Spectroscopic quantifiication of preservatives in different food matrices using QuEChERS extraction and multivariate calibration with comparison against liquid chromatography

2.1 Materials and reagents

Analytical standards of potassium sorbate PS, sodium benzoate SB, propylparaben PP, and methylparaben MP with > 99% purity were purchased from Sigma-Aldrich® (USA). Some characteristics of tested preservatives are provided in Table 1.

HPLC-grade solvents including acetonitrile, ethanol, methanol, acetone, and hexane were purchased from DBH® chemicals (UK). Analytical grades of MgSO₄, NaCl, and acetic acid were supplied from Sigma-Aldrich® (USA). Doubly distilled water was used in all preparations. Single-solute solutions were prepared in water-acetonitrile (50% by vol.) at 1000 mg/L. Diluted solutions were prepared at 100 mg/L by mixing suitable volumes of original standard solutions and diluting 50% water-acetonitrile mixture. The solutions were placed in a cold place at − 10 °C.

2.2 Instruments and software

Spectral analysis of preservatives over 200–300 nm was carried out using a Thermo Scientific UV–Vis spectrometer (Thermo Scientific, USA). Preservatives were separated using Shimadzu instrument (Shimadzu, LC6AD Liquid Chromatography, Japan) using C₁₈ (150 × 4.6 mm, 4.6 μm) columns. pH measurements were made using a Toledo Metler (Switzerland) pH meter with a calibrated glass electrodes. PLS calibration and NAS calculations were carried using MVC1 program which is freely available in the literature (Olivieri et al., 2004).

2.3 Multivariate calibration by PLS-1 for determination of preservatives

Initially, single-component calibrations were performed for the preservatives to find dynamic liner range, which is necessary when designing calibration and validation sets. A series of standards were prepared for each solute by diluting different volumes of stock solution using a 50% water-acetonitrile mixture. The dynamic range for each solute was determined by plotting the absorbance values measured at λ_max versus its concentration. Through Brereton’s tables, the optimised calibration standards of preservatives were prepared to train the PLS1 model, while another randomly prepared validation set is used for external validation (Brereton et al., 2007). The design of the calibration set was made to ensure a balanced combination of preservatives in the solution following a four-level full factorial design (Brereton, 2007). The selected levels were 4.0, 6.0, 8.0, and 10.0 mg/L for each preservatives.

Solutions that contain all preservatives were prepared by taking appropriate volumes of each standard, added to a 100 mL volumetric flask and then diluted with the acetonitrile–water mixture. Compositions of calibration and validation sets along with PLS1 prediction and other related parameters were provided in Table 2.

2.4 UV scanning and PLS1 calibration

UV measurements were achieved using a quartz cuvette of 1.0 cm length. UV spectra of preservatives were measured over 200–320 nm range, and the digitalised absorbance values (1.0 nm step, 121 datum/spectrum) were exported to matlab® (MATLAB® (version 7.0) before running PLS1 calibration. Spectral data were placed in matrix A, where preservatives were placed in matrix C, while spectral data of unknown sample placed in vector a. The main step in PLS1 calibration is to decompose A using the concentration information of calibrated solute (Brereton, 2007). The calibration vector b of the solute (perseverative) is obtained as (Eq. (1)) (Brereton, 2007; Olivieri, 2018):

Where W, P and q are A-weight matrix, A-loading matrix, and c-loading vector that estimated at the optimum number of PLS variables which often estimated by cross validation leave-one-out methodology (Brereton, 2007). Matrix inverse operation is donated as t. Prediction of calibrated preservative in new sample of spectrum a is carried out as (Eq. (2)) (Brereton, 2007; Al-Degs et al., 2012):

The performance of PLS1 was assessed by estimating standard error for prediction in calibration SEC and validation SEP sets as the following (Eq. (3)) (Brereton, 2007; Al-Degs et al., 2012):

Where C_act, C _pred, and n are actual concentration (mg/L), predicted concentration (mg/L), and the number of calibration or validation samples, respectively. The relative error of prediction (REP%), which measures the overall performance of PLS1 to predict preservatives in the validation set, was estimated as (Eq. (4)) (Brereton, 2007):

The goodness of fit of the model was also assessed by estimating the correlation coefficient (R²) as the following (Eq. (5)) (Brereton, 2007; Al-Degs et al., 2012):

Where $\overline{c}$ is the average concentration of the calibrated solute (see Table 3).

