A Feasibility Study on Monitoring Shelf Life of Bottled Natural Fruit Juice Using Laser-Induced Autofluorescence

. Shelf life of bottled natural fruit juice (BNFJ) provides relevant information on quality and authenticity for consumer protection. However, existing techniques for monitoring the shelf life of BNFJ are destructive and time-consuming. We report on using laser-induced autofuorescence (LIAF) spectroscopic technique in combination with multivariate analysis for shelf life monitoring of BNFJ. Te LIAF spectra data were acquired for nine (9) continuous days on three batches of BNFJ samples purchased from a certifed retailer. Deconvolution of the LIAF spectra revealed underlying peaks representing constituents of the BNFJ. Principal component analysis (PCA) was able to monitor the trend in the changes of the BNFJ as it aged. Partial least square regression (PLSR) predicted the exact day from the production of the BNFJ accurately at 96.6% and 98.8% in the training and testing sets, respectively. We, therefore, propose the LIAF combined with multivariate analysis as a potential tool for nondestructive, rapid, and relatively inexpensive monitoring of the shelf life of BNFJ.


Introduction
Natural fruit juice (NFJ) contains a lot of nutritional ingredients such as vitamins, minerals, antioxidants, and fbres which are essential for human health [1]. Even though there are diferent kinds of NFJ in production today, bottled natural fruit juice (BNFJ) remains the most produced in the beverage industry [2]. Its shelf life provides vital information on quality and authenticity for consumer protection, explaining why expiring dates are relevant and provided. Te shelf life for BNFJ is therefore an essential indicator for quality control in the industry and is known to be short [3]. Shelf life is the period of time under defned conditions of storage after manufacture or packaging, during which food products will remain safe and suitable for consumption [4]. During this period, food products retain their sensory, chemical, physical, functional, microbiological, and nutritional characteristics in the best conditions for acceptance by consumers [5]. Within this period, the products are expected to meet any label declaration of nutritional information when stored according to the recommended conditions. Te shelf life of any food product depends on factors such as composition, processing methods, packaging, and storage conditions [4]. Tis can be determined by monitoring physical, chemical, microbiological, and sensory changes during storage by measuring deterioration characteristics [5]. Shelf life studies have previously been conducted on diferent fruits such as orange, carrot, apple, cider, cranberry, mango, and tomato, as well as a mixture of two or more of these fruits [6][7][8][9][10][11][12]. In these studies, physicochemical parameters (pH, titratable acidity, total soluble solids, etc.) and sensory tests are employed to determine the deterioration associated with the fruit juices as they aged. Methods of determining these parameters are cumbersome and destructive, and therefore there is a need for a nondestructive, sensitive, and convenient method of which laser-induced autofuorescence (LIAF) ofers the needed advantage.
Laser-induced autofuorescence is a nondestructive, noninvasive, and sensitive analytical technique based on light absorbance and emission intensities that can be used to rapidly identify the presence of fuorescent molecules in a sample [13]. Fluorescent molecules in food include aromatic amino acids, vitamins, polyphenolics, and a variety of favouring compounds which are suitable and reliable to detect using the LIAF technique [14]. Te application of this technique to food samples has been suggested for the analysis of sugar, yoghurt, cheese, oils, honey, and distilled beverages [15][16][17][18][19]. Despite such successful applications of the LIAF technique, its application to the study of the shelf life of BNFJ is yet to be fully exploited.
Te LIAF measurements normally produce large sets of data which at times could be tricky to analyze by mere visual inspection. Multivariate statistical methods can be adopted for data exploration, data reduction, classifcation, calibration, regression, and wavelength selection to assist with this type of analysis. Multivariate analysis utilizes mathematics and statistics to extract relevant information from spectral data. Multivariate analysis methods such as principal component analysis (PCA) allow the extraction of principal components or eigenvectors of a correlation matrix in a dataset, fnds the main sources of variability in the dataset, and establishes the relationship between/within objects and variables [20,21], whereas partial least square (PLS) regression can be used to relate the values of predictive data to the studied response [22]. LIAF has been combined with PCA and PLS for several studies [23][24][25].
Tis research aimed to evaluate the feasibility of using LIAF and multivariate analysis methods to monitor the shelf life of BNFJ.

Experimental Methods and Procedure
2.1. Bottled Natural Fruit Juice Samples. Nine (9) BNFJ samples comprising 3 samples from 3 diferent production batches were purchased from a certifed retailer for the study. Tese samples had been prepared from a mixture of mango, pineapple, and passion fruits with no preservatives added. Te shelf life as indicated on the BNFJ samples was seven (7) days. Each batch of the three samples was bought fresh on the frst day of production and kept in a refrigerator at 4°C in the same laboratory with the LIAF setup. Te samples were brought out one after the other each day for their LIAF spectra measurement to be conducted and kept back in the refrigerator within fve minutes.

