Consumers must be assured that bought food supplements contain both bilberry extract and the anthocyanin amounts that match the declared levels. Therefore, a Fourier transform near-infrared (FT-NIR) spectroscopic method was validated based on principal component scores for the prediction of bilberry extract adulteration and partial least squares regression model for total anthocyanin evaluation. Anthocyanins have been quantified individually in 71 commercial bilberry extracts by HPLC-DAD, and 6 of them were counterfeit. The anthocyanin content in bilberry extracts was in the range 18–34%. Authentic bilberry extracts (
Anthocyanins (ACNs) are one of the most important pigments of vascular plants and are responsible for the shiny orange, pink, red, violet, and blue colours in the flowers and fruits. Anthocyanins are harmless and water soluble, which makes them interesting for use as natural colorants. In addition, in the past decades, there has been increased interest in anthocyanins because of their potential impact on human health [
Standards of cyanidin (Cy-), delphinidin (D-), petunidin (Pt-), peonidin (Pe-), and malvidin (Mv-) and their 3-O-glucoside (glc), Cy-3-O-galactoside (Cy-gal), Pt-gal, Pe-gal, Mv-gal, Cy-arabinoside (Cy-ara), and Cy-rutinoside (Cy-rut) were purchased from Polyphenols Laboratory (Sandnes, Norway). Methanol, acetonitrile hydrochloric, and phosphoric acid were from Merck (Darmstadt). Water was obtained from arium pro apparatus (Sartorius, Milan). Refined and standardised dry extract from the bilberry fruit (Sref) (CRS 1602, code Y0001059) was provided by the European Pharmacopoeia Reference Standard (EDQM, Strasbourg), and bilberry extracts
Approximately 100 mg of powder was dissolved in approximately 20 ml of a solution of methanol :H3PO4 1% in water (10 : 90, v/v). The suspension was sonicated for 10 min at room temperature and centrifuged at 1000 ×g for 5 min, and the supernatant was recovered. The residue, if present, was extracted and treated as described above. The supernatants were combined, and then the final volume was adjusted to 50 ml with 1% H3PO4 in water.
The HPLC system was an Alliance 2695 (Waters, Milford, MA, USA) equipped with a model 2998 photodiode array detector (Waters). A 2.6
Fourier transform near-infrared (FT-NIR) spectra were recorded in the reflectance mode using a model Tango spectrophotometer (Bruker Optics, Ettlingen) equipped with a gold integrating sphere. Two aliquots for each sample were analyzed, recording spectra in duplicate in order to account for the instrumental or sampling variability. Spectra were recorded in the range of 4000–12500 cm−1, from an average of 64 scans and with a resolution of 8 cm−1. Approximately 20 g of dry extract powder was put in the sample cup, and the data were collected three times for each sample. Principal component analysis (PCA), partial least-squares regression (PLS) modelling, and Mahalanobis distance were performed using OPUS Quant 7.5 (Bruker). Due to the limited number of samples in the data set, cross validation (leave-one-out method) was applied. Thus, authentic bilberry extract samples with known anthocyanin content (
NIR spectra contain large quantities of data that require a combination of statistical and mathematical sciences for their understanding. Therefore, preprocessing is needed to remove noise and background information. Spectral preprocessing was performed including no spectral data preprocessing, smoothing by the Savitzky–Golay (SG) method, multiplicative scatter correction (MSC), first derivative (1stDer) and second derivative (2ndDer) by the SG method, vector normalization (VN), straight line subtraction (SLS), minimum maximum normalization (MMN), subtraction of a constant offset (CO), rank optimization, 1stDer + SLS, 1stDer + VN, and 1stDer + MSC.
In brief, smoothing improves the quality of the spectra by removing noise, mainly consisting of moving average filters and applying the SG algorithm. MSC is used to diminish effects in the spectra caused by artifacts or imperfections such as undesirable scatter effect. This method is often used in diffusive reflection measurements. First and second derivatives eliminate baseline drifts, and small spectral differences are enhanced. To avoid enhancing the noise, which is a consequence of the derivative, spectra are first smoothed by the SG algorithm. VN is used to normalize the spectrum by first calculating the average intensity value and subsequent subtraction of this value from the spectrum. Basically, in diffusive reflection, the interferences from different material densities or particle sizes can often be minimized. MMN is used to transform the data into a desired range by subtracting the minimum value from each individual spectrum and then dividing the range of this spectrum. In SLS, preprocessing a straight line is fitted to the spectrum, using the PLS method, and then subtracted from the respective spectrum. In this way, a linear tilt of the baseline shift is eliminated. In the CO, the spectra are shifted in order to set the
The rank value, which defines the optimal number of principal components chosen for the analysis, was calculated by plotting the root-mean-square error of calibration (RMSEE) and prediction (RMSEP) values against the correspondent’s
If
After the group separation, the identity test was performed to determine the
Statistical analyses were performed with Statistica software (StatSoft Inc., Tulsa, OK, USA). Accuracy was determined by the Wilcoxon test considering significant a level of
The chromatogram relating to the reference bilberry extract, obtained at 520 nm, showed the presence of 15 main anthocyanins, and the respective aglycones were lower than 0.1%. Quantification of the ACNs was based on authentic standards and for D-gal, D-ara, Pt-ara, and Mv-ara by the Cy-glc calibration curve because pure compounds were not available. The monomeric anthocyanin pigment content of the analyzed bilberry extract samples ranged approximately from 18 to 34%. The precision of the method was tested by both repeatability (
A total of 260 FT-NIR spectra of 65 samples of authentic bilberry extract were recorded in the range 4000–12500 cm−1 at room temperature (23 ± 1°C). Spectral data were analyzed by PCA carried out with a validation to search for linear combinations of variables, which best explain the obtained data without taking into account external information. The PC1 and PC2 were responsible for about 61 and 26%, respectively, of the total variance among the examined samples. The scores plot of these PCs indicated that the bilberry group had different spectral patterns, and it could be distinguished from other groups present in library. In particular, the threshold for the group of the bilberry extracts was 0.353. After validation of the library, the adulterated bilberry extracts (Sa–Sf) were subjected to the identity test, and the
FT-NIR mean spectrum from the “outlier” (a), authentic (b), and adulterated (c) bilberry sample acquired from 4000 to 12500 cm−1.
