3.2 | Metabolomic comparison of the four species in fenced and unfenced conditions.
To compare the global metabolome of the four species under fenced and unfenced conditions, we conducted untargeted metabolomic analysis, which resulted in the identification of 2,333 metabolite features. A significant portion (84.3%) of these metabolite features were unique to each species: 950 metabolites in E. alatus (19.99%), 1,190 metabolites in R. multiflora (25.04%), 849 metabolites inN. sylvatica (17.87%) and 1,017 metabolites in L. benzoin(21.40%). While some metabolites were shared by two or more species (Appendix Table 1 ), only 27 metabolite features (1.1%) were commonly found in all the four species (Figure 2A ). Principal component analysis (PCA) conducted to assess the overall relatedness of the samples with respect to the 2,333 metabolite features generated four distinct clusters, each cluster corresponding to a species; the first two principal components PC1 and PC2 explained 25.4% and 23% of the total variability among the samples, respectively (Figure 2B ). The same pattern of relatedness among the samples is also observed on the dendrogram constructed using Pearson correlation and Ward’s clustering algorithm. The dendrogram grouped the two nonindigenous species (R. multiflora and E. alatus ) in the same clade, indicating that the nonindigenous species are the closest to each other with respect to their metabolite profiles. For most species, samples collected from the fenced (indicated by red color) and unfenced (indicated by green color) plants did not form distinct subclades on the dendrogram; however, N. sylvatica samples were grouped into two distinct subclades that correspond with the presence/absence of fence (Figure 2C ).
Using the metabolite features, we conducted functional analysis, which putatively assigned the metabolite features to 61 metabolic pathways; 14 (22.9%) of these metabolic pathways were found in all the four species while a few were shared by two or three species. Among the pairwise comparisons, the two indigenous species (L. benzoin and N. sylvatica ) shared the largest numbers of predicted metabolic pathways (30 metabolic pathways, 49.2%) (Figure 2D ). The complete list of predicted metabolic pathways that are unique to each species or shared among any of them is given in Appendix Table 2 .
To identify metabolites that differentiate the samples based on species and/or treatment (fence vs. no fence), we conducted partial least squares-discriminant analysis (PLS-DA), which resulted in two main clusters corresponding to treatment (presence or absence of fence) and four subclusters within each of these clusters that correspond to species (Figure 3A ). The accumulation of the top 15 most important features that contributed to the separation of the samples into the two main PLS-DA clusters is influenced mainly by the presence or absence of fences (Figure 3B ). However, close inspection of the relative accumulation of some of the top metabolite features (m/z: 149.0597; 176.8251; 188.0995; 215.0810 and 187.0967) indicates that their abundance was influenced both by species and/or treatment (Figure 3C ).
Among the four species and fencing treatments, the accumulation of metabolite 1 (m/z = 149.0597) was significantly lower in fenced N. sylvatica and fenced and unfenced L. benzoin plants (P < 0.05; Tukey HSD). Within species, we observed no significant differences in the accumulation of this metabolite between the fenced and unfenced samples for R. multiflora , L. benzoin andE. alatus , but unfenced N. sylvatica accumulated significantly more of this metabolite (P = 0.03; ANOVA). Fenced N. sylvatica plants also accumulated significantly lower (P < 0.05; Tukey HSD) amount of metabolite 2 (m/z = 176.8251) than the other species, while unfenced L. benzoin samples had significantly more (P < 0.05; Tukey HSD). Comparing fenced and unfenced plants within species, we found that the accumulation of metabolite 2 increased significantly (P < 0.05; Tukey HSD) in N. sylvatica andL. benzoin samples from unfenced plots. On the other hand, the accumulation of metabolite 3 (m/z = 188.0995) did not vary much among or within the species in both fenced and unfenced conditions. The exception was N. sylvatica , which, under fenced conditions, accumulated a significantly lower amount than all other species except for unfencedL. benzoin (P < 0.05; Tukey HSD). With respect to metabolite 4 (m/z = 215.0995), fenced N. sylvatica , L. benzoin and E. alatus plants accumulated significantly lower amounts compared to R. multiflora (P < 0.05; Tukey HSD), whereas in unfenced conditions, L. benzoin and E. alatus accumulated significantly lower amounts (P < 0.05; Tukey HSD). Within species, comparison of fenced and unfenced samples revealed a difference only for N. sylvatica , with significantly more metabolite 4 in the unfenced plants (P < 0.05; Tukey HSD). Unfenced R. multiflora , N. sylvatica , and E. alatus plants accumulated significantly lower amounts of metabolite 5 (m/z = 187.0967) compared to L. benzoin (P < 0.05; Tukey HSD). Under unfenced conditions, the accumulation of this metabolite feature was not significantly different among R. multiflora , N. sylvatica and E. alatus , but L. benzoin plants accumulated significantly more of this metabolite (P < 0.05; Tukey HSD). Comparing fenced and unfenced plants within species, we observed that unfenced R. multiflora , N. sylvatica and E. alatus plants had significantly reduced accumulation of metabolite 5 (P < 0.05; Tukey HSD) (Figure 3C ).