Application of Nonparametric Multivariate Analyses to the Authentication of Wild and Farmed European Sea Bass (Dicentrarchus labrax). Results of a Survey on Fish Sampled in the Retail Trade

The aim of this study was to apply biometric measurements and analyses of proximate composition, fatty acid composition, and ratios of stable isotopes of carbon (delta C-13) and nitrogen (delta N-15) in muscle tissue to reliably differentiate between wild and farmed European sea bass (Dicentrarchus labrax). Farmed (n = 20) and wild (n = 19) European sea bass were purchased between March and May 2008 and used as standard samples. In the same months, a survey was conducted to evaluate the truthfulness of the statements on the labels of European sea bass sold in retail markets (declared farmed n = 34 and declared wild n = 33). In addition, data from the literature (reference) were employed to build the profile type of wild and farmed European sea bass. Primarily, an exploration and comparison of the analytical data of the standard data set based on principal component analysis and permutation test were performed. Afterward, an inferential statistical approach based on nonparametric combination test methodology (NPC) was applied on standard samples to check its suitability in discriminating the production method. This multivariate statistical analysis selected 30 variables on a total of 36 available. The validation of standard fish data set was accomplished by a novel nonparametric rank-based method according to profile type (just 1 misclassification over 39 samples). Both the NPC test and nonparametric rank-based method were then applied to survey fishes using the selected variables with the aim to classify the individual European sea bass as "true farmed" or "true wild". The former test segregated 10 fishes over 33 declared wild, whereas the results obtained by the nonparametric rank-based method showed that 11 of 33 declared wild European sea bass samples could be unquestionably attributed to the wild cluster. Moreover, considering the comparative contribution of profile type, a few surveyed farmed samples were ascribed to the wild cluster.