Many of the reagents were obtained from Sigma-Aldrich Corporation (St. Louis, MO, United States), including aluminum chloride, cholesterol, copper(II) acetate monohydrate, copper(II) sulphate, 3,5-dinitrobenzoic acid, Dragendorff reagent, gallic acid, iron(III) chloride, magnesium ribbon, methanol, 1-naphtol, ninhydrin, nitric acid, picric acid, pyridine, quercetin and sulphuric acid. Glacial acetic acid, chloroform, glucose, Lugol’s iodine solution and sodium chloride were purchased from Rochelle, and blood agar plates (Selecta-Media Columbia Agar, ThermoScientific) were donated by the Biomedical Technology division of the Department of Health Sciences (VUT, South Africa). Acetic anhydride, aspartic acid, ethanol, gelatin, glycine, hydrochloric acid, lead acetate 3-hydrate, potassium acetate, potassium iodide, sodium carbonate anhydrous, sodium citrate dihydrate, sodium hydroxide, sucrose and tyrosine were obtained from Merck (Germany). Iodine and mercury(II) chloride were obtained from Thomas Baker Chemicals (Mumbai, India) and Radchem (Pty) Ltd (Gauteng, South Africa), respectively. All chemicals were analytical grade. We used Ultrapure water that was obtained from a Millipore Direct-Q water purifier system (Merck Millipore, Darmstadt, Germany).

The plant material for this study was collected from the dunes along the Bloubergstrand coast (latitude: −33°47’49.92” S and longitude: 18°27’43.20” E) in the Western Cape Province, South Africa. The plant was authenticated by Professor Stefan Siebert, curator of the AP Goossens Herbarium (Northwest University, Potchefstroom), where two voucher specimens (numbers 148 300 and 148 301) are housed. The fresh leaves of the plant were prepared as described by Laloo et al. (2024). The dried pulverised plant material was stored at −20°C in an airtight container for further analysis.

The variables affecting the two solvents used to extract phytocompounds from C. edulis were screened using a factorial design comprising three independent variables, pH, extraction time and temperature, each at two levels. Box 1 shows the independent variables, their designated letters and the levels used in this study.

The solid-to-solvent ratio (1:20) was kept constant for all experimental runs. The experimental runs were randomised to minimise the effects of unexpected variability in the observed response. The extraction yield (%) was the dependent variable or response.

The independent variables (pH, extraction time and temperature) for each experimental run were chosen as dictated by the experimental design matrix (Table 1).

A shaking incubator (Labotec, Model 355, Johannesburg, South Africa) was used for the extractions at 150 rpm. After filtering the extracts through Whatman No. 1 filter paper, the aqueous filtrates were lyophilised, and the methanolic extracts were concentrated using a rotary evaporator (HB 10 basic, IKA^®-Werke GmbH & Co. KG, Staufen, Germany). The concentrated methanolic extracts were then dried further until the weight remained constant. The extracts were stored at −20°C in dark containers until further analysis. The extraction yield (%) was calculated using Equation 1: %Extraction yield= Amount (g) of dried crude extract obtainedAmount (g) of finely ground plant material usedx100[Eqn 1]

The experiments were performed in triplicate to estimate experimental error, reduce noise and minimise bias in the observed response values (Vaux, Fidler & Cumming 2012). The Design Expert version 11 (Stat-Ease Inc., MN, United States) statistical software was used to analyse the response variables, as described in Laloo et al. (2023) The following equation expressed the regression model in this study (Equation 2): y=β0+β1A+β2B+β3C+β12AB+ β13AC+β23BC+β123ABC+ε[Eqn 2]

In Equation 2, the predicted response is represented by y, and β_n represents the regression coefficient correlated with variable n. The regression coefficients were obtained from the analysis of the experimental data. The main variables are represented by A, B and C, the two-way interactions are by AB, AC and BC, the three-way interaction by ABC and the experimental error by e. Analysis of variance (ANOVA) was used to determine the statistical significance (p < 0.05) of the model, and the accuracy was determined using the coefficients of determination (R²). The Fisher F-test was used to determine whether the constructed models were adequate to describe the observed response values. The relationship between the independent variables (main effects), their intervariable effects (interaction effects) and the response variable was demonstrated through statistical plots.

Qualitative phytochemical screening was performed on all reconstituted (50 mg/mL) aqueous (Terblanche et al. 2017) and methanolic crude extracts using standard methods (Godghate, Sawant & Sutar 2012; Nayak et al. 2011; Obouayeba et al. 2015; Samejo et al. 2013). The following phytochemicals were analysed: carbohydrates, proteins, amino acids and secondary metabolites, including alkaloids, anthocyanins, diterpenes, flavonoids, glycosides, phenols, phytosterols, saponins, tannins and terpenoids.

