Bioprocess monitoring and chemometric modelling of wine fermentations

ENGLISH ABSTRACT: Wine fermentation is a continuously changing biological process whereby the raw product, grape juice is transformed into a high value product wine. In an ideal situation the fermentation kinetics of batch fermentations should follow the same trend over time. This however is not the case in industrial wine fermentations where significant batch-to-batch variation is present. The time trajectories of fermentation processes are therefore often unpredictable in absolute terms. The monitoring of substrate (sugar) and product concentrations (ethanol) as well as other quality parameters during a wine fermentation, is therefore of extreme importance to ensure effective control and management of wine fermentation processes. Conventional methods for fermentation monitoring are however costly, time consuming and often unreliable. For these reasons the modern wine industry requires rapid, reliable, non-destructive monitoring techniques which would meet the criteria of providing critical real-time process information that is displayed in easily interpretable graphical format, in order to ensure the highest quality and continuous consistency throughout all the stages of a process. This research study in particular, addressed the current need for alternative fermentation monitoring strategies that meet these criteria, by evaluating the potential use of spectroscopy as an analytical technique for fermentation monitoring. The overall objective of this study was to use chemometric modelling of information obtained by Fourier transform mid-infrared (FT-MIR) and near-infrared (FT-NIR) spectroscopy, to quantitatively and qualitatively monitor both alcoholic (AF) and malolactic fermentation (MLF) processes. Towards this objective 11 batch fermentations elaborated with Oenococcus oeni and Lactobacillus plantarum strains in respectively a co-inoculation and sequential inoculation scenario, were sampled and analysed at regular time intervals with FT-MIR and FT-NIR spectroscopy and enzymatic reference methods during 2011. Samples were also analysed by gas chromatography flame ionisation detection (GC-FID) and mass spectrometry (GC-MS) at two critical stages during fermentation, namely 50% completion of MLF and 100% completion of MLF, in order to obtain a profile of the evolution of the aroma compounds associated with each inoculation scenario. Three clearly defined research objectives were set for this project. The first objective involved the expansion of the existing quantitative platform for fermentation monitoring. Towards the outcomes of this objective, partial least squares (PLS) calibration models for prediction of malic acid and lactic acid in fermenting must and wines elaborated in our study, were established, based on the MIR and NIR spectra. The models showed excellent predictive abilities in independent test set validation. This outcome made a significant contribution to our existing PLS calibration capacity, particularly towards monitoring of MLF. Quantitative data obtained with the PLS models were also used to graphically project the rate of AF in the different batches, by non-linear fitted regression plots that easily visualised the overall patterns of sugar and ethanol metabolism in the different fermentations. The second research objective involved the qualitative monitoring of fermentations. This approach used FT-MIR and FT-NIR spectra together with chemometrics to identify trends between the different fermentation treatments. Principal component analysis (PCA) clearly projected the time trend from the onset of fermentation, through AF and MLF. No unique bacterial trend was however observed with spectroscopy. These results illustrate the potential of these techniques to be used for modelling of fermentations in industrial situations, through providing critical information about the evolution of the process. Furthermore, these techniques provide tools for identifying problematic and deviating fermentations. A spectral conformity test based on simple calculations of the standard deviation between the absorbance at each recorded wavenumber in the spectra, further confirmed identification of the critical fermentation stages. This technique by-passes the need for spectral interpretation and is a very useful addition, particularly from the industry perspective, to the portfolio of methods established in this study. The third research objective adressed the need to evaluate the possibility to discriminate between the different process stages and LAB treatments using univariate (ANOVA) and multivariate chemometric techniques such as PCA, PLS discriminant analysis (PLS-DA) and Soft Independent Modelling of Class Analogy (SIMCA) for possible future interpretative and classification purposes. The exploratory tool of PCA was used to investigate the similarities and differences between the chemical footprints of the different treatments. PCA showed clear differentiation between the two process stages using chemical quantified data. Differentiation between the LAB treatments was visible with PCA, showing a more prominent separation at 50% completion of MLF. Furthermore the ability of spectroscopy for potential classification was shown using PLS-DA and SIMCA. This profiling study can be seen as a preliminary study setting the ground work for further in-depth research into the profiling of different LAB treatments and inoculation strategies.