Development of an HPLC-DAD Method for the Quantification of Ten Compounds from Moringa oleifera Lam. and Its Application in Quality Control of Commercial Products

An HPLC-DAD separation method for the simultaneous quantification of ten compounds from Moringa oleifera plant was developed. The method was validated with pure solvent and different matrices of M. oleifera products. This method was found to be linear in the concentration range of 1 to 10 mg L−1 for all the compounds in the solvent and from 3 to 10 mg kg−1 in the different matrices. The correlation coefficients ranged between 0.9900 and 0.9999. Intra-day and inter-day variability showed that the developed method is both repeatable and precise with percent relative standard deviation values less than 10% and 20%, respectively. Limits of detection ranged between 0.06 and 0.8 mg L−1 for the solvent and 0.1–1.5 mg kg−1 for the matrices, while the limit of quantification ranged between 0.2 and 2.8 mg L−1 and 0.4–4.8 mg kg−1, respectively. The validated method was applied successfully to thirty-two different M. oleifera products, whereby all ten compounds were detected in one of the samples. Principal component analysis was used to assess the correlation and variance between the products. Variations were observed in products from different regions and from different manufacturers.


Introduction
Moringa oleifera Lam. is the most cultivated species in the Moringa genus. It is a valued natural product because of its many uses and hence it has earned names such as the miracle tree and the magic bullet. All parts of the plant are useful in some way or another, either medical, nutritional, or cosmetic [1]. The plant is used traditionally to treat cholera, diabetes, skin infections, respiratory problems, and ear infection, among others [1,2]. The seeds of the plant are also used to purify water, as demonstrated in studies reported by Ndabigengesere et al., Sanchez-Martin et al. and Suhartini et al. [3][4][5]. Its popularity has led to commercialization of products prepared from the plant.
There are efforts in emerging economies to use natural products as alternatives to synthetic drugs that are usually associated with side effects and are expensive. Growing interest in research on these products observed lately attests to this. Phytochemical studies support traditional use of the plant and validate the commercialization of M. oleifera products [6][7][8]. The phytochemical approach is generally used to evaluate the activity of the plant extracts, and thereafter, the active compounds are isolated and identified. However, sometimes the activity of plant extracts has been reported to be greater than that of the individual compounds [9]. This is probably due to the synergistic effect of the compounds within the extract, as shown in studies conducted by Brahmbhatt et al. [10]. Most commercial natural knowledge gap has been identified in terms of quality control and quality assurance measures for various natural products in the market. Quality control is important to avoid low-quality products being delivered to consumers as well as to prevent potential harmful side effects. It is therefore important to develop analytical methods for quality control of medicinal plant products. Shanker et al. developed a quantification method for two nitrile glycosides in the leaves and bark of M. oleifera [11]. Similarly, Pachauri et al. developed a quantification method for 1,3-dibenzyl urea and aurantiamide acetate found in the roots of M. oleifera [12].
The purpose of this study was to develop and validate a simple method for the simultaneous separation and quantification of ten compounds that are commonly found in different parts of M. oleifera plant (illustrated in Figure 1) in order to perform their analysis in different M. oleifera plant and commercial products (n = 32) [13][14][15][16]. HPLC-DAD was the selected method because it fulfils the purpose. Small production entities and/or companies could use this method to assess the quality of their products, which is an efficient and low-cost method with adequate sensitivity for the studied metabolites. In addition, the sensitivity offered by an HPLC-DAD is adequate for most compounds found in plants. 3,5,7-trihydroxy-2-(4hydroxyphenyl)chromen-4-one (Compound 10) To the best of our knowledge, no work has been reported on simultaneous quantitative determination of the ten compounds in M. oleifera selected for this study.

HPLC-DAD Method Development for Separation and Quantification
Compounds 1-10, illustrated in Figure 1, were used as reference standards to develop an HPLC-DAD separation method that could be used for qualitative and quantitative analysis of the Moringa oleifera products. Figure 2 shows a typical chromatogram of a baseline separation of the ten compounds eluting in the following order with retention times (tR) as given: compound 1 (tR: 1.4 min), 2 (tR: 2.1 min), 3 (tR: 3.8 min), 4 (tR: 4.5 min), 5 (tR: 4.8 min), 6 (tR: 6.5 min), 7 (tR: 7.0 min), 8 (tR: 7.6 min), 9 (tR: 7.8 min), and 10 (tR: 8.7 min), under the conditions described in Section 3.3. Baseline separation for all the compounds was achieved and allowed accurate quantification of the compounds. To the best of our knowledge, no work has been reported on simultaneous quantitative determination of the ten compounds in M. oleifera selected for this study.

