3.1. Data Acquisition for the Responses
A mixture of seven standards of the food additives and caffein was injected into the HPLC system in accordance with the BBD. The analytes were eluted in the following order of increasing retention time: 1. ACE; 2. BEN; 3. SOR; 4. SAC; 5. TAR; 6. CAF; 7. SUN; 8. ASP. Hereafter, these order numbers indicate the corresponding compounds, as cited in the peak resolution. Rs
1–2 means resolution between ACE and BEN, and so forth. The resulting Rs
values of the peaks of the eight studied analytes are presented in Table 2
. Complete separation was found for Rs
equal to 1.5 or higher. However, an Rs
with a value of 1.0 was acceptable because it produced 98% separation. Values below 1.0 were considered to give poor separation [31
]. Henceforth, specific Rs
showing some values less than 1.0 were included in the optimization process with the target of maximizing the values reaching full separation.
In addition to Rs, the total time to elute the last peak, indicating the analysis time, was also included in the optimization as the response because it should be kept as low as possible. The analysis run time ranged from 14.7 to 25.4 min. Hence, by counting the number of Rs values that that were lower than 1.0 and the analysis run time, the total number of responses for the optimization was eight.
3.2. Optimization of the Separation Method
Prior to the use of the multi-response optimization (MRO), the response surface methodology (RSM) data were calculated to generate a model for each response through regression analysis. As a result, eight models were constructed for seven Rs
values and analysis time. Each model demonstrates the empirical relationship between the three studied variables and the responses. The calculated regression coefficients are given in Table 3
, and these results explain the effects of the studied variables and interactions.
Analysis of variance (ANOVA) was used to assess the variability in responses and separate the responses into pieces for each of the effects. It was then used tested the statistical significance of each effect by comparing the mean square against an estimate of the experimental error. Variables with a p value coefficient of less than 0.05 were defined as having a significant effect on the corresponding response. The regression coefficients were then used for a final predictive equation of RSM for Rs and analysis time by using only significant variables.
The significant variables for each response were varied. The positive effect of %B initial
) was important for Rs
6–7, while Rs
4–5 and analysis time were negatively affected by %B end
). The mobile phase composition defines the affinity of analytes to the mobile phase. A stronger affinity between analytes with the mobile phase could lead to a higher elution power. %B end
indicates the use of the organic methanol in the gradient program. With lower %B end
, the elution power is reduced, resulting in better separation for SAC-TAR (Rs
4–5). Similar findings were reported for the use of methanol as eluent for food additives separation by chromatographic techniques [16
Additionally, according to the ANOVA results, both main and quadratic effects of pH significantly contributed (p
< 0.05) to the separation for most peaks (Rs
5–6, and Rs
6–7). The importance of pH in the mobile phase was considered to improve the selectivity of the chromatographic method [18
]. The positive linear term (x3
) indicated that the higher the pH, the higher the resulting resolution. The increase in pH facilitated the ionization process of acidic analytes in the polar mobile phase, which increased the solubility of the analytes. In contrast, the buffer solution with a higher pH hardly eluted the basic analytes because they were less ionized [35
]. Hence, a complete separation of adjacent peaks consisting of the acidic and less acidic analytes, e.g., BEN-SOR (Rs
2–3) and TAR-CAF (Rs
5–6), can be achieved by increasing the pH. Additionally, a significant effect of pH on the resolution is related to the available chemical forms for the compounds during the elution. If the working pH is near the pKa, more than one chemical form could be found; therefore, broad peaks or even more than one peak would be produced from one compound.
Using only the significant variables, models were built for predicting each response. The models were validated by the coefficient of determination (R2) that ranged from 0.7414 (Rs3–4) to 0.9745 (Rs1–2). In addition, the mean absolute error (MAE) ranged from 0.0830 (Rs7–8) to 0.5456 (analysis time). Because of the high values of R2 and low error, the models can be used for reliable prediction in multi-response optimization (MRO).
The MRO was employed to optimize the eight responses simultaneously. The desirability function d(y) was then built based on the values obtained for each optimized response. The MRO approach assumes the response values equal to (y) can be modeled through the d(y), where the desirability ranges from 0 to 1.
