On the Prediction of Polyphenol Properties

The increasing population demands increasing food supply with the consequent increase in food waste. Moreover, not only the quantity of the demanded food increases but so does the quality. This situation poses an interesting challenge to the food industry that is moving towards the revalorisation of the food waste and, as a part of it, towards the recovery of nutritional compounds that provide health benefits (nutraceuticals) still present in those streams. However, those streams possess very complex chemical and physical characteristics, namely multiple phases, ionic and hydrophobic interactions, etc. Moreover, the concentration of these nutraceuticals tends to be low, making regular processes costly and inefficient if applied to these cases. Therefore, it is extremely important to have models that describe the behaviour of the different components of the system to be able to design and develop a feasible process to recover these molecules. The objective of this research, therefore, was to apply or develop and validate models that could predict important properties for the design of a downstream process in each of the cases. The main focus of the work was on green tea catechins and soy isoflavones. From the beginning, their molecular similarities made clear that Group Contribution (GC) models were the obvious way to reduce complexity. Therefore solubility experiments of the four main catechins (Chapter 2) were performed in order to validate a well know GC activity coefficient model such as Mod. UNIFAC (amongst other models). The validation proved unsuccessful since Mod. UNIFAC did not give good predictions for the interaction between the molecules and water. Activity coefficients are a key property in design since they account for the non-idealities in the system which made the development of a suitable activity coefficient an important part of the research. An extensive bibliographic work was done and solubility values of flavonoids (a class of polyphenols) were gathered in a database that was used for the regression of new interaction parameters for the Mod-UNIFAC model with a new definition of groups more applied to polyphenols. Those parameters were regressed and validated in what is called Modified PolyPhenols-UNIFAC or MPP-UNIFAC (Chapter 3). After having a good understanding of the behaviour of the polyphenols in solution, the next logical step was to model their behaviour in equilibrium with a resin since Solid-Liquid Extraction is one of the most used unit operations. A set of food-grade resins was tested for their suitability to separate catechins from caffeine using green tea as a starting material (Chapter 4). A model-based Design of Experiments was used in order to perform the needed experiments to minimize the uncertainty of the regressed parameters which were further used as criteria for a resin selection. A similar approach was used for the okara. However since okara is a solid matrix, a pretreatment was needed in order to bring the polyphenols to solution. This pretreatment consisted of a solvent extraction with different ratios of ethanol and water. After testing the behaviour of the different extracts with the resin, it was decided to further study the water extract since it maximized the overall yield. Adsorption isotherms were measured (Chapter 5) showing an almost linear behaviour, typical of small concentrations. In some cases it was observed a non-linearity that was hypothesized as a bilayer adsorption of polyphenols onto adsorbed proteins. This hypothesis was then proved for the interaction of medium-hydrophobic isoflavones and highly hydrophobic resins. Moreover, a linear relation was discovered between the logarithm of the slope of the isotherm and the logD value of the compound, pointing into possible predictability of the isotherms for isoflavones. This relationship was further studied for higher concentrations in which the isotherms do not behave linearly. A lattice model was developed to calculate the molar fraction of the compounds in the adsorbed phase in contact with the resin (Chapter 6). This allowed for the experimental calculations of activity coefficients in the adsorbed phase that were modelled with a simple activity coefficient model (Margules). As in previous chapter, a linear relation was discovered between the logarithm of the equilibrium constant and the logD value. Moreover, the regressed interaction parameters showed as well a linear correlation with the logD values of the compounds in equilibrium. This way a “toolbox” of models is established and validated for the prediction of soy and tea polyphenol properties. These models can be readily used for the design of a separation process or to gain insight and develop an integrated process. Moreover, because of the nature of the models, either group contribution or mechanistic, this toolbox is easily applicable to different polyphenols from different sources.BioTechnologyApplied Science