Simultaneous Spectrometric Determination of Caffeic Acid, Gallic Acid, and Quercetin in Some Aromatic Herbs, Using Chemometric Tools

Caffeic acid (>98%), gallic acid (>99%), quercetin (>95%) are Sigma-Aldrich reagents. All the other reagents used are of analytical grade and used without purification.

Appropriate volumes of caffeic acid, gallic acid, and quercetin were transferred to 25 mL volumetric flasks. The solutions were thoroughly mixed and made up to the mark with doubly distilled water. The final concentrations of these solutions were kept in their linear determination ranges. All the samples were prepared in triplicate to minimize errors of measurements.

The studied material consisted of some herbs commonly used in households: basil (Ocimum basilicum), chive (Allium schoenoprasum), laurel (Laurusnobilis), mint (Mentha piperita), parsley (Petroselinum crispum), and thyme (Thymbra spicata). These studied herbs were purchased directly from local markets in the north of Algeria. They included both imported and locally made products. All herbal samples were treated in an identical manner. The extracts of basil, chive, laurel, mint, parsley, and thyme, are obtained by infusion of 0.1 g of dry matter in 100 mL of doubly distilled water for 15 minutes, then filtered and cooled in the dark. For each herb, we used at least three Algerian brands and we considered and presented the average in the “results and discussion” section.

All absorbance measurements were collected on a Shimadzu UV2102PC double bean UV-Vis spectrophotometer, using a 10 mm quartz cell. The recorded spectra were digitized with an interval of 0.5 nm between consecutive points. Therefore, 301 data points were used to represent a spectrum in the range of (250–400) nm. For each measurement, about 2 mL of the above solution were transferred to a spectrophotometric cell and the spectra of all prepared solutions were recorded against the reagent blank.

Chemometric analysis was performed with Solo_MIA 8.5.2 software from Eigenvector Research, Inc., Manson, WA USA. Partial least squares regression (PLS) was applied on UV visible spectroscopic data. Before the calibration, all the spectral data were mean-centered.

The calibration curves of caffeic acid, gallic acid and quercetin were calculated after reading the absorbance of the standard solutions, and corresponding equations were reported in Table 1. The linear ranges were 3–30 mg/L, 2–50 mg/L and 1.5–25 mg/L for caffeic acid, gallic acid and quercetin, respectively. Table 1 also presents the calibration equations at λ_max of each compound, linear range and determination coefficients.

Experimental Design. The design used for the calibration experiments maximizes the information content in the spectra. A training set of 21 samples, s1 to s21, was produced and arranged, as indicated in Table 2. In addition, seven further samples, s22 to s28, were chosen for a test set to span the calibration range. When working with three components (in this case caffeic acid, gallic acid, and quercetin), the experimental domain corresponds to a triangle. A point in this plane can identify all possible mixtures. In this work, a six concentration level design was chosen (a reduced 63 design).

The reduction was performed by choosing a subset of all possible 216 experiments with a number of properties. There are six mixtures for each compound at each concentration level. This design spans the mixture space fairly well. Each level corresponds to a concentration, the lowest level to the minimum, and the highest to the maximum.^32,33 The true concentrations of each compound at six concentration levels are given in Table 2.

Selection of Optimum Number of Factors. In this work, the cross-validation method was used to select the number of significant PLS components for modeling the system without over fitting the concentration data. The root mean square error cross-validation RMSECV obtained by optimizing the calibration matrix of the absorbance data with PLS was shown in Fig. 2. A good rule for choosing the number of PLS components to retain corresponds to the minimum of the RMSECV, so the optimum numbers of components obtained was 2.

Figure2.

Representation of the RMSECV vs. Latent Variable Number.

The relation between samples (scores) in Fig. 3 shows a triangle formed with 21 calibration samples as explained above in the experimental design. The scores on latent variable 1 (61.52%) increase with the concentration of caffeic acid (sample s1 contains a maximum of caffeic acid and a minimum of gallic acid and quercetin) while the scores on latent variable 2 (38.24%) increase with the concentration of quercetin (sample s16 contains a maximum of quercetin and minimums of caffeic and Gallic acids).

Figure3.

Representation of bi-plot scores and loadings on (LV1, LV2) plan.