Generally, the spectra obtained with the NIR analysis require optimization of the width of the interval of wavelengths considered for the large number of test samples, which introduces a high number of variables. In this way, the signal extracted from the spectra is decomposed by means of PCA to select and eliminate nonrelevant variables (principal component with low eigenvalues) and improve the quality of the calibration model. Thus, spectral components of the signal representing the conditions of minimum error are selected. The choice made in this way allows the optimal bands for characterisation of the samples to be identified. Mathematical pretreatments of the NIR data were carried out to enhance the prediction ability of the models and the qualitative interpretation of the spectra. The best preprocessing strategies chose for the spectra to develop the NIR model were obtained by smoothing (9 points) and straight line subtraction (SLS). Smoothing to remove the noise of the data and SLS fits a straight line to the spectrum and subtracts it. Spectral and chemical data are acquired in the form of matrices, in which each row represents a sample spectrum and then reduced to a few latent variables. Not all principal components are relevant to describe the spectral features, so only the most relevant ones should be used to perform the regression model. In this way, “overfitting” model can be avoided. On the contrary, less latent variables give a lower adaptability because of lacking enough information. Their number in the chemometric model is termed the “rank,” and a value below 10 is desirable. In our chemometric model, automatically created after forming the spectral and concentration data and after choosing the optimal preprocessing method, the optimal rank obtained by calibration and prediction was 8 and 7, respectively (Figure
Reference measured versus predicted value of calibration (a) and validation (b) samples for total anthocyanins (%) in bilberry extracts using the PLS model.
Summary of the NIRS calibration model statistics.
Parameters | ACN content in bilberry extracts |
---|---|
ACN content | % dried matter |
SEL | 0.31 |
|
38 |
Outliers | 0 |
Min | 18.8 |
Mean | 24.2 |
Max | 33.2 |
SD | 4.1 |
RMSEE | 0.28 |
|
99.63 |
RPD | 12.5 |
RANK | 8 |
Segments/range (cm−1) | 2/9400–6096, 5456–4248 |
Acquisition range (cm−1)/step | 4000–12500/8 |
Pretreatments tested | NSDPP, CO, MSC, 1stDer, 2ndDer, VN, SLS, MMN, RO, 1stDer + SLS, 1stDer + VN, and 1stDer + MSC |
Pretreatments used | SLS |
Regression method | PLS |
Threshold | 0.353 |
Mahalanobis distance | 0.91 |
Summary of the NIRS validation statistics.
Parameters | Total ACNs in bilberry extracts |
---|---|
ACN content | % dried matter |
|
27 |
Outliers | 0 |
Min | 18.3 |
Mean | 25.7 |
Max | 33.9 |
SD | 4.4 |
RMSEP | 0.303 |
|
99.51 |
SEP | 0.28 |
Rank | 7 |
RPD | 15.4 |
Bias | −0.114 |
Slope | 0.9704 |
Offset | 0.8755 |
The results obtained in this study showed the potential of FT-NIR spectroscopy with chemometric techniques to discriminate genuine bilberry extracts from those adulterated with anthocyanins extracted from other berries. The Mahalanobis distance method was successfully used to exclude the outliers, and the differences were removed by preprocessing procedures. The results also showed a relationship between near-infrared spectra and the amount of anthocyanins in the bilberry extract. Moreover, this technique allowed the rapid, accurate, and nondestructive quantification of total anthocyanins in commercial bilberry extracts that are commercially used for the production of food supplements. Thus, FT-NIR spectroscopy could be applied in a quality control laboratory to monitor adulteration and/or contamination and assess anthocyanin content in bilberry extract.
The data used to support the findings of this study are available from the corresponding author upon request.
The authors declare that there are no conflicts of interest regarding the publication of this paper.
Supplementary file 1 (S1). Sref and Sa–Se chromatograms. Supplementary file 2 (S2). RMSEE and RMSEP versus rank. Supplementary file 3 (S3). Residuum versus True ACNs% normal distribution of the residuum. Supplementary file 4 (S4). Spectrum residual versus Mahalanobis distance.