HPLC-DAD Method Development for Separation and Quantification
Compounds 1-10, illustrated in Figure 1, were used as reference standards to develop an HPLC-DAD separation method that could be used for qualitative and quantitative analysis of the Moringa oleifera products. Figure 2 shows a typical chromatogram of a baseline separation of the ten compounds eluting in the following order with retention times (t R ) as given: compound 1 (t R : 1.4 min), 2 (t R : 2.1 min), 3 (t R : 3.8 min), 4 (t R : 4.5 min), 5 (t R : 4.8 min), 6 (t R : 6.5 min), 7 (t R : 7.0 min), 8 (t R : 7.6 min), 9 (t R : 7.8 min), and 10 (t R : 8.7 min), under the conditions described in Section 3.3. Baseline separation for all the compounds was achieved and allowed accurate quantification of the compounds.

Linearity and Matrix Effect
Linearity of the method was determined using the external standard calibration and matrix-match calibration methods. The calibration curves were obtained by plotting the average peak area vs. the concentration of the compounds. The calibration curves for all the compounds in the solvent and five different matrices showed a good linear relationship between peak area and concentration in the range of 1-10 mg L −1 and 3-10 mg kg −1 , respectively. This is shown by the regression coefficients (r 2 ) of the compounds, that were in the range 0.9942-0.9999 and 0.9900-0.9998 in the solvent and matrices respectively, which confirms good linearity ( Table 1). The percentage matrix effect for each compound in the different matrices were calculated using Equation (1) given in Section 3.4. As shown in Figure 3, the compounds experience either a positive or negative matrix effect. A negative matrix effect means that the detector response for the compound was decreased/suppressed, while a positive matrix effect means that the detector response was increased/enhanced [17]. The % matrix effect (ME) for the compounds in the different matrices ranged from −29% to 45%, with compound 1 experiencing the highest suppression in the teabags and the highest enhancement in the porridge sample as compared to the other compounds. Compounds 3, 5, and 10 were enhanced in all the products shown by the positive matrix effect. Overall, 60% of the compounds were enhanced and 40% were suppressed in the different products, as shown by the positive and negative matrix effects, respectively. Matrix effect is classified as mild, medium, or strong when % ME is <±20%, ≥±20% but <±50% and ≥±50%, respectively. In the leaves, porridge, and teabags matrices, 60% of the compounds experienced mild matrix effect while 40% experienced medium matrix effect (see Supplementary Figure S1). With regards to the seeds, 80% and 20% of the compounds experienced mild and medium matrix effect, respectively (Supplementary Figure S1). On the other hand, for the pill matrix, there was equal distribution of mild and medium matrix effects (Supplementary Figure S1). Strong matrix effect was not observed for any compound in all the products. This means that the matrix effect experienced by the compounds in the different products is not significant and hence calibration equations in the solvent can be used for quantification. However, matrix-matched calibration equations were used to give more accurate results.

Linearity and Matrix Effect
Linearity of the method was determined using the external standard calibration and matrixmatch calibration methods. The calibration curves were obtained by plotting the average peak area vs. the concentration of the compounds. The calibration curves for all the compounds in the solvent and five different matrices showed a good linear relationship between peak area and concentration in the range of 1-10 mg L −1 and 3-10 mg kg −1 , respectively. This is shown by the regression coefficients (r 2 ) of the compounds, that were in the range 0.9942-0.9999 and 0.9900-0.9998 in the solvent and matrices respectively, which confirms good linearity ( Table 1). The percentage matrix effect for each compound in the different matrices were calculated using Equation (1) given in Section 3.4. As shown in Figure 3, the compounds experience either a positive or negative matrix effect. A negative matrix effect means that the detector response for the compound was decreased/suppressed, while a positive matrix effect means that the detector response was increased/enhanced [17]. The % matrix effect (ME) for the compounds in the different matrices ranged from −29% to 45%, with compound 1 experiencing the highest suppression in the teabags and the highest enhancement in the porridge sample as compared to the other compounds. Compounds 3, 5, and 10 were enhanced in all the products shown by the positive matrix effect. Overall, 60% of the compounds were enhanced and 40% were suppressed in the different products, as shown by the positive and negative matrix effects, respectively. Matrix effect is classified as mild, medium, or strong when % ME is <±20%, ≥±20% but Both limit of detection (LOD) and limit of quantification (LOQ) were determined using the linear regression equation according to the ISO11843 method, where LOD is signal-to-noise ratio (S/N) × 3, and LOQ is signal-noise-ratio (S/N) × 10. Table 1 shows LODs and LOQs in the range 0.1-0.8 mg L −1 in the solvent, while LOQs were in the range 0.2-2.8 mg L −1 . For the leaves, the LODs ranged from 0.5 to 1.4 mg kg −1 , while LOQs were 1.5-4.8 mg kg −1 . For the remaining products (i.e., porridge, pill, seeds, and teabag), the LODs and LOQs ranged from 0.1 to 1.2 mg kg −1 and 0.4 to 4.6 mg kg −1 , respectively. Table 1. Linear regression equations (correlation coefficients) where Y is the response of the compound and x is the concentration of the compound, limits of detection # , and limits of quantification * for the solvent and different matrices.   in the solvent, while LOQs were in the range 0.2-2.8 mg L −1 . For the leaves, the LODs ranged from 0.5 to 1.4 mg kg −1 , while LOQs were 1.5-4.8 mg kg −1 . For the remaining products (i.e., porridge, pill, seeds, and teabag), the LODs and LOQs ranged from 0.1 to 1.2 mg kg −1 and 0.4 to 4.6 mg kg −1 , respectively.