The relation between variables (loadings) in Fig. 4 shows optimums in (258 nm, 290 nm, 320 nm, and 375 nm) for LV1 and (270 nm, 323 nm, and 375 nm) for LV2.

Figure4.

Loadings on LV 1 and LV 2.

We can also see similarities in some regions with the individual spectra of caffeic acid, gallic acid and quercetin shown in Fig. 1. The loadings on LV1 (61.65%) closely resembles the spectrum of caffeic acid in the range of (250 nm – 375 nm) and is opposite to the spectra of gallic acid (250 nm – 320 nm) and quercetin (250 nm – 400 nm). On the other hand, the loadings on LV2 (38.10%) closely resembles the spectrum of quercetin in the range of (320 nm – 400 nm) and is opposite to the spectrum of gallic acid in the range of (250 nm – 320 nm).

It is useful to introduce several measures of the models fit to the data and predictive power. Two general statistical parameters: Root mean square error calibration (RMSEC) and Root mean square error cross validation (RMSECV) were selected to evaluate the calibration step of the model using 21 calibration samples. Three statistical parameters: Root mean square error prediction (RMSEP), Recovery (%), and determination coefficient (R²) were selected to evaluate the prediction ability of the model for the simultaneous determination of caffeic acid (CA), gallic acid (GA), and quercetin (Q) in the test step using seven samples. The composition of the test samples on CA, GA, and Q, and the results obtained by applying PLS model to the seven test set samples are listed in Table 3. Recovery values are very close to 100% and this result verifies the good performance of the PLS model in predicting the concentrations of the phenolic compounds in ternary mixtures. Other statistical results are summarized in Table 4. The complexity of the model is very low (only two latent variables with 99.76% of variance), the values of RMSEC, RMSECV, and RMSEP are very close together, and the R² values are close to one. The results obtained allow us to confirm the successful applicability of the proposed model for the simultaneous determination of the three phenolic compounds tested.

In order to verify the robustness of the PLS model used; we applied it on the spectral data of six aqueous herb extracts. The UV spectra obtained from the different solutions were shown in Fig. 5.

Figure5.

Representation of the UV spectra of the six aqueous herb extracts.

The UV spectra of mint, thyme, basil, and laurel have two absorption maxima (at 284 nm and 324 nm), and the shape of these spectra look like that of caffeic acid. The spectrum of parsley has two absorption maxima one at 266 nm like gallic acid, and another at 332 nm which resembles that of quercetin. The chive spectrum shows two absorbance maxima; one at 266 nm like gallic acid, and the other 284 nm like caffeic acid. The band intensities of the chive spectrum are the smallest, whereas those of the mint spectrum are the greatest. The predicted concentrations of CA, GA, and Q in the studied herbs, using the PLS model, were presented in Fig. 6, they were found to be 3.918–18.055, 9.014–23.825, and 9.040–13.350 mg/g of dry sample, respectively. The mean content (mg/g of dry sample) in descending order was GA (18.307), Q (11.85), and CA (7.214).

Figure6.

Predicted concentration of CA, GA and Q in the six aqueous herb extracts.

The highest levels of CA, GA, and Q were found respectively in mint, chive, and parsley. The lowest levels of GA and Q were found in mint, and that of CA was found in chive.

Caffeic acid contents announced in other works were found to be 1.2–60 mg/100 g of dry sample in Greek aromatic plants,³⁴ 0–0.4 mg/g in thyme and spearmint,³⁵ 5.8–30 mg/100 g in thyme, and 2.8 mg/100 g in sweet basil. Gallic acid contents in other works were found to be 11.7, 0.7, and 0.9 mg/100 g fresh weight in thyme, parsley, and spearmint, respectively. Quercetin contents in other works were found to be 9 mg/kg in chives,³⁶ generally below 10 mg/kg in most vegetables,³⁷ 0−170 mg/100 g in Danish fresh herbs,³⁸ 1.3 mg/100 g and 10.4 mg/100 g in parsley and chives, respectively.³⁹ Some of the amounts of caffeic acid, gallic acid, and quercetin found in this study were different compared with other studies. This gap could be explained by the difference in the varieties and growing conditions of the herbs analyzed in these studies.