Untargeted Metabolomics Coupled With Machine Learning Unravels Crop-Specific Versus Generalized Effects of Biostimulants in Cucumber, Pepper, and Tomato.

In recent years, agricultural practices have shifted toward sustainability, aiming to reduce the use of agrochemicals and rely more on bio-based solutions. However, the effectiveness of these latter suffers from inconsistency. Understanding how different crops respond to biostimulants, instead of referring to a specific trial or crop, is a challenge in this field. This study attempts to identify common metabolite signatures associated with different commercial biostimulants across three crops, moving from trial-specific to more generalized effects in horticultural crops. To this aim, advanced metabolomics data integration and supervised statistical methods have been used. Advanced multivariate analyses included analysis of variance (ANOVA)-multiblock orthogonal partial least squares (AMOPLS), and Data Integration Analysis for Biomarker discovery using Latent variable approaches for Omics studies (DIABLO). HCA and AMOPLS revealed differences in metabolic profiles among the biostimulant treatments, while confirming crop-specific responses. Data integration indicated that three metabolites, betaine, N-caffeoylputrescine, and 2-amino-4-hydroxypyrimidine-5-carboxylic acid, were consistently modulated across all three crops treated by the multi-component biostimulant containing osmolytes and zeatin. Notably, these metabolites are known to be involved in plant growth and adaptation to different abiotic stresses. Overall, the applied analytical approach enabled the identification of putative markers within complex metabolic datasets that included different crop species. The use of independent validation methods increases confidence in these markers and supports the integration of complementary datasets in biostimulant studies.