2.5 Analytical figures of merit by net analyte signal

Figures of merit of the proposed multivariate calibration method and spectral overlapping among preservatives were estimated using the net analyte signal NAS concept (Lorber et al., 1997; Goicoechea & Olivieri, 2002; Hemmateenejad et al., 2006; Al-Degs et al., 2012; Yousefinejad & Hemmateenejad, 2012; Salameh et al., 2020; Al-Degs et al., 2021). For analyte k in a given mixture, the net analyte signal NAS of k is the part of its spectrum that is orthogonal to space spanned by the spectra of all other analytes or interferences (Yousefinejad & Hemmateenejad, 2012). Lorber’s algorithm for inverse multivariate calibrations was adopted (Lorber et al., 1997). Initially, PLS regression was performed on the calibration data (A and c_k) at an optimum number of variables. The A was reconstructed (Â) using an optimum number of PLS variables. The matrix Â_-k containing the absorbance data of all analytes in the calibration sample except the solute k was then estimated by the rank annihilation method (Yousefinejad & Hemmateenejad, 2012). A projection matrix H that is orthogonal to Â_-k is defined as (Eq. (6)) (Salameh et al., 2020):

Where “+” superscript present the matrix pseudo-inverse, and I is the identity matrix. Net analyte signal vector (r^∗_k) is estimated by multiplying H with the vector spectra of the mixture r (Hemmateenejad et al., 2006; Al-Degs et al., 2021):

Equation (7) where r is substituted by the pure spectrum of analyte k (s_k) and the left-hand side of this equation produces the net pure-spectrum of k (the part of a pure spectrum of k is orthogonal to the spectra of other absorbing solutes (s^∗_k) (Hemmateenejad et al., 2006). NAS calculations are used to estimate analytical figures of merit, including sensitivity, selectivity and limit of detection (Yousefinejad & Hemmateenejad, 2012; Al-Degs et al., 2021). These figures are used to assess the quality of a given analytical technique. The sensitivity is defined as the norm of s^∗_k vector as (Eq. (8)) (Yousefinejad & Hemmateenejad, 2012):

The selectivity SEL is related to the amount of signal that is used for prediction. Moreover, SEL can also be defined as the relative amount of signal not lost by spectral overlap with other analytes. Hence, a higher SEL value indicated less spectral overlapping. SEL is defined as (Eq. (9)) (Yousefinejad & Hemmateenejad, 2012):

Detection limit DL can be estimated using different approaches. However, the most practical definition is derived from the classical univariate detection limit (Eq. (10)) (Yousefinejad & Hemmateenejad, 2012; Al-Degs et al., 2021):

where ||ε|| is a measure of the instrumental noise as outlined in literature (Salameh et al., 2020).

2.6 Chromatographic separation of preservatives

The separation of preservatives was performed according to the analytical protocol outlined in the literature with minor modifications (Arias et al., 2019). The mobile phase was made up of acetonitrile and 0.05 M acetate buffer maintained at pH 4.0. The four preservatives were separated using isocratic mode using 45% acetonitrile and 55% buffer at a 1.1 mL/min flow rate. The buffer was freshly prepared by adding glacial acetic acid to the ammonium acetate solution to attain the necessary pH value. The detection wavelength was set at 240 nm, and the injection volume was set at 20 µL.

