Reply to Evens and Niedz: Multivariate ionomics models are robustly validated

We strongly disagree with Evens and Niedz (1) that our article (2) is “fundamentally flawed,” for several reasons:

1. Our goal was to establish the existence of ionomic signatures in Arabidopsis shoots and to use these signatures to predict a plant's Fe and P nutritional status. Contrary to Evens and Niedz (1), our goal was not to understand the relative contributions of each predictor. Instead, we aimed at achieving a high predictive ability and validating the predictions using growth conditions established to alter the expression of known Fe- and P-responsive genes.

2. The model used for prediction was validated against an independent dataset based on the Fe and P nutritional status of the plants. Contrary to Evens and Niedz, these were not the ionomics signatures. We utilized this validated model to identify known Fe-deficient mutants frd3 and frd1 in a screen of ionomic data from 880 Arabidopsis mutants. Evens and Niedz provide no comments on this important test of the model. Evens and Niedz characterize this rigorous process of model validation as follows: “Subsequent experiments to validate the model were insufficient and merely served as unverified negative controls.”

3. Evens and Niedz raise the question of potential confounding effects. Our article shows that alterations of the concentration of several mineral nutrients in the fertilization solution do not cause the model to incorrectly predict the Fe or P nutritional conditions under which the plants were grown. We understand that dropping salts from the nutrient solution changes anion and cation concentrations. However, our conclusion that ionomic signatures provide high-quality predictions of the Fe and P nutritional status of Arabidopsis remains valid.

4. Evens and Niedz suggest an experiment to establish the contribution of each element in the ionomic signature to the response. This is an excellent idea, but it is beyond the scope of this study.