Rapid dioxin assessment in fish products by fatty acid pattern recognition

Abstract

A novel, cost- and time-effective dioxin screening method relying on fatty acid profile was developed for fish products. The method is based on multivariate covariance between fatty acid composition and dioxin. A dioxin range varying from 1.1 to 47.1 ng TEQ-WHO kg fat⁻¹ was investigated using 64 fish meal samples. An optimal multivariate dioxin prediction model was developed based on reduction from the original 32 to 13 fatty acids, thus increasing the parsimony and the robustness of the model. The model obtained with three partial least squares regression (PLS) components included the following 13 fatty acids: C14:1 n-5, C16:4 n-1, C18:1 n-9, C18:2 n-6, C18:3 n-6, C18:3 n-3, C20:0, C20:1 n-9, C20:4 n-6, C20:3 n-3, C22:1 n-7, C22:6 n-3, C24:1 n-9. Considering the whole investigated dioxin range, the performance of the PLS model based upon full cross-validation yielded a correlation of 0.90 (r²) and a prediction error of 3.31 ng PCDD/F TEQ-WHO kg fat⁻¹. A submodel of samples in the lower dioxin range 1 to 15 ng PCDD/F TEQ-WHO kg fat⁻¹ returned a r² of 0.88 and an error of 1.85 ng PCDD/F TEQ-WHO kg fat⁻¹.