In this optimization by the desirability function, the response of analysis time was set as less important (the lowest importance with impact coefficient of 1) than Rs
(the default importance with impact coefficient of 3). These settings were due to the range of the analysis run time being adequate for a fast HPLC method. In the MRO, the importance of the responses for computational analysis was indicated by the impact coefficient given to the responses. By using the STATGRAPHICS Centurion XVI, values of the impact coefficients can be set from 1 to 5. On the basis of these settings, a 3D contour plot was built for MRO (Figure 2
). It can be seen that the maximum value for the desirability function can be found at %B initial
= 0.69 (x1
; 8.5%), %B end
= 0.49 (x2
; 90.0%), and pH = 0.88 (x3
, 6.7). Using those conditions, a complete separation was achieved in roughly 15 min (Figure 3
The optimum separation condition was then confirmed by performing the analysis of the studied analytes in a mixture of standard solution to check the selectivity of the method. The resulting chromatogram by the optimum condition suggested by the MRO is presented in Figure 3
The proposed method could completely separate the eight analytes, as indicated by the resolution values higher than 1.5. The resulting resolution values by the optimized method ranged between 1.51 (Rs
3–4) and 9.56 (Rs
5–6). Additionally, this separation method could be considered as a rapid chromatographic method because the analysis time was less than 16 min. This report is faster than previous studies since it provides faster analysis time to separate similar or even higher numbers of analytes yet with complete resolutions [18
3.3. Validation of the Separation Method
Validation of the analytical method is a procedure to prove whether an analysis method meets the specified requirements so that the results of the analysis can be justified. In this study, the system suitability test (SST) was also assessed prior to evaluating the other validation parameters. Table 4
summarizes the results of the HPLC-DAD method validation.
The results showed that the system suitability test (%CV) values of each set of parameters were less than 2%, indicating the high precision of the HPLC-DAD system. The regression of the calibration curves for all analytes provided a high coefficient of determination (R2), 0.9991 or more in the studied range 1 to 50 mg L−1. The LOD values ranged from 1.16 mg kg−1 (SAC) to 3.00 mg kg−1 (ACE), while the LOQ values ranged from 3.86 mg kg−1 (SAC) to 10.02 mg kg−1 (ACE). The low LOD and LOQ facilitate a reliable detection and quantification of seven food additives and caffeine with very low concentration in instant flavored drinks.
Recovery represents a measure that indicates the degree of closeness of the analysis results with the actual level of the analyte. The standard addition experiments resulted in recoveries that ranged from 95.30% to 101.41%. Providing recoveries near to 100%, the results indicate the confidence of the method to measure the level of the studied compounds in the sample.
Intraday and interday precision were calculated to establish the precision of the method. The CVs for intraday precision of the retention time and the signal of the area of the studied analytes, on average, were 0.28% and 2.14%, respectively, while the intermediate precisions were 0.25% and 2.53%, respectively. It was also observed that CAF and ASP have the highest precisions. Based on the acceptance values suggested by the AOAC through the International Manual for the Peer-Verified Methods Program, the proposed method has been validated due to the high precisions [30
The validated HPLC-DAD method was then applied for the simultaneous determination of seven food additives and caffeine in nine popular powdered drinks in the market. The chosen samples consisted of four powdered drinks with fruity flavor (samples 01 to 04), one powdered drink with sweet tamarin flavor (sample 05), and the rest of the samples (samples 06 to 09) were powdered drinks based on tea. The composition of the relevant ingredients of the samples is given in Table 5
Applying the HPLC-DAD method to the nine powder drink samples, seven food additives and caffeine were successfully identified and quantified (Table 5
b). The method detected and provided the levels of the analytes mentioned on the products’ labels. Although CAF was not listed as an ingredient in samples 06 to 09, it was found to be present in the products. Samples 06 to 09 consisted of tea, and this information was claimed in the label as an ingredient, which provided the natural CAF to the product.
National and international regulations are applied to control the use of food additives in powdered drinks. In Indonesia, the actual regulation is referred to as the National Agency of Drug and Food Control (NADFC) guidance. The established guidance rules for the studied additives in powdered drink products are as follows: ACE, max. 600 mg kg−1
; BEN, max. 600 mg kg−1
; SOR, max. 1000 mg kg−1
; SAC, max. 120 mg kg−1
; TAR, max. 300 mg kg−1
; CAF, max 250 mg kg−1
; SUN, max. 300 mg kg−1
, and ASP, max. 600 mg kg−1
. The amount of food additives in the nine samples assessed here were all below the limits defined by the NADFC legislation for food additives in powdered drinks [3