Laser-Induced Autofuorescence Measurements.
Te LIAF setup used in this study is shown in Figure 1 Te laser source was coupled to one arm of the bifurcated fbre optic probe using a fbre port micro-positioner (PAF-SMA-5-B, Torlabs). Tis arm of the bifurcated fbre optic probe was incident on the sample, and the backscattered fuorescence light was collected by the other arm of the bifurcated fbre into the detection system which consisted of the high-pass absorptive edge flter and spectrometer. Te flter was placed before the spectrometer to cut of the excitation wavelength, and the spectrometer was interfaced to an HP laptop computer (Intel (R) Core i3-2310M CPU @ 2.10 GHz, 796 MHz, and 2.94 GB of RAM) to display and record the LIF spectra using OOIBase32 software, an interface for the USB 2000 Ocean Optics spectrometer.
Te LIAF spectral data were frst obtained for an emptied BNFJ bottle and then on the NFJ flled samples. Te spectral data collection on each sample was done within a space of fve minutes in 24-hour intervals for nine (9) continuous days using the LIAF setup. For each measurement, 92 LIAF spectra data were recorded for 60 s using an integration time of 300 ms at ambient temperature. In all 276 spectra, data (i.e., 92 spectra × 3 replicates) were recorded and averaged for each BNFJ sample. A total of 81 spectra were obtained comprising spectra from 3 samples × 3 batches of production × 9 days of measurement. Te recorded data were then exported into MATLAB (R2019a MATLAB 9.6, MathWorks Inc., USA) for analysis.

Deconvolution of the Laser-Induced Autofuorescence
Spectra. PeakFit software (4.12 version, Jandel Scientifc, Germany) was used to deconvolve each of the spectra into separate bands as was done in a previous study [26]. Te PeakFit software combined the Loess smoothing function and Marquardt-Levenberg and Lorentzian spectral functions for analyzing the LIAF spectra. Te Lorentzian spectral function helped the choosing of a reasonable corresponding ft of the spectra. Tis enabled the determination of the peak amplitude, centre wavelength, and full width at half maximum (FWHM) for further analysis. Te individual peak wavelength positions of the peak were compared with the literature to identify corresponding molecular constituents of the BNFJ.

Principal Component Analysis of Laser-Induced
Autofuorescence Spectra. Principal component analysis (PCA) was applied to the averaged LIAF spectra data of the BNFJ samples using self-written MATLAB codes. Te PCA technique is useful for reducing the dimensionality and exploring underlining patterns in the spectral data [20]. In this work, the fuorescence spectra matrix is represented as Q with i rows (81 observations) and j columns (2048 variables). PCA decomposes Q as a sum of series combinations of scores (u i ) and the loadings (v i ) as in equation (1). Te scores (u i vectors) contain information on how the fuorescence spectra relate to each other in the principal component (PC) space, whereas the loadings (v i vectors) contain information on how the wavelengths relate to one another.  (2) and (3) to construct a regression model to predict the response for unknown sample X test using equation (4).
where M and N are the scores and P and Q are the loadings of X and Y, respectively. U is a diagonal matrix with the regression weights.
Te accuracy of prediction model was measured using Pearson's correlation coefcient (R 2 ), rroot mean square error of calibration (RMSEC), and root mean square error of prediction (RMSEP) with equations (5)-(7), respectively. An acceptable robust model should have a high correlation coefcient, a low root mean square error (RMSE), and a very low prediction bias [27].
where n is the number of samples (in the training set and test set), y i is the reference measurement results for sample i, y i is the estimated result for sample i when the model is constructed with sample i removed, y i is the estimated results of the model for the sample i, and y is the average of the reference measurement results for all. In addition, the systematic error of the average diference between the actual and predicted values known as bias was calculated using the following equation.
where y i is the prediction of the removed sample.

