**4. Statistical Analysis**

Regression analysis as well as analysis of variance (ANOVA) was applied to establish the nature of the relationship between the responses and the three independent variables. To fit the regression models to the experimental data with the objective of achieving the overall optimal region for all response variables studied [33]. For all tests the *p*-value adopted was less than 0.05. The generalized polynomial model proposed for relating the response to independent variables is given below:

$$y\_i = b\_0 + b\_1 \mathbf{x}\_1 + b\_2 \mathbf{x}\_2 + b\_3 \mathbf{x}\_3 + b\_{12} \mathbf{x}\_1 \mathbf{x}\_2 + b\_{13} \mathbf{x}\_1 \mathbf{x}\_3 + b\_{23} \mathbf{x}\_2 \mathbf{x}\_3 + b\_{11} \mathbf{x}\_{12} + b\_{22} \mathbf{x}\_{22} + b\_{33} \mathbf{x}\_{32} \tag{1}$$

where *y*i represents the predicted dependent variables; b0 is the offset term (constant); b1, b2 and b3 are the linear effects; b11, b22 and b33 are quadratic effects; and b12, b13, b23, b31 and b32 are the interaction effects. The terms *<sup>x</sup>*i*<sup>x</sup>*j and *<sup>x</sup>*i<sup>2</sup> (i = 1, 2 or 3) denote the interaction and quadratic terms respectively [34,35]. The adequacy of the model was tested using model analysis, lack of fit test and coefficient of determination (*R<sup>2</sup>*) analysis.

#### **5. Optimization and Validation Procedure**

The final reduced models in optimization can be presented as 3-D response surface plots. These can reveal the significant interactive effects of the independent variables on the response [36]. Here, the relationship of each response to the independent variables was expressed with 3-D plots by fixing two variables at the centre point while varying the third within the chosen experimental range. The levels of the independent variables for achieving the optimum goal of the individual and overall responses were determined with the aid of the response optimizer.

The response optimizer allows the attainment of a fair balance in the optimization of several response variables by identifying the best combination of input variable settings that favour maximum value of response(s) [37]. It was thus applied in the current study in order to simultaneously reduce the 11 target mycotoxins. The final reduced models were verified by conducting five replicate experimental runs at the optimal settings and comparing the observed results with the predicted responses. Significant difference between the predicted and experimental results were further confirmed by one sample *t*-test Experimental design, model generation, prediction and other statistical analysis were done using a statistical package (Minitab 17 software, State College, PA, USA).

### **6. Adsorption Studies**

About 2 kg of fresh representative PKC samples were kept at 4 ◦C ahead of sample extraction and subsequent analysis. Three different concentrations (5.0, 25.0 and 100.0 ng/g) of AFB1, AFB2, AFG1, AFG2, OTA, ZEA, DON, HT-2, T-2, FB1 and FB2 standards were mixed with approximately 5 g of PKC, each in triplicate solvent evaporation was allowed to occur in the spiked samples by storing them overnight in the dark. Preliminary studies conducted within the range of 0.005 to 0.04 g of CTS revealed that decrease in mycotoxins was not observed beyond 0.035 g. Therefore, this amount was utilized in the adsorption experiments. In the sorption experiments, 350 mg of CTS adsorbent was added to 5 g of mycotoxin-contaminated PKC samples in 50 mL flasks. A 20 mL volume of solvent (acetonitrile/water/formic acid at 70:29:1, *v*/*v*/*v*) was added to the flask and the pH (3–6) was adjusted as needed using 0.01 M HCl. Adsorption was then carried out at controlled temperatures (30–50 ◦C) and at desired equilibrium times (4–8 h) under constant shaking (300 rpm). The mixture of solution was centrifuged at 3000 rpm for 10 min and 1 mL of the final solution was mixed with 3 mL of water for dilution [38]. The purpose of sample dilution during the sample preparation procedure was to reduce the possible matrix effect [39]. The extract obtained was then filtered with nylon syringe filter (0.22 μm). At the end of this process, residual mycotoxins present were measured using LC/MS-MS [40]. Adsorption was estimated based on the initial and final amounts of mycotoxins present in the aqueous, as presented in the following Equation (2) [41]:

$$\mathbf{E} = (\mathbf{C}0 - \mathbf{C}\mathbf{e})\mathbf{C}0\mathbf{\hat{}}100$$

In Equation (2), C0 is the concentration (ng/mL) of the mycotoxin in the blank control and Ce is its concentration (ng/mL) in the supernatant.

#### **7. Optimization for Maximum Mycotoxin Removal**

To identify the optimum settings of the independent variables for the desired goal of mycotoxin removal, multiple response optimizations (numerical and visual) were conducted. Two stages may be considered in optimization: (a) visualize the significant interaction effects of independent variables on the response variables and (b) the actual optimization, where the factors are further examined in order to determine the best applicable conditions. Presented in Figure 2 are the respective response plots for simultaneous removal of eight mycotoxins obtained with different settings of the studied variables. Maximum removal of all mycotoxins was predicted to occur at the optimal condition of pH 4, time 8 h and temperature 35 ◦C (Figure 2).

**Figure 2.** Response optimization, parameters, predicted response (*y*) and desirability of multi-mycotoxin by CTS.

#### **8. Reduced Response Model Validation**

Adequacy of the response-regression equations was evaluated using t-test. The corresponding experimental responses were compared with predicted values and the results are presented in Table 5. In the validation process, there must be no significant difference (*p* > 0.05) between the predicted and actual experimental values; this implies good agreemen<sup>t</sup> between the two values. This observation verifies adequate fitness of the response equations by RSM. Applying the optimum conditions predicted by the reduced models in this study, reduction for AFB1, AFB2, AFG1, AFG2, OTA, ZEA, FB1 and FB2 were 94.35, 45.90, 82.11, 84.29, 90.03, 51.30, 90.53 and 90.18% respectively. The total desirability was 0.77 as shown in Table 5. Hence, the final reduced models fitted by RSM were adequate.


**Table 5.** Comparison between predicted and experimental values based on the final reduced model.

*y*0: predicted value, *y*i: experimental value, *y*0–*y*i: residue.

Results of the recovery all the experimental response values showed that this method is acceptable to be used for mycotoxin removal with chitosan in PKC. Recovery values ranged from 81% to 112% for all mycotoxins as reported [38,42].

Results from this study have shown that CTS is promising as an adsorbent for removal of various types of mycotoxins. Its application for the removal of OTA in contaminated drinks has been previously demonstrated [9,43]. Dietary supplementation in poultry for removal of AFB1 and ZEA [44], demonstrated the e ffectiveness of CTS in reducing the levels of one or two mycotoxins. In this study, CTS showed a moderate to high adsorbent capacity against eight of the eleven mycotoxins evaluated simultaneously, though it showed poor adsorption against HT2, T-2 and DON. These findings indicate better performance against a wider range of mycotoxins than was achieved in a similar study [45] and with cross-linked chitosan [30]. This study further suggests that CTS can remarkably bind all eight of the mycotoxins assessed simultaneously.
