Predicting pathogen susceptibility
As a demonstration of how these data can be applied, we used the data collected on pathogen impacts on invasive plants in Table 2 and applied a multivariate model to associate plant traits with pathogen effects (Figure 2). We compiled trait data from the TRY database for each invader and then selected all the traits that were available for at least 70% of species. We then ran principal component analysis (PCA) with imputed data for missing values using the ‘FactoMineR’ and ‘missMDA’ packages in R (Lê et al. 2008; Josse et al. 2016). We calculated the pathogen effect as the relative difference in performance (e.g. survival, reproductive output, growth) between the control and pathogen infected plants averaged across all available studies of each plant-pathogen combination (Table S1). For studies that tested multiple pathogens on a single invader only the most pathogenic was used in the analysis. The pathogen effect was included as a supplementary variable (i.e. not used in the calculation of the axes but projected over the trait axes) to the PCA. For three species in our dataset a quantitative pathogen effect was not available (Ambrosia artemisiifolia, Asclepias syriaca, and Centaurea diffusa ), so we added these as supplemental species to the PCA based on their trait values as a demonstration of how trait data can be used to predict pathogen susceptibility. From this framework, we can begin to determine the trait axes that are most associated with pathogen susceptibility. The first two axes of the PCA explained 75.9% of the total variation. Axis 1 was significantly associated with SLA (r = 0.91; P < 0.01) and leaf nitrogen (r = 0.66; P = 0.03). Axis 2 was significantly associated with leaf mass (r = 0.90; P < 0.01), vegetative height (r = 0.90; P < 0.01), and seed mass (r = 0.82; P < 0.01). Pathogen susceptibility was most strongly associated with Axis 2 of the PCA (r = 0.55; P = 0.08). Therefore, we can hypothesize that taller invasive plants with larger leaves and seeds are more susceptible to pathogens, and recommend methods to test this hypothesis by expanding and improving the model in the future (Box 2).
Increasing the availability of both trait data and pathogen susceptibility data will help to refine this model. There is a particular need for additional data on root traits, for which there were none available for enough species to use in the analysis, despite the recent creation of a root specific trait database incorporated into the TRY database (Iversenet al. 2017). It is also important to note that this model does not necessarily imply direct causation of the traits correlated with pathogen effect but instead is meant to be predictive.