2.7 Samples collection, conditioning and matrix cleaning by QuEChERS

Food samples of different items were selected, including cereals (corn flakes), dairy products (yoghurt), beverages (juice drinks), and condiments (soya sauce). The selected items have high consumption by a wide selection of consumers. The selected food samples were locally produced and saved in different packaging materials, including cartoon for corn flakes, plastic for yoghurt, and glass for juices/ soya sauce. The samples were selected from different supermarkets and placed in a dry and cold place before analysis. The solid samples (corn flakes) were grounded and homogenised, while liquid samples were sonicated to remove gases in preparation for QuEChERS extraction. The rest of the samples were homogenised and weighed before extraction. Matrix cleaning of food samples by QuEChERS was performed as the following (Arias et al., 2019): To a 50 mL polypropylene tube, 1.5 g sample and 10 mL distilled water was added and mixed to get a slurry, and the 150 μL acetic acid and 10 mL acetonitrile was added to the extraction tube. The whole mixture was sonicated for 1.5 min before adding 4.0 g MgSO₄ and 1.0 g NaCl to the mixture. The mixture was agitated and centrifuged (5500 rpm) for 5 min to separate solid residues. The extracted solution was then further refined using solid-phase extraction with chitosan as outlined in the litetrure (Arias et al., 2019). Finally, the upper acetonitrile layer was collected and diluted appropriately with distilled water before spectral scanning or chromatographic injection. Prior analyses, the extracts were filtered using a 0.45 µm cellulose filter. Each food item was analysed in triplicate, and the average of measurements was provided.

2.8 Preparation of matrix-based calibration standards and method validation

In order to validate the QuEChERS-PLS1 method, a series of calibration standards were prepared by spiking 1.5 g of preservative-free food with standards of preservatives to get final ranges 5–100 μg/g for PS and 10–200 μg/g for SB, MP, and PP. Spiked matrices were subjected to the outlined QuEChERS extraction methodology. The proposed method was further validated by reporting method limit of detection (LOD_m), method limit of quantification (LOQ_m), dynamic method range, accuracy (recovery), precision (repeatability and intermediate precision). The method LOQ (LOQ_m) was defined as the lowest spiked level of preparative whose recoveries ranged between 70 and 120% and relative standard deviation (RSD) ≤ 20% (Arias et al., 2019). Method linearity was investigated over a wide concentration range of 5–200 mg/kg by using standard solutions prepared in solvent and in the blank of the control sample. Method accuracy was assessed by a recovery study (R%) at 10 mg/kg in the selected food item. Both repeatability and reproducibility were assessed by estimating the RSD of triplicate measurements of 10 mg/kg spiked food sample. To assess the influence of QuEChERS on matrix cleaning and quality of analysis, matrix effect was estimated by comparing calibration curves prepared in the pure mobile phase and the matrix as the following (Eq. (11)) (Matuszewski et al., 2003):

Where S_m is the slope of calibration line obtained from spiked samples (matrix effect) and S_s is the slope of calibration line reported from the pure standards.

3.1 Intense spectral overlapping among preservatives and importance of PLS-1 calibration

In order to avoid intense UV absorption of preservatives in solution, the composition of the calibration set was carefully designed. Higher UV intensities would result in a non-linear correlation with solutes concentration and affect PLS calibration. Moreover, intense spectral overlapping is also contributed to a non-linear correlation with concentration. Single-solute calibration lines indicated the dynamic ranges, 0.3–10.0, 0.4–15.0, 0.8–15, and 0.5–20 mg/L for PS, PP, MP and SB. The dynamic range of PS was relatively small, and this was attributed to its high UV absorption compared to the rest of the preservatives. To avoid intense UV absorption, the mixture should contain lower levels of PS. The selected concentration range of preservatives was 4.0–10.0 mg/L. As indicated in Fig. 1, PS has the highest light absorption among all preservatives. Hence, in the mixtures containing high content of PS (10 mg/L) should be diluted (1:1) to get more sensible spectral measurements for the mixture. Accordingly, the solutions (11, 12 and 14) were diluted before UV scanning. UV spectra of preservatives are shown in Fig. 1.

