Statistical analysis
All data from targeted analyses were tested for normality (Kolmogorov–Smirnov test) and for homogeneity of variance (Leven median test). Thereafter, three-way ANOVA was used to investigate the effect of treatments (control, drought, heat, and drought+heat), root type (primary and seminal), root zones (apical and sub-apical) and their interactions. The means were separated by the Tukey’s honest significant difference (HSD) test (p<0.05).
The metabolomics-based analyses were performed according to García Pérezet al. (2021). Briefly, raw mass features were elaborated using the software Agilent Mass Profiler Professional B.12.06 for normalization and baselining, and then the multivariate unsupervised hierarchical cluster analysis (HCA) was performed (Euclidean distance, Ward’s linkage rule) to describe similarities and dissimilarities among treatments from a fold-change based heat map. A Volcano analysis (α = 0.05; fold change, FC > 1.2) was performed to identify compounds varying significantly between treatments and control. In addition, supervised modeling by Orthogonal Projection to Latent Structure Discriminant Analysis (OPLS-DA) was performed using SIMCA 16 software tool (Umetrics, Sweden). The Variable Importance in Projection (VIP) analysis was performed to further identify markers responsible of the discrimination. The obtained multivariate model was then cross-validated by Cross Validation-Analysis of Variance (CV-ANOVA, α < 0.01) and its fitness and prediction were evaluated by R2Y and Q2Y parameters, respectively.