Ethical clearance to conduct this study was obtained from the Vaal University of Technology Faculty Research Ethics Committee (No. FACSREC-16102020-A00015).

This study determined how varying the pH, extraction time and temperature can affect the extraction yield of phytochemicals from the leaves of C. edulis using maceration. A factorial design shows the influence of the independent variables and their interactions. Table 1 shows the design matrix with the values of the independent variables, the experimental response values (extraction yield as a percent) and predicted values (based on the regression analysis) for each of the 16 experimental runs (8 aqueous extracts and 8 methanolic extracts).

Conspicuously higher extraction yields (41.70% – 64.21%) were obtained for the methanolic extracts than the aqueous extracts (21.55% – 31.11%). The amphiphilic nature of methanol can form intermolecular forces between polar and nonpolar functional groups, resulting in the solubilisation and extraction of a larger range of phytochemicals (Tiwari et al. 2011). Lapornik, Prošek and Wondra (2005) stated that methanol is more efficient in the degradation of cell walls, which have nonpolar characteristics, releasing more phytochemicals from the cells. Methanol was also found to be a better extraction medium by Magar et al. (2023) and Nortjie et al. (2024). The observation from this study underscores the importance of extracting solvents in medicinal plant research.

Statistical plots (i.e. Pareto plots, main effect and interaction plots) for the water and methanol extraction conditions were generated to: (1) identify the statistically significant variables; and (2) describe the effect of each of the independent variables and their interactions on the percent extraction yield.

The weight and statistical significance of the independent variables and the intervariable interactions are displayed on Pareto charts (Figure 1), main effect plots (Figure 2) and interaction plots (Figure 3).

According to Ozbay and Yargic (2015), Pareto charts show the absolute values of the standardised effects in decreasing order and establish two limit lines, namely, the Bonferroni limit line (t-value of effect = 3.08209) and the t limit line (t-value of effect = 2.11991), which defines the minimum statistically significant effect magnitude for the 95% confidence level. Regression coefficients with t-value effects above the Bonferroni line are labelled as significant, while those with t-value effects below the t-limit line are considered statistically nonsignificant (Shah & Pathak 2010). If the t-value of the effect of a regression coefficient lies between the Bonferroni and t-limit lines, the coefficient is likely to be significant. Main effect and interaction plots refer to simple line graphs obtained from connecting the mean values of each treatment. An effect with a zero-slope horizontal line (main effect plots) and an interaction displaying parallel lines (interaction plots) were interpreted as not significant effects, whereas increasing (positive effect – the response is higher at the high level) or decreasing (negative effect – the response is lower at the high level) effect lines were considered significant.

Figure 1 for the aqueous extraction conditions revealed that the extraction time (B) had the most significant effect in a negative mode (Figure 2) on the extraction yield (%), whereas it showed a positive significant effect (Figure 2) on the percent extraction yield (%) for the methanolic extraction conditions. The temperature (°C) had the most significant effect in a positive mode (Figure 2) on the extraction yield (%) for the methanolic extraction conditions (Figure 1) and exhibited a significantly positive effect (Figure 2) for the aqueous extraction conditions. The interactive effects, AB (pH*extraction time) and BC (extraction time*temperature), demonstrated both negative significant effects (Figure 3) for the aqueous extraction conditions but displayed no significant effects (Figure 3) on the extraction yield (%) for the methanolic extraction.

The contribution of pH (A) was positively significant (Figure 2) for the methanolic extraction conditions but was not significant (Figure 2) for the aqueous extraction conditions. The interaction between pH and temperature (AC) exhibited a weak, positively significant effect (Figure 3) on the extraction yield for the aqueous extraction conditions but had no significant role (Figure 3) during the methanolic extraction conditions. The three-way interaction ABC (pH*extraction time*temperature) was above the Bonferroni limit and displayed a significantly negative effect on the extraction yield (%) for both the aqueous and methanolic extraction conditions.

After fitting the obtained response values for each extraction condition tested to the regression model equation (Equation 2) and performing ANOVA to determine the accuracy of the suggested models, the p-values for the three independent variables and the intervariable interactions were obtained and are presented in Table 2 (aqueous), Table 3 (aqueous), Table 4 (methanolic) and Table 5 (methanolic), respectively. The variables and interactions with p-values greater than 0.05 were considered not significant and excluded from the model development. These mathematical models, based on the regression equation (Equation 2), were built and are presented in Equation 3 (aqueous conditions) and Equation 4 (methanolic conditions):    y^=β0−β2B+β3C−β12AB+β13AC−     β23BC−β123ABCYield(%)=24.81−1.62*E+0.8871*T−1.21*pH*     ET+0.6838*pH*T−1.02*ET*T−     0.8738*pH*ET*T[Eqn 3]  y^=β0−β1A+β2B+β3C−β123ABCYield(%)=52.88+3.79*pH+1.77*     ET+6.92*T−1.31*pH*ET*T[Eqn 4] where ŷ represents the predicted extraction yield, ET the extraction time and T the temperature.

