LCMS data processing and analysis
Mass peak picking and alignment were performed using Metalign software (Lommen 2009). Mass features in the resulting peak list were considered as a real signal if they were detected with an intensity of more than 3 times the noise value and in 3 out of the 4 biological replicates of at least one treatment. Mass features originating from the same metabolites were subsequently reconstituted based on their similar retention window and their intensity correlation across all measured samples, using MSClust software (Tikunov et al. 2012). This resulted in the relative intensity of 725 putative metabolites in Arabidopsis, 868 in Artemisia and 1908 in broccoli detected in positive and negative ionization mode, in which the metabolite abundance was represented by the Measured Ion Count (MIC), i.e the sum of the corrected intensity values of all mass features ions within the corresponding cluster. ANOVA and a threshold of at least a 2-fold change were applied to pinpoint compounds that were significantly different between rhizobacteria-treated and control samples. Log transformation and scaling of the data was performed in GeneMaths XT 1.6 (www.applied-maths.com). Transformed and scaled values were used for hierarchical cluster analysis using Pearson’s correlation coefficient and Unweighted Pair Group Method with Arithmetic Mean (UPGMA).
Annotation of differential metabolites was performed after manually identifying the putative molecular ions within the clustered masses. In-house databases were used to annotate metabolites detected in Arabidopsis and Broccoli by considering the observed accurate masses and retention times of the molecular ions. If selected compounds were not yet present in this experimentally obtained database, detected masses were matched with compound libraries, including Metabolomics Japan (www.metabolomics.jp), the Dictionary of Natural Products (www.dnp.chemnetbase.com), KNApSAcK (www.knapsackfamily.com), and Metlin (www.metlin.scripps.edu) using a maximum mass deviation of 5 ppm. To annotate metabolites detected in Artemisia, we used the online Magma (Ridder et al. 2013) in combination with the above-mentioned publicly available databases.