland et al. , 2001, used to inform many of the estimates included in analyses here [see Supporting Information- Table S1], is a more accurate sampling approach vs. total population counts, considerable error surrounds single estimates and cannot fully account for season-to-season or year-to-year fluctuations or legacy effects of previous deer populations. Additionally, only deer-access plot data were used in these analyses; thus, the paired-plot baseline provided by fenced-plot data is lacking. Due to these limitations, we exercised caution in interpreting results of regression analyses. All sites were included in deer density analyses except West Point, for which deer population density estimates were unavailable.

Deer effects on individual introduced and native species' abundance.
To follow up the MRPP and determine which species might be driving community-level differences, we used indicator species analysis (ISA) to test for species and genera affinities for deer access or deer exclusion (Dufrêne and Legendre 1997). The ISA results show which plant species or genera associate with deeraccess or with deer-exclusion plots. We calculated indicator values for each species by multiplying the relative abundance across all plots by the relative frequency across plots within each treatment. We used a Monte Carlo randomization test to determine significance of indicator values, which range from 0 (not detected) to 100 (exclusive association). We conducted separate ISAs for presence/absence, cover and density data. We used PC-ORD software for the ISAs (McCune and Grace 2002). We report species as significantly associated with a treatment when α < 0.05.
We used linear mixed effects ANOVA to test the main effect of deer access/exclusion on abundance of individual introduced and native plant species. We conducted these species-level abundance analyses for the most frequently occurring introduced plants (defined here as species present in >5 % of all plots; a total of 13 introduced species) and for the 20 most frequent native plant species (present in >12 % of all plots). Plot pair and site were random effects included in models to control for within-and between-site variability, respectively.
Non-linearities are pervasive in ecology (e.g. Lockwood et al. 2013;Turner and Gardner 2015), yet we did not analyse them in the data presented here, opting instead to transform the data and test for linear patterns. For community-and species-level mixed models, we used square root or natural log transformations of response variables when necessary to meet statistical assumptions of normality and homogeneity of residuals. In community-level analyses, we report 95 % confidence intervals for significant fixed effects (α < 0.05) and, for mixed effect models with a significant deer treatment effect (α < 0.05), we determined least square means using t-tests (based on the Satterthwaite approximation for denominator degrees of freedom). To determine significance of random factors, we used log-likelihood ratio tests (chi-square with one degree of freedom, i.e. one effect tested at a time). We used the lme4 (Bates et al. 2015), lmerTest (Kuznetsova et al. 2013) and vegan (Oksanen et al. 2013) packages for mixed model analyses in R version 3.1.2 (R Development Core Team 2014). We report plot-level means and standard errors.

Results
We recorded 50 introduced and 345 native species in the regional forest understory species pool. Fifty-four species, six of which were introduced, only occurred in deeraccess plots. In contrast, 72 species, 16 of which were introduced, only occurred in deer-exclusion plots. Of the introduced species, 32 % occurred only in deer-exclusion plots; 16 % of native species occurred only in deerexclusion plots. A higher proportion of native species occurred in both deer-access and deer-exclusion plots (70 %) than of introduced species (56 %). Introduced and native species richness and abundance were significantly positively correlated (Table 2; see Supporting Information- Fig. S2). We detected no effect of deer on the ratio of introduced to native vegetation (nonsignificant interactions between deer effect and native species vegetation) (Table 2). At five sites, no introduced species were observed. Total species richness was 23 % higher at sites where introduced species were present vs. where they were absent. For species richness and abundance by deer access/exclusion treatment and site, see Supporting Information-Tables S4 and S5, respectively.