The regression coefficients (β_n) were obtained from the ANOVA and describe the potential influence of each of the independent variables and their interactions on the response value (Montgomery 2013). The higher the value of the regression coefficient, the more significant the corresponding effect (Alcheikh Hamdon, Darwish & Hilal 2015). The positive sign of the regression coefficients for the aqueous extraction condition (β₀, β₃ and β₁₃) and the methanolic extraction condition (β₁, β₂ and β₃) is indicative of a synergistic effect, whereas the negative sign of the regression coefficients β₂, β₁₂, β₂₃ and β₁₂₃ (aqueous) and β₁₂₃ (methanolic) denotes an antagonistic effect.

The Fisher F-test was used to evaluate the statistical significance of the constructed regression models. The F-values of 24.06 for the full model (Table 2) and 29.46 for the reduced model (Table 3), both with p-values of < 0.0001, suggested that the models were highly significant in describing the response values for the aqueous extraction. The improved F-value obtained for the reduced model was because of the elimination of the nonsignificant model terms.

The coefficient of determination (R²) and adjusted coefficient of determination (Radj2) were used to validate the accuracy of the model. The coefficient of determination (R²) determines the proportion of variability in the response value that can be explained by the independent variables, indicating the proximity of the predicted values to the fitted regression line (Ogee et al. 2013). The R² for the full (R² = 0.9132) and reduced (R² = 0.9123) models (aqueous extraction conditions) indicated that the predicted values from the models are a good fit.

Note that the accuracy of the model should not solely be based on R², as R² always increases with an increase in the number of model variables, even if these variables are not significant (Kukreja et al. 2011). The presence of multiple variables and intervariable effects warrants the evaluation of the adjusted R² values, which compensates for the introduction of additional variables on the normal R² calculation by only increasing the value, should the additional variables significantly contribute to the model (Grant & Kenton 2019). The value of Radj2 for the full (Radj2 = 0.8753) and reduced (Radj2 = 0.8813) models, even though lower than R², still affirms an acceptable model fit.

Similar to the validation criteria used for the aqueous extraction conditions, F-tests, R² and Radj2 were used to ascertain the significance and validity of the regression models for the methanolic extraction conditions. The F-values of 58.96 for the full model (Table 4) and 107.17 for the reduced model (Table 5), both with p-values of < 0.0001, affirm the statistical significance in describing the response values for the methanolic extraction conditions. The R² for the full (R² = 0.9627) and reduced (R² = 0.9576) models (methanolic extraction conditions) points to a better model prediction than that of the aqueous extraction conditions. This is underscored by the improved Radj2 calculations (Radj2 = 0.9464) and reduced (Radj2 = 0.9486) for the methanolic extraction models.

The regression models for maximum extraction yield were optimised by using the maximum values of the aqueous and methanolic extraction conditions. The optimal yield for the aqueous model was obtained at a pH of 9, an extraction time of 72 h and a temperature of 40°C. The methanolic model produced the best yield at a pH of 9, extraction time of 168 h and temperature of 40°C.

Table 6 shows the results of the qualitative phytochemical screening of the extracts, showing the impact of the various extraction conditions. The extraction variables did not affect the presence of carbohydrates (Molisch and Benedict’s test), phenols, tannins, flavonoids, cardiac glycosides, diterpenes, phytosteroids and triterpenoids because they were present in all extracts under all experimental conditions. Many of the therapeutic properties of C. edulis are attributed to the presence of these metabolites. Phytochemical tests for the xanthoproteic reaction, ninhydrin alkaloids, anthraquinone glycosides, leucoanthocyanins, haemolysin test for saponins and phlobatannins were negative for both extracts. On the contrary, anthocyanins, starch, phytosteroids, triterpenoids, the Biuret test for proteins and saponins were present in some of the experimental runs. The sulphur test showed the presence of proteins only in the methanolic extracts. The presence or absence of these phytochemicals provides evidence that a factorial design is essential in obtaining a substantial quantity of a desired active compound from plants. The results also show that certain solvents are required to isolate specific compounds, as shown in the case of sulphur in this study. The negative results for starch, aromatic amino acids, free amino acids, alkaloids, anthocyanins, leucoanthocyanins, anthraquinone glycosides, saponins and phlobatannins may be because of the choice of extraction solvent or the absence of these compounds in detectable quantities in the extracts.