Understanding odour thresholds: A Quantitative Structure-Property Relationship study of the role of the molecular structure

International audienceWithin the huge variety of aroma compounds identified, there are some to which humans are more sensitive than others [1]. This characteristic is usually associated with the molecule's detection threshold (DT). However, the relationship between a molecule's structure and its DT has not a straightforward answer. The understanding of odour perception requires being able to link the molecular properties of aroma compounds to their DT values. In that way, the Quantitative Structure-Properties Relationship (QSPR) approach is especially appropriate. Several studies have proposed models that relate DT and some structural features, such as the molecule's size and the distribution of positive charges, but this has been done essentially in sets of alcohols or pyrazines [2]. In this context, the aim of this work was to understand the role of the chemical structure of aroma compounds by means of a QSPR model that included a vast variety of chemical functions. In this study, we have collected from the literature the orthonasal DT values in water of 407 molecules belonging to a variety of chemical families representing the diversity of odour molecules commonly found in foods and beverages: alcohols, aldehydes, pyrazines, sulphides, among many others. Then, we have used the Biovia Discovery Studio package (Dassault Systèmes, San Diego, CA, USA) to calculate the values of 106 molecular descriptors. The original set was split into a training set (327 molecules) and a test set (80 molecules). Several QSPR models were generated using Genetic Functional Analysis (GFA) being the descriptors as independent variables and ln DT as the dependent variable. The optimisation was done by maximisation of the regression coefficient r². The best models involved at least 5 terms (r² 0.46, RMS Residual Error 3.09, LOF 37.8, F 68.5, p-value 1.254.10-51 a=0.05). By increasing to 6 terms allowed to improve the quality of the model (r² 0.50, RMS Residual Error 2.97, LOF 34.9, F 67.9, p-value 5.565.10-58 a=0.05). Four variables that encode the role of hydrophobic features of the molecules were found in both models: ES_Count_dsCH (number of CH methine groups), Jurs_RASA (relative apolar surface areas), Molecular_Weight, and V_DIST_equ (topological descriptor related to the environment of the covalent linkages). Besides, the descriptors encoding the negative charges of the molecules differed depending on the model: Dipole_mag was found to have a contribution in the 5-variable model, whereas ES_Count_ssO (number of non-carbonyl oxygens) and Jurs_RNCS (relative negatively charged surface areas) were involved in the 6-variable model. The analysis of the standardised coefficients put forward the role of the skeleton of the molecules (V_DIST_equ, positive contribution), which partially counterbalanced the global negative contribution of the Molecular_Weight. In conclusion, we have developed QSPR models that explained the contribution of different molecular characteristics to DT in water for a set of more than 400 molecules. These models will help understand odour perception and predict orthonasal DT to a certain degree. 1. Keller A et al. BMC Neurosci., 2016, 17, 55. 2. Edwards PA et al. Chem. Senses, 1991, 16(5),447-465