Close

Fig. 1

UV spectra of preservatives at 8.0 mg/L (solvent 50% H₂O/CH₃CN mixture).

Export to PPT

Fig. 1 indicated that PS has a high UV absorption over the range (220 – 250 nm), and MP-PP has severe overlapping while SB has modest light absorption. The UV spectra of the mixture indicated high light absorption over the range 240–270 nm, which back to the combined absorption of PS-MP-PP as they have comparable λ_max (250–260 nm). For a better assessment of the analytical results, multivariate sensitivity, selectivity, and spectral overlapping for each solute were estimated using NAS calculations (Hemmateenejad et al., 2006; Al-Degs et al., 2012; Yousefinejad & Hemmateenejad, 2012; Salameh et al., 2020; Al-Degs et al., 2021). Analysis indicated high multivariate sensitivity SEN of PS with values of 3.22, 1.32, 1.03, and 0.57 for PS, MP, PP, and SB, respectively. The high PS sensitivity was due to high UV absorption over the spectrum. In the meantime, the low multivariate sensitivity of SB was due to modest absorption compared to the rest of the preservatives (Fig. 1). In addition, multivariate selectivity (which defined as the relative amount of UV spectrum that is not lost by spectral overlapping with signals of other solutes) can provide a better assessment of spectral overlapping and the need for multivariate calibration. The estimated multivariate selectivity SEL values (Eq. (9)) were 0.69, 0.48, 0.33, and 0. 29 for PS, SB, PP, and MP, respectively. The high multivariate selectivity of PS (0.69) was also attributed to its high sensitivity compared to the rest of the solutes. The estimated signal overlapping which estimated as 100×(1-SEL) were 31, 52, 67, and 71% for PS, SB, MP, and PP, respectively. The intense overlapping reported for MP and PP (67–71%) is clear in their UV spectra as shown in Fig. 1. Analysis indicated a serious spectral overlapping in the signals of SB, MP, and PP (50–70%), and this limited the application of direct spectrometry (i.e., calibration and prediction at λ_max) for quantification purposes. Direct spectrometry was applicable to predict PS only in the mixtures with REP% values < 10% in all cases, and this is attributed to the low signal overlapping with other solutes. Hence, the application of PLS1 calibration as a reasonable alternative was applied for mixture analysis.

PLS1 calibration was separately applied for each preservative. The performance of PLS1 calibration was provided in Table 1. The PLS variables needed for optimum calibration were 1, 2, and 5 for PS, SB and both MP/PP, respectively. The estimated number of variables (obtained using the leave-one-out cross-validation method) can be accounted for high signal overlapping or poor sensitivity. Only one variable was needed for PS (high sensitivity and low overlapping), while five variables were required to handle intense signal overlapping in parabens (MP/PP). On the other hand, 2 PLS variables were used to account for the low SB sensitivity. All preservatives were predicted in validation mixtures with REP% values 1.7, 4.8, 7.6, and 9.5 for PS, SB, MP and PP, respectively. In addition to the REP% value, other statistical parameters, including R²(0.9994), SEC (0.032) and SEP (0.071) indicating an excellent prediction of PS compared to the rest of the preservatives. Before running PLS, each solute’s most informative spectral regions were pre-selected using wavelength selection interval PLS (iPLS). In iPLS, the entire spectrum is divided into equal-width intervals and creates a sub-PLS model for each solute (Norgaard et al., 2000). The interval with the lowest value of SEC is then selected for prediction purposes (Norgaard et al., 2000). The most informative regions were 235–265 nm, 210–240 nm, and 230–275 nm for PS, SB, and PP/MP. Except for PS, using informative interval or full range did not affect PLS performance. In general, running PLS calibration over the 200–300 nm range led to optimum outputs for all preservatives under the study. The applicability of PLS to predict preservatives in real food matrices is highly dependent on the presence of interfering substances which should be removed to get informative results. In general, food preservatives are added to processed foods at high levels (i.e., >500 mg/kg). Hence, their detection by PLS is highly possible but in the absence of strong signal-overlapping from other food ingredients. Including all food ingredients in the PLS model maybe not practical considering their identity and concentration. With the help of QuEChERS extraction, detection and quantification of four preservatives in different food items were successfully achieved. The proposed QuEChERS-PLS was of comparable performance with a liquid chromatographic method, as will be outlined.

