Figure 2

PLS modeling of metabolomic data. PLS plots correlating metabolic profiles with dose of PB at day 14 using ¹H-NMR (A), GC-MS analysis of the aqueous extract (B), and GC-MS analysis of the lipid extract (C). In addition to the use of Q² values, models were validated to prevent over fitting by leaving samples out of the model building stage and then predicting their dose of PB (D) (mean ± standard deviation of 5 iterations) and also by using the validate function in SIMCA-P+ with 200 iterations (E). In this approach the actual model is compared with models formed where the Y value (in this case dose) is randomly permuted to produce models where there should be no correlation between the X-block and Y-block matrices. As demonstrated in panel E, the decreases in R² and Q² as the dose is randomly permuted is indicative of a robust model. PLS analysis of MAS ¹H-NMR data separated the control and low dose from the mid and high dose groups (F). The corresponding loadings column plot (G i) and assigned spectra (G ii) showed that exposure to PB increased lipids and decreased sugars. The ratio of total lipid to total carbohydrate was increased in the mid and high dose groups compared to the control group (H). Plot show average ± standard deviation (n = 5) with significant changes as determined by ANOVA with Dunnett's post test for multiple comparisons labelled ** (p < 0.01). Key: Control (filled square) 50 ppm (open circle) 500 ppm (filled triangle) 1000 ppm (open circle).