Recovery, Precision, and Uncertainty Measurements
The percentage recoveries for the compounds at the lowest spiking concentration (3 mg kg −1 ) for the compounds in the leaves, porridge, pill, seeds, and teabags ranged between 75. 7 Table 2. The recoveries in different matrices had acceptable random errors, as indicated by the percent relative standard deviation (%RSD) values which were below the 20% threshold as stipulated by the EU Commission Decision 2002/657/EC directive [18]. These results indicate that the method is accurate.
Intra-day and inter-day variability were determined using the lowest, middle, and highest concentrations. The results were expressed as percent relative standard deviation (%RSD). Table 2 shows intra-day results, where %RSD values for all the compounds in the solvent were found to be in the range of 0.3-1.1%, 0.2-0.8%, and 0.2-0.4% at the lowest, medium, and highest concentrations. The %RSD values for the compounds in the different matrices were 0.2-4.7, 0.2-3.2, and 0.1-1.4 at the lowest, medium, and highest concentrations, respectively. Inter-day variability was used to assess reproducibility over five consecutive days. The %RSD values (Table 2) obtained for the solvent ranged between 0.2% and 1.6%, 0.05% and 0.6%, and 0.02% and 0.2% for the lowest, medium, and highest concentrations. In the different matrices, the %RSD values for the compounds ranged between 0.3 and 6.3, 0.1 and 4.4, and 0.2 and 4.6 for the lowest, medium, and highest concentrations. The values obtained for intra-day and inter-day variability were below 10% and 20% respectively, which is considered acceptable and indicates that the method is both repeatable and reproducible according to the EU Commission Decision 2002/657/EC directive [18]. Table 2. Percentage recoveries, (percentage relative standard deviation (%RSD)) Intra-day + , (%RSD) Inter-day *, and measurement uncertainty for the compounds in the solvent and different matrices. The measurement uncertainty due to precision and recoveries were calculated using Equations (3)-(5), the results are given in Table 2. The results are reported at the 95% confidence level. The measurement uncertainty ranged from 0.3% to 9.8% with compound 2 exhibiting the highest uncertainty in the teabag matrix. This is because of the low recovery of the compound at the lowest concentration level. Compound 3 showed the lowest uncertainty in the leaves which is due to the good recovery and %RSD of the compound in the matrix.

Specificity
Specificity of the method was determined using the DAD Chemstation software to assess the purity of the compounds of interest in the different matrices. Their Ultraviolet (UV) spectra and retention times were also compared with those in the pure solvent. The purity indices were in the range of 897 to 998 for all the compounds in the different matrices, meaning that they were not co-eluting. The UV spectra and retention times of all the peaks corresponding to the compounds of interest in the different matrices were similar to those in the solvent with standard deviations between 0.02 and 0.14. This showed that the method was specific for the compounds of interest.