3.2 Method validation

As the developed analytical procedure should be applied for quantification of preservative in real food matrices, different samples presenting different food items were spiked and extracted. For instance, corn flakes, yoghurt, un-carbonated liquid drinks, and soya sauce were selected to present cereals, dairy product, beverages, and condiments, respectively. The selected samples have a variable degree of complexity and highly consumed. To better assess the QuEChERS-PLS method, yoghurt was selected due to its complexity, which contains carbohydrate, fat, and protein. Analytical figures of merit for QuEChERS-PLS included method limit of detection, method limit of quantification, method linearity, determination coefficients, accuracy, and precision were estimated and the results are provided in Table 3. Moreover, the matrix effect was evaluated. For comparison purposes, the samples were analysed by the independent chromatographic method as provided in Table 3.

For both PLS1 and HPLC, sample clean up by QuEChERS resulted in better detection of preservatives. For QuEChERS-PLS, the solutes would be detected down to 1.1 mg/kg for PS and 2.4 mg/kg for SB while parabens down to 2.0 mg/kg as shown in Table 3. The method quantification limits seem reasonable (3.6–8.0 mg/kg) considering sample complexity and intense overlapping among solutes. Moreover, the proposed method is rather sensitive as tested additives added in large levels (>500 mg/kg). Convenient recovery (R% 97–102%), precision (RSD 1.4–6.7%), and linearity (R² 0.979–0.999) were reported for quantification of preservatives by QuEChERS-PLS, and this was attributed to the efficient matrix cleaning by QuEChERS extraction and accurate prediction by PLS1 regression. The better detection of PS (1.1 mg/kg) was attributed to its high multivariate sensitivity and selectivity, as outlined earlier. Except for PS, the preservatives would be detected in food up to 200 mg/kg as shown in Table 3. With a lower detection limit (0.6–1.7 mg/kg) and wider dynamic range (1.9–800 mg/kg), QuEChERS-HPLC outperformed the QuEChERS-PLS, and this was expected due to the actual separation of solutes before UV detection. In other words, In HPLC-QuEChERS the solutes were separated before detection while in PLS-QuEChERS no physical separation was performed, therefore, HPLC-QuEChERS achieved lower detections and wider linearity ranges for preservatives. The provided chromatograms may evidence the high efficiency of QuEChERS extraction toward matrix cleaning in Fig. 2.

Close

Fig. 2

Chromatograms of 20 mg/L mixture of preservatives A: corn flakes sample containing 77.3 mg/kg methylparaben B; Juice sample containing 131.7 mg/kg sodium benzoate and 27.4 mg/kg methylparaben C. Chromatographic parameters: mobile phase, acetonitrile – pH 4.0 acetate buffer (45:55), flow rate 1.1 mL/min, detection 240 nm, and injection volume 20 μL.

