Janina Wulf, Dieter Treutter1, Susanne
Rühmann1, Manuela Zude
1
Technische Universität München, Fachgebiet Obstbau, Freising
Due to actual discussions and new findings of the relation between nutrition and healthiness based on clinical studies, the demand of food stuffs promoting a better healthiness such as fruits and vegetables became more and more popular for their internal quality parameters such as vitamins, polyphenols and other products of the secondary metabolism with health promoting effects. However, the contents of these compounds are severely influenced by pre- und postharvest conditions. Very often such influence is not visible by means of product appearance. In this context, consumers became also more conscious about quality controls and product-orientated monitoring in the entire preharvest and postharvest supply chain.
In this study the laser-induced fluorescence spectroscopy (LIFS) (EXC 337nm/EMI 400-820nm) was non-destructively applied (Figure 1) aiming to develop a calibration protocol for fluorescent polyphenols in fresh strawberries using high-pressure liquid-chromatography (HPLC) for the reference analyses. For building the calibration protocol different data pre-processing steps were evaluated. Additionally reflectance spectra were measured to correct the fluorescence spectral data. Robustness of calibrations after different data pre-processing were compared and validated using an independent test-set of fruit samples.
When building partial least squares (PLS) regression based on the fluorescence spectra of the strawberry fruits (X matrix) and the chemically analysed content of p-coumaroyl-glucose (y) different pre-processing methods for the spectral data (smoothing, standardisation, derivation) were compared with respect to the regression coefficient of calibration (rc²) and the root mean squares error of cross validation (rmsecv). Smoothing is reducing the noise in the spectral data; this may explain the better rmsecv value as by using the raw fluorescence spectra. The derivation generally leads to increasing noise in spectral data, consistently a higher rmsecv value has been found with a smaller rc² at the same time. A better calibration model has been calculated by using autoscaled fluorescence spectra. These pre-processing methods are feasible to reduce noise, to correct baseline shifts or to deal with unequal weightings, but such conventional calibration leads to low rc² and high error values. Based on the assumption that variation in the overall spectral fruit fluorescence may be masked by reabsorption, the direct orthogonal signal correction (DOSC) algorithm as well as corrections using fruit reflectance data were tested. Applying the PLS calibration models, calculated on standardized fluorescence spectra followed by DOSC, on an independent test-set resulted in high rv² values in the validation (> 0.93) and relatively low standard errors of prediction (rmsep = 15.03 %) (Figure 2). In contrast the often applied approach in fluorescence spectroscopy to correct with the information included in the simultaneously recorded reflectance spectra leads to less robust models. It can be assumed that the use of the reflectance spectral data includes information in the spectral data matrix that may not be relevant for the prediction of the phenol content.
The fluorescence spectroscopy is an appropriate method to non-destructively predict the fruit polyphenol content developing adequate calibration models. Furthermore mathematically data pre-processing methods such as DOSC are suitable for the correction of scattering and reabsorption effects in fruit tissue.
|
|
 |
|
Schematic view of the experimental set-up. Strawberry fruits were measured at different ripening stages. |
Calibration model built on the spectral data of strawberries and their content of p-cumaroyl-glucose [mg/g dry weight] after pre-processing the fluorescence spectra by autoscaling and DOSC (green) and validation on a independent test-set of fruits (orange). |
