The challenges of climate change on the consistency of the cocoa flavour quality

Objective tools to trace sustainable cocoa production are necessary, especially with regard to climatic changes and the political situation in the producing countries. Fingerprinting is a good approach for monitoring and authenticating food [1-2]. Food authentication is often based on the degree of similarity of the fingerprint between the sample under investigation and a representative reference. This process is known as food identitation [2-4] and its reliability depends on its correctness. Cocoa (Theobroma cacao L.) is a tropical perennial plant and one of the most important economic factors in the countries where it is grown. It is also a raw material of great economic relevance for various market fields, of which confectionery and functional foods and beverages (cocoa and chocolate derivatives) account for more than 60% of the market. For chocolate products, flavour is one of the key characteristics associated with product quality, along with brand and price. Furthermore, flavour quality identitation requires analytical platform able to produce detailed diagnostic profiles that can be correlated with sensory characteristics to be monitored for an objective evaluation in quality control (QC) specifications. In this study, HS-SPME-GC-MS in combination with machine learning tools was applied to the identification of a range of cocoa samples from different origins. This approach was applied to one hundred and sixty samples of cocoa beans and cocoa liquors. Untargeted fingerprinting and profiling approaches were tested for the information, discrimination and classification capabilities given by the volatilome of incoming beans and liquors. Machine learning tools PCA, NEAR indices and PLS-DA modelling were applied to the data set for flavour identitation of origins, to find flavour similarity of origins with the industry quality standard and to develop new blends to be compared with the flavour of standard references [2,5]. The results indicate a coherent, clear clustering of samples according to their origin with the two analytical strategies, both on raw beans and on cocoa liquors, albeit with differences at the molecular level. Predicting the classification of cocoa beans with the untargeted fingerprint on an external test set gave excellent results for beans with a classification rate of 100% and very good results for liquors (88%), despite the processing they underwent. Better results were obtained for targeted approaches with a classification rate above 92% (i.e., 92.86% for beans and 92.31% for liquors). The NEAR index, calculated using the OAVs of selected origin markers, showed Ecuador origin as to be the most similar to the reference in terms of taste. Through an OPLS-DA modelling of the chemical-specific origin marker and sensory similarity, blends with different proportions of Ecuador cocoa were created.