Deer effects on floristic composition, diversity and community-level abundance
Species composition was significantly different between deer-access and deer-exclusion plots based on all three MRPP analyses despite high heterogeneity among plots within each treatment (Table 3). Deer-access plots had lower Shannon diversity (H′) than deer-exclusion plots (Table 4a; Fig. 2) and, among deer-access plots, H′ (cover but not density) was negatively correlated with deer density (Table 4b). Deer exclusion did not affect introduced plant species richness or the proportion of introduced plant species (non-significant interaction between deer access/exclusion and total species richness) (Table 5a; Fig. 3). However, as deer density increased, the proportion of introduced species increased (significant interaction between deer density and total species richness) (Table 5b). Deer-access plots had 16 % lower native plant species richness and 10 % lower total plant species richness than deer-exclusion plots (Table 5a; Fig. 3). Table 2. Mixed model effects of white-tailed deer a) access/exclusion and b) population density and native vegetation on introduced plant richness and abundance (per cent cover and stem density) a . Results are based on floristic composition data collected from deer-access (unfenced) and deer-exclusion (fenced) plots at 23 sites in east-central and north-eastern USA. The ratio of introduced to native vegetation was evaluated based on the interaction of the deer effect with native vegetation; the lack of significant interactions indicates that the ratio of introduced to native vegetation does not vary with the deer effect. See Supporting Information-Fig. S2 for the relationships between introduced and native vegetation. For random effect results, see Supporting Information- Table S3. P values are in bold print if significant at the alpha level α < 0.05. While deer-access plots tended to have higher absolute introduced plant abundance than deer-exclusion plots, these trends were not statistically significant ( Fig. 2; Table 4). However, in deer-access plots, native stem density was 16 % lower and native cover was 46 % lower than in deer-exclusion plots ( Fig. 2; Table 4). Total stem density was 2 % lower and total cover was 27 % lower in deer-access plots than in deer-exclusion plots. The proportion of introduced plant cover was 44 % higher in deer-access vs. deer-exclusion plots ( Fig. 2; Table 4a). The proportion of introduced plant stems was numerically, but not statistically, higher in deer-access plots ( Fig. 2; Table 4a). The proportion of introduced plant abundance (cover and stem density) was positively correlated with deer density, while native and total plant cover were negatively correlated with deer density (Table 4b).

Introduced species richness
In summary, deer had no effect on absolute introduced abundance but they increased the proportion composed of introduced species. The difference between the absolute and proportional metrics is native vegetation, which is reduced by deer. Thus, deer indirectly increase the proportion of introduced vegetation via their negative influence on native vegetation. Notably, we detected these effects after accounting for within-and between-site variability, which were significant random factors [see Supporting Information- Table S3]. More unknown species occurred in deer-access plots than in deer-exclusion plots but unknown species abundance was unaffected [see Supporting Information- Table S2].

Deer effects on individual introduced and native species' abundance
Deer access/exclusion differentially affected introduced and native plant species. Indicator species analysis results showed that three introduced species and four native species were indicators of deer-access plots, while three introduced and 15 native species were indicators of deer-exclusion plots (Table 6). Two introduced plants, the grass Microstegium vimineum and the herb Alliaria petiolata, occurred in a large number of plots and sites and were the best indicator species (by indicator value) of deer-access plots (Table 6). Indicator species of deer-exclusion plots included the introduced vine Lonicera japonica and shrub Rosa multiflora (Table 6). In general, introduced indicator species were found to be more abundant in their respective deer-access or deerexclusion plots using mixed model analyses (Table 7). In deer-access plots, absolute abundance of three introduced species, M. vimineum, A. petiolata and the tree Ailanthus altissima, was higher than in deer-exclusion plots. In contrast, three other introduced species, L. japonica, R. multiflora and Duchesnea indica, occurred in lower abundance in deer-access vs. deer-exclusion plots ( Table 7).
The native herbs Pilea pumila and Oxalis stricta were indicators of deer access, but others, including Maianthemum racemosum and Polygonatum biflorum were indicators of deer exclusion. Native trees P. serotina, Ulmus rubra, two Carya spp. and Quercus rubra and native shrubs Viburnum acerifolium and Rhododendron periclymenoides were indicator species of deer exclusion ( Table 6). The cover of native trees Fraxinus americana, Acer rubrum and Celtis occidentalis and the native shrub Lindera benzoin was reduced in deer-access plots relative to deer-exclusion plots ( Table 7). The native vines Parthenocissus quinquefolia and Toxicodendron radicans were indicators of deer exclusion and occurred in greater abundance in deer-exclusion vs. deer-access plots (Tables 6 and 7). Unknown species in the genera Polygonum, Oxalis, native Prenanthes and Solanum were associated with deer access, while unknown Rubus spp. indicated deer exclusion. For frequencies of each taxon recorded in deer-access plots, in deer-exclusion plots and overall, see Supporting Information- Table S6.

