Environmental filtering determines patterns of host trait covariation and the ability of host traits to predict pathogen infection

Background/Question/Methods

The Leaf Economics Spectrum (LES) describes covariation among leaf traits and can predict host susceptibility to pathogen infection. Support for LES, however, comes mostly from communities where environmental filtering shaped the trait pool. We hypothesized that, in the absence of filtering (e.g., rapid environmental change), patterns of leaf trait covariation would deviate from the LES, causing trait-based models of plant susceptibility to pathogen infection to become less accurate. We experimentally tested this prediction in the greenhouse, using 23 grass species and nitrogen addition to create three scenarios of environmental filtering: (1) unfiltered: perennials at high nitrogen and annuals at low nitrogen, such that, relative to the field, all individuals were in novel conditions; (2) semi-filtered: perennials and annuals at low and high nitrogen, so half the individuals were in novel conditions while half were in field conditions; and (3) completely filtered: perennials at low nitrogen and annuals at high nitrogen, so all individuals were in field conditions. In each scenario, we measured several leaf traits and used principal components analysis to quantify support for the LES. All individuals were then inoculated with a virus, and we compared the ability of host traits to correctly predict successful infection across scenarios.

Results/Conclusions

Support for the LES increased with the strength of environmental filtering: Principal Component 1 (PC1), our measure of the LES, explained 39.5% of the trait variation in the unfiltered scenario, 45.9% of the trait variation in the semi-filtered scenario, and 59.4% of the trait variation in the completely filtered scenario. The ability of host traits to predict susceptibility to virus infection also increased with environmental filtering. Moving from the unfiltered to the completely filtered scenario, model accuracy increased by 10.6%, regardless of whether PC1 or specific traits were used to predict infection. In the completely filtered scenario, PC1 predicted infection with 77.5% accuracy and specific traits predicted infection with 80.4% accuracy. Our results suggest that environmental filtering plays a fundamental role in generating multivariate axes of trait covariation, and that trait-based models of ecological processes may fail when communities have not been filtered by the current environment, for example, following rapid environmental change.