Direct analysis of coffee aroma compounds with Proton Transfer Reaction-Time of Flight-Mass Spectrometry: traceability, perceived quality and processing

This thesis involves developing fast, rapid, and non-invasive headspace and nosespace analysis techniques based on PTR-ToF-MS coupled with an autosampler and tailored data analysis tools. The investigated case studies are related to coffee flavour also in connection with different technological and fundamental aspects as roasting and origin. In a first study, an automated headspace sampling method was developed by combining a GC autosampler to PTR-ToF-MS to analyse the aroma profiles of three monoorigin (Brazil, Ethiopia and Guatemala) roasted and ground Coffea arabica samples from different batches. Unsupervised and supervised multivariate data analysis techniques were applied for data exploration and to classify coffees according to origin. Coffee samples were successfully separated according to origin by unsupervised methods (Principal Component Analysis, PCA). This separation was confirmed with Partial Least Square Regression-Discriminant Analysis (PLS-DA). The samples of one batch could be used as training set to predict geographic origin of the samples of the other batch, suggesting the possibility to predict further batches in coffee production by means of the same approach. As a follow-up study, the developed headspace sampling method was applied to analyse six roasted Coffea arabica coffees, both brew and powder, of different geographical origins (Brazil, Ethiopia, Guatemala, Costa Rica, Colombia, and India). For the first time, the volatile compounds released from coffee were analysed with PTR-ToF-MS in Switching Reagent Ion (SRI) mode by using different ionization agents: H3O + , NO+ and O2 + . Significant differences were found among volatile concentrations for the different origins both for powders and brews, in particular high concentrations of terpenes for Ethiopia, sulphur compounds for Colombia and thiazoles for Brazil and India. Effective classification models have been set for the different ionization modes and data fusion of the data obtained by different reagent ions further reduced the classification errors. The next project was the development and application of an experimental protocol to monitor the volatile compounds released from single coffee beans at different stages of roasting. A laboratory scale oven was used to roast the green coffee beans (Coffeea arabica) from different geographical origins (Brazil, Guatemala and Ethiopia) by sampling every one min up to 25 min. Two batches of one coffee origin were selected and at each time point, 3 coffee beans were roasted. This resulted in volatile profiling of a large sample-set: 468 coffee beans (3 origins x 2 batches x 3 replicates x 26 time points). The weight losses due to roasting process were calculated for each coffee bean and time point. The effect of coffee geographical origin was reflected on the final weight losses and therefore volatile compounds formation. We observed a reduction in the amount of terpene fragments and an increase in heat induced volatile compounds. Clear origin signatures, which are in agreement with previous findings, were observed especially in the concentration of the volatiles released. Depending on the phase of roasting, some mass peaks were released earlier than the others and vice versa (e.g. m/z 82.065 around 6th min and m/z 101.060 around 10th min). Finally, nosespace analysis (NS) was performed via simultaneous combination of PTR-ToFMS with a dynamic sensory method called “Temporal Dominance of Sensations (TDS)” to gain insight about in-nose volatile release with the perceived aroma during espresso coffee drinking. One goal was to combine real-time instrumental and sensory methods and the second goal was to investigate the impact of roasting degree and sugar addition on aroma release and perception. A set of four coffee samples, two roasting degrees and two sugar levels, has been used for both sensory and instrumental measurements. Eighteen trained judges joined the study and they selected the dominant sensations among a 9-attribute list (sweet, sour, bitter, astringent, roasted, burnt, caramel, nutty and vegetal). The volatile compounds released in the nose of judges were monitored by NS analysis. A significant effect of roasting was observed with both techniques. More compounds and in larger quantity were released when increasing roasting degree, which was described in sensory perception as a greater dominance of the attributes burnt, roasted, astringent and bitter. Sugar addition did not significantly affect the aroma release of volatile compounds as demonstrated by the NS profiles of judges while changing completely the way the coffee was perceived by TDS. As expected, sweet taste became dominant over bitter and sour but it increased global flavour complexity with caramel and nutty notes via reducing the roasted or burnt notes. This result emphasized the presence of taste–smell perceptual interactions, due to congruence effect between sweet taste and some flavours of coffee, and the potential of combining dynamic methods to study the interactions. Besides, the treatment of NS data using clustering methods revealed two different release behaviours, which permitted identifying potential volatile compounds as TDS markers. The developed methods are of general interest and have been tested also in the case of other food matrixes such as tea. The automated headspace sampling method was applied to measure the volatile compounds emitted from black (n=63, from 12 different countries) and green tea (n=38, from 9 different countries) leaves and their infusions. Black and green teas were correctly classified by the volatile compounds emitted from tea leaves and their infusions independent from their geographical origins. A fixed time and temperature was applied for preparing black and green tea infusions to reduce the possible variability. Results showed that the release of volatile compounds is higher for tea samples which have smaller leaves or contain broken leaves as compared with the tea samples with bigger leaves, in particular after tea infusion. Depending on the processing method, teas produced in different countries have diverse appearance and flavour. For this reason we built classification models to investigate the possibility to link tea aroma with geographical origin. Results provided a good separation of tea origins, classification errors being mostly between countries geographically close to each other. These findings suggested that a better discrimination of tea samples might have been achieved if teas were classified according to production region rather than just country of origin. The promising outcomes of this thesis suggest PTR-ToF-MS as a successful tool for monitoring the release of volatile compounds from different aspects in coffee flavour science from coffee bean roasting to coffee drinking and as well as from product discrimination to traceability