Discussion
White-tailed deer alter plant species composition and reduce community-wide plant diversity, upholding our first prediction. Deer facilitate some but not all introduced plant species and strongly negatively affect native plant species, offering partial support to our second prediction that deer would benefit introduced plants and disadvantage native plants. This work substantially clarifies previous conflicting reports of deer effects on introduced plants. By exploring deer-plant patterns across the region, our results provide evidence that attribute the seemingly contradictory findings in individual sitelevel studies to species-level differences, illustrating Table 3. Results of MRPPs, testing the effect of white-tailed deer on species composition in east-central and north-eastern USA. Separate analyses were conducted for species presence/ absence and abundance, per cent cover or stem density. The agreement statistic, A, indicates within-group homogeneity compared to random; A varies between 0 (heterogeneous plots) and 1 (homogenous plots). The P value and the number of plots within each group, deer access or deer exclusion, are shown. The number of plots was constrained in analyses due to plots with zero vegetation [see Supporting Information-Text S1].  the presumed consequences of differential palatability to deer (Senft et al. 1987;Keane and Crawley 2002;Averill et al. 2016). Our results are consistent with previous research showing that deer can increase (Knight et al. 2009;Eschtruth and Battles 2009b;Beasley and McCarthy 2011;Kalisz et al. 2014;Dávalos et al. 2015b), decrease (Rossell et al. 2007;Shelton et al. 2014), have no effect (Bowers 1993;Levine et al. 2012;DiTommaso et al. 2014) or mixed effects (Cadenasso et al. 2002;Webster et al. 2005;Knapp et al. 2008;Shen et al. 2016) on introduced plants. Where deer facilitate an increase in introduced plant abundance, plant invasion via enemy release (Elton 1958;Keane and Crawley 2002;Colautti et al. 2004) might be responsible. In contrast, where deer decrease introduced plants, biotic resistance to plant invasion is a possible outcome (Levine et al. 2004;Parker and Hay 2005;Parker et al. 2006). Despite within-and between-site heterogeneity, the fact that deer had negative impacts on native plants and indirect, facilitative effects on the proportion of introduced plant abundance elucidates the overarching effects of deer on vegetation at a regional scale (Russell et al. 2001). That introduced and native species richness and abundance patterns are positively correlated (Table 2; see Supporting Information- Fig. S2) is consistent with research showing that introduced plant species invade 'hot spots' of diversity at large spatial scales (Stohlgren et al. 1999;Stark et al. 2006). Site characteristics, such as spatial heterogeneity in abiotic conditions (Davies et al. 2005), including land-use history and soil nutrients (Fraterrigo et al. 2006) across sites likely are responsible for the positive relationship between native and introduced plant richness and abundance.

Deer effects on floristic composition, diversity and community-level abundance
Deer do not directly impact introduced plant species richness (Table 5; Fig. 3) or abundance (Table 4; Fig. 2), which is evidence against our second prediction. These results are surprising, as many sites and large areas within the region and across the world currently are dominated by introduced species and also have high deer or other large herbivore populations (Waller and Alverson 1997;Rooney et al. 2004;Vavra et al. 2007). Such observations prompted our second prediction that greater richness and abundance of introduced plants would accompany deer access vs. deer exclusion. The fact that deer increase the proportion of cover of introduced plants appears to arise from the substantial decrease in the native flora imposed by deer. These results imply that the positive deer effect on the relative cover of introduced plants is caused indirectly by greater susceptibility of native vs. introduced plants to deer (however, see species-level results below). Deer have a markedly stronger negative effect on native species than on introduced species both in forest understories, as found in this work, and in an old field (DiTommaso et al. 2014). This result stands in contrast to reports that native and introduced species behave similarly in dynamic systems (Meiners 2007;Stromberg et al. 2009), albeit because of a native herbivore. The perspective that species be judged based on function and not on where they originated is gaining  Table 1 for additional site information). An asterisk between bars indicates a significant effect of deer; ns = not significant; n = sample size (number of plots). The number of plots was constrained in the proportion introduced richness and Shannon diversity analyses due to plots with zero vegetation [see Supporting Information-Text S1].
ground (Davis et al. 2011), yet our results show an important difference between native and introduced plants, namely their general susceptibility or response to herbivory, suggesting that native status has a deserved role in future research and in management decision-making. Our analyses suggest that declines in plant community diversity (McKinney and Lockwood 1999;Rooney et al. 2004) result more from deer herbivory than from the Table 5. Mixed model effects of white-tailed deer a) access/exclusion and b) population density on introduced, native and total plant species richness based on floristic composition data collected from deer-access (unfenced) and deer-exclusion (fenced) plots at 23 sites in eastcentral and north-eastern USA a . Proportion introduced plant species richness was evaluated based on the interaction of the deer effect with total species richness; a significant interaction indicates that the ratio of introduced to total plant species richness (i.e. proportion introduced) varies with the deer effect. For random effect results, see Supporting Information- Table S3. P values and LSmeans treatment test results are in bold print if significant at the alpha level α < 0.05. Our finding that deer increase the proportion of cover of introduced plants (Fig. 2) opposes the broadly observed biotic resistance pattern in which native herbivores reduce the relative abundance of introduced vegetation (Parker et al. 2006) as a result of differential palatability among introduced and native species (Parker and Hay 2005). This global meta-analysis found that native herbivores (e.g. insects, rodents and cervids) suppress introduced plants more than native plants. While informative for plant-herbivore interactions generally, such an extensive analysis is less likely to be predictive for a particular herbivore. Nonetheless, deer herbivory is a constant and important filter of regional species pools ) and could have a role in biotic resistance for certain introduced species (Maron and Vilà 2001), even preventing them from appearing in floristic census records.

