2.7. Metabolomics Data Analysis and Statistics

The analysis of metabolites was performed as described previously [27]. The data table was filtered by deleting metabolites of unknown identity. The data table containing peak intensities was uploaded to MetaboAnalyst 4.0 (https://www.metaboanalyst.ca/) (accessed on 31 July 2020), followed by log transformation and Pareto scaling for normalization of the data. From the same table, fold changes were calculated by comparing the mean of log transformed peak intensity values with respect to the control (i.e., LSHTN or AHTN/Control) using a mathematical formula (mean of LSHTN or AHTN/mean of control). The ‘p’ values were reported as generated from the ANOVA post hoc results, for the comparison of individual groups and the control. A metabolite was ‘significantly altered’ based on fold change >2 and ‘p’ value < 0.05.

Univariate analysis for all three datasets (Control, LSHTN, and AHTN) were compared with one-way ANOVA, followed by Tukey’s multiple comparison test (Control vs. LSHTN; Control vs. AHTN; LSHTN vs. AHTN). Multivariate analyses were performed within MetaboAnalyst and included PCA. Hierarchical clustering was performed with the hclust function in package stat, and the results were represented in the form of a dendrogram. Based on the ranking of features, a heatmap of all metabolites identified by ANOVA as being significantly different between groups was generated. Cutoffs based on rank were chosen over a specific significance threshold, since different statistical approaches will yield different absolute scores of significances, but most of the top-ranked features are expected to be consistent. For a heatmap, the top 50 features subjectively provide the ability to visualize both the trends and the variability of the features across samples.