Quantification of the Compounds in the Plant and Products' Extracts
The performance of the developed and validated method was tested on M. oleifera products which included the pills (P), porridge (PRD), teabags (TB), leaves (L), and seeds (S) from different regions ( Table 3). Table S2 shows the information about when and where the leaves and seeds samples were collected. Retention times and UV spectra of the identified compounds were compared with those of the reference standards (See Figure 4 below for a chromatogram representing the sample L4 and Supplementary Figures S3-S8 for chromatograms of the standards and selected samples plus UV spectra of the peaks of interest). The matrix-matched linear regression equations were used to calculate the concentrations of the compounds in the different products. It is noted that even though all the compounds in this study have been reported previously in the seeds and leaves (except for compound 9 which has only been reported in the seeds), only compounds 6 and 7, known for their anticancer properties, were detected in all the samples. Compound 6 was present in higher amounts in leaves L3 from Limpopo, South Africa (9.7 mg kg −1 ), while the PRD1 from Manufacturer I contained the highest amount of compound 7 (9.6 mg kg −1 ). As much as compounds 6 and 7 were detected in all the samples, they could not be quantified in S6, PRD3 and PRD4 samples because they were present in quantities below LOQ. L4 from Mpumalanga contained all the ten compounds, however, five out of the ten were below the LOQ. L5, TB1, TB2, and TB3 samples contained five compounds each, corresponding to the least number among all 32 samples. There are some prominent peaks which were observed in the matrices but could not be identified due to insufficient data. These prominent peaks were observed at retention times of 1.2 to 1.4 min in the samples L4, P4, and PRD1. Other prominent peaks were observed at 5.2 to 5.8 min in L4, P4, S2, PRD1, and TB7 (See Supplementary Figures S4-S8). Table 3. Quantification data for application of the validated method to real samples (reported in mg kg −1 of the compound per 280 mg kg −1 of the sample).

Region
Compound 1  all the ten compounds, however, five out of the ten were below the LOQ. L5, TB1, TB2, and TB3 samples contained five compounds each, corresponding to the least number among all 32 samples. There are some prominent peaks which were observed in the matrices but could not be identified due to insufficient data. These prominent peaks were observed at retention times of 1.2 to 1.4 min in the samples L4, P4, and PRD1. Other prominent peaks were observed at 5.2 to 5.8 min in L4, P4, S2, PRD1, and TB7 (See Supplementary Figures S4-S8).   The variance and correlation between the different products were assessed using rincipal component analysis (PCA) and hierarchical clustering. Figure 5 shows a PCA biplot of the products and the compounds. The plot shows that PC1 explains 30.3% of the variance while PC2 explains 25.1%. The leaves in general apart from L9 are negatively correlated to the seeds along the PC1 axis. This is also observed on the hierarchy clustering plot where groups 2 and 4 containing these samples are correlated at a lower level (Supplementary Figure S2). The leaves from Mpumalanga are clustered together because they all contain significant amounts of compound 3. The same is true for PRD3 and PRD4 which contain compound 10. PC2 shows the difference between the leaves from Limpopo and those from Mpumalanga, though the variation is not significant. The significant difference between the leaves from Ethiopia and those from South Africa is shown by their separation along PC 1. This difference could be due to geographical effects. The variation between manufacturers of the same products was also assessed by the PCA plot. The plot shows that even though products from the same manufacturer are closely correlated, there is a significant variation between products from different manufacturers. For example, P1 and P2 (from MI) are negatively correlated to P5 and P6 (from MIII) along PC1, which is also confirmed by the hierarchical cluster plot (Supplementary Figure S2) where these are correlated at a lower level (groups 3 and 4). These results prove the need for quality control methods for the products in the market.

Materials and Methods
All the solvents used in this work were of HPLC grade with a purity > 99.9%. Some of the reference standards (Compounds 6, 7, and 9) were isolated in our laboratory (see Supplementary Table S1) with a percentage purity ranging from 90 to 99 [19]. The rest were purchased from Sigma-Aldrich (Steinheim, Germany) and had percentage purity between 95 and 97. The commercial M. oleifera products used as samples such as the pills, porridge, and teabags were purchased from local pharmacies. The leaves and seeds were collected from different regions (see Table S2) and were authenticated by the College of Agriculture and Environmental Sciences at the University of South Africa. The aqueous solutions were prepared using ultra-high-purity (UHP) water (18.2 mΩ) from a Milli-Q water purification system (Molsheim, France) and filtered using a 0.22 µm membrane filter from Sigma-Aldrich (Steinheim, Germany).

Instrumentation
The compounds were extracted using a Microwave reaction system: Multiware 500 (Anton Paar, Graz, Austria). The HPLC method for separation and quantification of the compounds was developed using an Agilent HPLC 1260 system (Agilent Technologies, Waldbronn, Germany) which consisted of a binary high-pressure pump, autosampler, a thermostatted column compartment, a diode array detector, and a fluorescence detector. Instrument control, data collection, and processing were achieved using the Chemstation (version 1.9.0) software.

Extraction Procedure
The leaves and seeds were grinded using mortar and pestle to give fine powders. Two grams of each sample were extracted using an optimized microwave-assisted extraction method for phenolic compounds in M. oleifera reported by Rodriguez-Pérez et al. [20]. The extraction was performed for each sample using 20 mL of water:ethanol (58:42 v/v) at a temperature of 158 • C for 20 min. The extracts were then filtered using Whatman filter paper and dried using a freeze-drier. The dried extracts were stored at 4 • C until further use.