Deer effects on individual introduced and native species' abundance
Overall, we found a few graminoid and herbaceous species are favoured in the presence of deer, while trees, shrubs, vines and many herbaceous species lose out (Tables 6 and 7). These findings are consistent with assessments of winning and losing species in Northern Wisconsin Wiegmann and Waller 2006) and globally (McKinney and Lockwood 1999). Our finding that many woody and herbaceous plant species are negatively impacted by deer contrasts with results from a meta-analysis showing that woody, but not herbaceous species are negatively impacted by deer (Habeck and Schultz 2015), a discrepancy possibly owing to publication bias detected in the meta-analysis. Our work clearly shows that deer facilitate several notorious invasive plants in east-central and northeastern USA, including A. altissima (tree-of-heaven), A. petiolata (garlic mustard) and M. vimineum (Japanese stilt-grass) (Tables 6 and 7). Positive effects of deer on A. petiolata and M. vimineum have been found in site-level experiments (Eschtruth and Battles 2009a, b;Knight et al. 2009;Kalisz et al. 2014;Dávalos et al. 2015a, b) and deer have been implicated in the establishment of A. altissima (Knapp and Canham 2000). The facilitative effect of deer on these species is likely due to their unpalatability relative to other plants. In deer preference trials, A. petiolata and M. vimineum were the least palatable of 15 introduced and native species (Averill et al. 2016). Ailanthus altissima is apparently also unpalatable (Forgione 1993), yet anecdotal evidence of browsing has been observed (K. L. Caraher, Hood College, pers. obs.) and thus the species' rapid growth rate (Knapp and Canham 2000) could outweigh herbivory. These results show how unpalatable plants can gain an apparent competitive advantage relative to palatable plants (Holt 1977), i.e. native plants, and become more strongly represented in the flora or even invasive (Senft et al. 1987;Keane and Crawley 2002;Royo and Carson 2006;Arcese et al. 2014).
While deer facilitated an increase in the abundance of several unpalatable invaders in unfenced plots, deer exclusion in fenced plots resulted in higher abundance of several other invaders, including L. japonica (Japanese honeysuckle) and R. multiflora (multiflora rose) (Table 7). Lonicera japonica, L. maackii, and R. multiflora were indicator species of deer-exclusion plots (Table 6), reinforcing previous findings (Shelton et al. 2014) and suggesting these fleshy-fruited species perform better where protected against deer browsing. Even if species perform well enough where deer occur to be considered invasive, they might perform better Figure 3. Effects of white-tailed deer access/exclusion on mean (±SE) (A) introduced (white shading) and native (black shading) plant species richness, (B) total plant species richness (includes unknown species) and (C) proportion introduced plant species richness in east-central and north-eastern USA. An asterisk between bars indicates a significant effect of deer; ns = not significant; n = sample size (number of plots). The number of plots was constrained in the proportion introduced richness analysis due to plots with zero vegetation [see Supporting Information-Text S1]. Table 6. Indicator species analysis results showing plant species and genera associated with deer access or with deer exclusion in east-central and north-eastern USA. Introduced species are in bold type. Indicator values range from 0 (no indication of association with treatment) to 100 (perfect indication) and were determined according to species' presence/absence (p/a) and the metric used to record abundance, density and/or cover. The number of plots and sites where each species was observed is included to indicate frequency and distribution across the 23 sites analysed. Results are arranged by deer access/exclusion, then by indicator value and then by P value; each species' results are listed together.