Laser-Induced Autofuorescence Spectra.
Te LIAF spectra of the BNFJ samples for the 9 continuous days are shown in Figure 2. Each day's spectra represent the averaged spectra from all samples (9) of the 3 batches. 9 days were chosen to enable two (2) extra days of observation of the LIAF spectra from the BNFJ upon expiry after the recorded shelf life of 7 days. Initial measurement on the fuorescence spectra of the empty NFJ bottle only (not presented) shows that the bottle had no efect on the LIAF measurements. Visual inspection of Figure 2 shows variations that can be observed in the normalized intensities of the LIAF spectra for the diferent days in diferent wavelength regions of the spectra. For instance, a reduction in spectral intensity is seen each day within the spectral region of 520-560 nm (inset). Te observed wavelengths in this study are closely related to the spectral signatures of chemical compositions in fruits [28][29][30][31]. Molecules that exhibit numerous conjugated double bonds, for instance, carotenoids, chlorophylls, and porphyrins, show light absorption in these spectral regions. Teir absorption properties could be used to assess food products and may be used to predict shelf life [32]. Tis observation suggests that once BNFJ is freshly prepared, there are variations in the soluble solid substances with time, resulting in weakening in absorption and re-emission of incident light by endogenous fuorophores in the BNFJ. Tis phenomenon can be used as a basis for monitoring the shelf life of BNFJ.
Furthermore, to explore the constituents of the BNFJ, PeakFit analysis, as shown in Figure 3, revealed six (6) hidden peaks in the LIAF spectra of the samples. Peak 1 ranges from 475 nm to 500 nm; peak 2 overlaps peak 1 and International Journal of Optics ends at 550 nm; peak 3 exhibits the highest fuorescence intensity and spread from 500 nm to 600 nm. Te remaining peaks are peak 4 (525 nm-650 nm), peak 5 (600 nm-650 nm), and peak 6 (625 nm-725 nm). Peak 5 shows the lowest fuorescence intensity. Peaks 1 to 4 (400-600 nm) show the presence of antioxidants mainly carotenoids which are essential for the delay of oxidation in fruits [31,32]. Te antioxidants present in the BNFJ samples deteriorate with the increasing number of days causing a rise in oxidation as observed in Figure 2, whereas peaks 5 and 6 (600-750 nm) show the presence of chlorophyll pigments [28].

Principal Component Analysis.
Variance plot describing the contribution of two (2) principal components (PCs): PC1 (85%) and PC2 (7%), and loading plot showing wavelength-dependent factors infuencing the variations in the 2 PCs are presented in Figures 4(a) and 4(b), respectively. Figure 4(a) reveals variations, signifcance, and contributions by ten (10) PCs to the LIAF data. It shows that PC1 is the linear combination of the LIAF data with maximum variance and PC2 is the linear combination with the next maximum variance orthogonal to PC1. Te variation in the LIAF data can be further interpreted by inspecting the loadings [20] in this case corresponding to PC1 and PC2 (Figure 4(a)). Loading plot of PC1 and PC2 (Figure 4(b)) can be related to the deconvoluted peaks 1 and 2 ( Figure 3). Besides, this plot shows part of the visible region (450 nm-750 nm) of the electromagnetic spectrum where electronic transitions occur. In addition, some peaks can be observed at specifc wavelengths (475, 478, 480, 495, 520, 550, 630, 660, 754, and 760 nm) which are considered useful for PCs to be used for predicting shelf life of BNFJ.
A scatter plot of PC1 and PC2, which together accounted for 92% of the total variability in the LIAF spectral data for 9 days, is presented in Figure 5. Te coefcients of PC1 are more signifcant as the number of days increases and also show a pattern of distribution of the BNFJ samples. Te negative coefcients of PC1 show the scores of BNFJ samples monitored for the frst 4 days while the positive coefcients display the scores of those monitored for the last 5 days. PC2 coefcients show a majority of the scores above the origin with a marginal separation of the BNFJ samples. Tese observations indicate that changes in BNFJ samples followed a common trend for 9 days in PC space, especially when PC1 scores are considered. Tus, PCA could help shelf life monitoring of BNFJ.
However, as seen in Figure 5, even though the PCA (unsupervised algorithm) follows a trend with regard to the shelf life, it did not give clear boundary between some of the days (e.g., day 1 and day 2). Terefore, a supervised regression algorithm was used to investigate the feasibility of LIAF for BNFJ shelf life monitoring.

PLS Regression Analysis.
Te PLS regression analysis for predicting the shelf life of BNFJ on both training and testing datasets is shown in Figure 6. Te optimal PLS model was obtained on 6 latent factors. Prediction performance of 0.98 and 0.99 for R 2 , 0.36 and 0.22 for RMSE (C/P), and 0.047 and 0.046 for prediction bias of the training set and test set was observed, respectively. Te prediction performance demonstrates the suitability of the PLS model for shelf life monitoring of BNFJ. Te PLSR analysis showed that each day of the BNFJ life span can correctly be predicted. Previous work on fruit juice was mostly destructive and not as simple and straight forward as compared to the LIAF method  International Journal of Optics [12,33]. Our results using the LIAF combined with PLS regression produced a clear indication of shelf life of BNFJ.  International Journal of Optics BNFJ in clear containers, further studies on colored and opaque containers are needed. Also, the study can be extended to other types of bottled natural fruit juices. Tis LIAF setup is simple and can be developed into a compact miniaturized system for mobility and easy use by regulatory bodies, BNFJ producers, and consumers.

Data Availability
Te laser-induced autofuorescence spectra data from the bottled natural fruit juice samples used to support the fndings of this research are available from the corresponding author upon reasonable request.

Conflicts of Interest
Te authors declare that they have no conficts of interest.