Export to PPT

As indicated in Fig. 2, the adopted chromatographic procedure was efficient as preservatives were separated entirely in 13 min, and all peaks have Gaussian shape with no peak tailing or fronting. Interestingly, the measured chromatogram (at 20 mg/L for each solute) indicated high sensitivity of PS compared to the rest of the preservatives. Fig. 2B and 2C established the excellent efficiency of QuEChERS extraction to remove other foods extractives or interferences with convenient detection of targeted solutes. The stable and noise-free baselines of food chromatograms (Fig. 2B and 2C) proved the high extraction performance of QuEChERS. The same conclusions have been reported in the literature (Arias et al., 2019). The combination of QuEChERS with PLS or even HPLC has reduced matrix effect and improved detection of the analytical method. It is worth mentioning that eliminating other food ingredients (fat, proteins, sugar) has been positively reflected on the performance of PLS1 as all signal-generating compounds have to be in the calibration set (Brereton et al., 2007; Olivieri et al., 2018). The matrix effect was also assessed by comparing the slopes of the calibration lines of the matrix extract (the one obtained from the yoghurt sample) and the other one obtained from the standards. For QuEChERS-PLS, a low influence of food matrix on preservatives determinations (ME 1.23–5.31%) was reported, as indicated in Table 2. The chromatographic procedure was effectively handled other interferences with ME 0.14–0.67% for the tested matrices, which attributed to the physical separation of all components before detection. Generally, the matrix effect was notably reduced thanks to QuEChERS extraction; hence, calibration curves based on pure standards were often used for preservatives quantification in tested samples. It is necessary to mention that ME values of ± 20% are acceptable when analysing complex food matrices (Norgaard et al., 2000). With lower ME% values 0.14–0.76, QuEChERS-HPLC handled the negative influence of matrix on measurements due to the physical separation of all components before detection. The main advantage of QuEChERS-PLS against QuEChERS-HPLC is that it avoids using advanced chromatographic instruments and other related accessories, including columns, injectors, training, and regular maintenance.

3.3 Determination of preservatives in different food items

The developed analytical method was further tested to analyse preservatives in different food items collected from local markets in Amman, Jordan. Based on a quick interview with many food dealers, the tested food samples are highly consumed and cover four food items. Four identical determinations were carried out for each food sample following the optimised PLS calibration, and results are provided in Table 4. Moreover, spiking with the preservative claimed to be added to the food was necessary for specific situations. The analytical performance of QuEChERS-PLS was checked against QuEChERS-HPLC

As shown in Table 4, the final results revealed that MP was added to corn flakes, liquid drinks and soy sauce in variable levels. Among tested food items, no additive was added to yoghurt. The screened samples contain preservatives with varying level (0–393.7 mg/kg), being the most abundant in soya sauce. The significant difference between the contents of preservatives among tested food items was verified by running single-factor ANOVA testing at p = 0.05. Liquid drinks were the only food that contains two preservatives, MP and SB. The combination of MP and SB in liquid drinks is interesting as SB or PS are often added to this food item. Packaging type would be a possible source for some parabens, including MP (Maher et al., 2020). Hence, the detection of MP in liquid drinks would be attributed to the packaging material. Addition of MP to corn flakes and soya sauce due to its high activity as an antimicrobial agent (Maher et al., 2020). The absence of PP in all samples was expected as this preservative is less desirable in foodstuff products while it is used extensively in cosmetics (Piao et al., 2014). Although the levels of MP and SB appear to be high (393.7 mg/kg in soya and 128.4 mg/kg in liquid drinks), they are less than the regulated levels in processed foods (Piao et al., 2014). Based on Codex food standards, PS/SB and MP can be added to many food items up to 3000 and 500 mg/kg, respectively (CODEX, 2017). The outputs of spiking tests indicated the high performance of QuEChERS-PLS for preservatives in different food matrices.

In one sample of the tested samples, SB was detected and assayed with high recoveries (95.7–103.4%) and accepted RSD (3.7–7.2). Comparable results were reported for MP, which detected four samples with recoveries of (99.5–105.4) and standard deviation (4.5–8.2). Determination of preservatives by an independent QuEChERS-HPLC was also provided in Table 4. t and F tests were carried out on the reported results to check whether the results of both methods are comparable (Table 4). Tabulated t value (0.05, 6) and F value (0.05, 3, 3) are 2.45 and 15.44, respectively. In all cases, calculated t values were lower than tabulated ones, indicating no significant difference between both methods. Moreover, F-calculated values were also lower than tabulate one, which confirmed that both methods are of comparable precision.