Bioprocess monitoring and chemometric modelling of wine fermentations

[摘要] 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 fermentationkinetics of batch fermentations should follow the same trend over time. This however is not thecase in industrial wine fermentations where significant batch-to-batch variation is present. Thetime trajectories of fermentation processes are therefore often unpredictable in absolute terms.The monitoring of substrate (sugar) and product concentrations (ethanol) as well as otherquality parameters during a wine fermentation, is therefore of extreme importance to ensureeffective control and management of wine fermentation processes. Conventional methods forfermentation monitoring are however costly, time consuming and often unreliable. For thesereasons the modern wine industry requires rapid, reliable, non-destructive monitoringtechniques which would meet the criteria of providing critical real-time process information thatis displayed in easily interpretable graphical format, in order to ensure the highest quality andcontinuous consistency throughout all the stages of a process. This research study in particular, addressed the current need for alternative fermentationmonitoring strategies that meet these criteria, by evaluating the potential use of spectroscopy asan analytical technique for fermentation monitoring. The overall objective of this study was touse chemometric modelling of information obtained by Fourier transform mid-infrared (FT-MIR)and near-infrared (FT-NIR) spectroscopy, to quantitatively and qualitatively monitor bothalcoholic (AF) and malolactic fermentation (MLF) processes. Towards this objective 11 batchfermentations elaborated with Oenococcus oeni and Lactobacillus plantarum strains inrespectively a co-inoculation and sequential inoculation scenario, were sampled and analysedat regular time intervals with FT-MIR and FT-NIR spectroscopy and enzymatic referencemethods 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 ofthe aroma compounds associated with each inoculation scenario.Three clearly defined research objectives were set for this project. The first objectiveinvolved the expansion of the existing quantitative platform for fermentation monitoring.Towards the outcomes of this objective, partial least squares (PLS) calibration models forprediction 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 predictiveabilities in independent test set validation. This outcome made a significant contribution to ourexisting PLS calibration capacity, particularly towards monitoring of MLF. Quantitative dataobtained 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 patternsof sugar and ethanol metabolism in the different fermentations.The second research objective involved the qualitative monitoring of fermentations. Thisapproach used FT-MIR and FT-NIR spectra together with chemometrics to identify trendsbetween the different fermentation treatments. Principal component analysis (PCA) clearlyprojected the time trend from the onset of fermentation, through AF and MLF. No uniquebacterial trend was however observed with spectroscopy. These results illustrate the potentialof these techniques to be used for modelling of fermentations in industrial situations, throughproviding critical information about the evolution of the process. Furthermore, these techniquesprovide tools for identifying problematic and deviating fermentations. A spectral conformity testbased on simple calculations of the standard deviation between the absorbance at eachrecorded wavenumber in the spectra, further confirmed identification of the critical fermentationstages. This technique by-passes the need for spectral interpretation and is a very usefuladdition, particularly from the industry perspective, to the portfolio of methods established in thisstudy.The third research objective adressed the need to evaluate the possibility to discriminatebetween the different process stages and LAB treatments using univariate (ANOVA) andmultivariate chemometric techniques such as PCA, PLS discriminant analysis (PLS-DA) andSoft Independent Modelling of Class Analogy (SIMCA) for possible future interpretative andclassification purposes. The exploratory tool of PCA was used to investigate the similarities anddifferences between the chemical footprints of the different treatments. PCA showed cleardifferentiation between the two process stages using chemical quantified data. Differentiationbetween the LAB treatments was visible with PCA, showing a more prominent separation at50% completion of MLF. Furthermore the ability of spectroscopy for potential classification wasshown using PLS-DA and SIMCA. This profiling study can be seen as a preliminary studysetting the ground work for further in-depth research into the profiling of different LABtreatments and inoculation strategies.

[发布机构] Stellenbosch University