Species
Habit The native status based on genus alone is unknown. where deer are excluded. These findings might be an outcome of one or several processes, three of which are outlined here. (i) These species are palatable (Sheldon and Causey 1974;Ashton and Lerdau 2008;Averill et al. 2016) and, in heavily browsed plant communities, the most palatable species are the most susceptible to being consumed and reduced in abundance (Royo and Carson 2006 (Tables 6 and 7). Many other researchers (e.g. Abrams and Johnson 2012; Bressette et al. 2012;Nuttle et al. 2013;Abrams 2013;Owings et al., this issue) also report negative impacts of abundant deer on native tree species, implying that forest regeneration could be at risk. Shrubs, including L. benzoin, R. periclymenoides and V. acerifolium, also appear negatively influenced by deer (Tables 6 and 7), which jeopardizes organisms in other trophic levels that depend on forest understory shrub layers, e.g. birds (deCalesta 1994;McShea and Rappole 2000;Fuller 2001;Chollet and Martin 2013).

Site influences
Five sites were uninvaded by introduced plants. However, the native fern Dennstaedtia punctilobula, which is considered a native invasive plant (de la Cretaz and Kelty 1999), is dominant at two of the sites in north-eastern Pennsylvania (Long Run and Marienville). At the other three sites (Fernow, Monongahela and Zaleski), deer density estimates were considerably lower (~6 deer km −2 ) than the average across sites (mean = 35 deer km −2 ; median = 26 deer km −2 ) ( Table 1). At Fernow and Monongahela, deer were shown to increase herbaceous richness and abundance by reducing fast-growing early successional species (Royo et al. 2010a). Thus, sites without introduced invaders might instead have native invaders or low deer densities, which might be associated with increased biotic resistance to introduced plant invasion.
In addition to deer density, overstory species composition and duration of deer exclusion varied among sites (Table 1) and likely contributed to varying deer effect patterns at the site level [see Supporting Information-Tables S4 and S5]. While sites were not selected randomly, most were not established to study invasive plants and spanned a wide deer abundance gradient (Table 1). Furthermore, many concomitant and often interactive factors not limited to deer and invasive plants (e.g. forest successional age and proximity to centres of human activity, propagule pressure, resource availability, invasive earthworms, etc.) affect forest understory diversity (Abrams 1998;Pyšek et al. 2002) Baiser et al. 2008Eschtruth and Battles 2009b;Royo et al. 2010a;Martin and Baltzinger 2011;Fisichelli et al. 2013;Dávalos et al. 2014;Dobson and Blossey 2015;Forsyth et al. 2015;Nuzzo et al. 2015), yet were not included in analyses here. Including such factors in future work would improve understanding of community assembly and invasion processes. The influence of some site characteristics, including surrounding landscape structure and composition, on the relationship between deer and plant invasion is explored elsewhere (Averill 2014).
In floristic censuses, plant abundance is sometimes sampled using different metrics for different plant habits (Table 1), which presents issues for pooled or meta-analysis, such as requiring analysis and interpretation of multiple abundance metrics. The results reported here also show that using different abundance metrics can yield different results. For example, deer access increased the proportion of introduced plant cover, but not stem density (Table 4; Fig. 2), perhaps because herbivory influences cover more than stem density. Furthermore, determining total stem density or total cover, and therefore total vegetation abundance, depends on species being sampled in the same way. Total plant abundance is a useful metric for relating primary productivity to ecosystem functioning (Chapin et al. 2002), but cannot be calculated in data sets that use different abundance metrics for different plant habits, as was the case here.

Conclusions
This analysis deepens ecological understanding of some key factors in the invasion process. In this work, an abundant, native, large herbivore is shown to alter plant community composition, lower diversity, reduce native plant richness and abundance, and increase the relative cover of introduced plants. Unpalatable invasive plants seem to benefit under heavy herbivore pressure. While introduced plant invasion has been causally implicated in native plant decline (Wiegmann and Waller 2006), ruminant herbivory appears to be a key factor affecting both processes. Dominant native herbivores such as deer are important agents of ecosystem change as their presence (i) reduces native biodiversity and (ii) increases the relative abundance of introduced plants, two of the major drivers affecting modern plant communities and ecosystems (Hooper et al. 2012).

Supporting Information
The following additional information is available in the online version of this article-Text S1. Additional vegetation data processing methods.
Table S1. Sources and methods for deer density estimates. Table S2. Statistical results for effects of deer on unknown species. Table S3. Statistical results showing the influence of the random effects plot and site. Table S4. Species richness by plant native status, deer access/deer exclusion and site. Table S5. Vegetation abundance by metric used (per cent cover or stem density), plant native status, deer access/deer exclusion and site. Table S6. Taxa frequency by deer access/deer exclusion and plant native status, with sites of occurrence. Figure S1. Vegetation cover class categories used to estimate plant abundance across 15 sites. Figure S2. Relationships between introduced and native species richness and abundance.

Conflicts of Interest
None declared.