High Density Supercritical Carbon Dioxide for the Extraction of Pesticide Residues in Onion with Multivariate Response Surface Methodology

The excessive use of pesticides is a serious health problem due to their toxicity and bioaccumulation through the food chain. Due to the complexity of foods, the analysis of pesticides is challenging often giving large matrix effects and co-extracted compounds. To overcome this problem, a selective and “green” supercritical fluid extraction method was developed, using neat carbon dioxide as a solvent at pressures of up to 800 bars. A Box–Behnken response surface experimental design was used, with the independent variables of density (0.70−1.0 g mL−1), temperature (40−70 °C), and volume (10−40 mL) of solvent, and the dependent variable of extracted amount of pesticides. The optimum extraction condition was found at the use of 29 mL of supercritical CO2 at 0.90 g mL−1 and 53 °C (corresponding to 372 bars of pressure). It was observed that increasing the density of CO2 significantly increased the extraction recovery of endrin and 2,4′-dichlorodiphenyldichloroethane. Matrix-matched calibration curves showed satisfactory linearity (R2 ≥ 0.994), and LODs ranged from 0.2 to 2.0 ng g−1. Precision was lower than 11% and recoveries between 80%–103%. Thus, the developed method could efficiently be used for trace analysis of pesticides in complex food matrices without the use of organic solvents.


Introduction
The use of chemical pesticides in modern agriculture to ensure food security has increased over the past decades [1,2]. Organochlorine insecticides and s-triazine herbicides for effective control of a variety of insects and weeds are commonly used in developing countries [3,4]. The extensive application of these compounds and their degradation products contaminate the environment and agricultural products [5,6]. Moreover, water-soluble compounds are transported to surrounding water and eventually enter the food chain [7,8]. Lipophilic compounds like organochlorine can also concentrate in fatty tissues and bioaccumulate in the food chain [2].
Although applications of organochlorine and s-triazine pesticides have been banned in developed countries, developing countries like Ethiopia are still using them for agricultural and medicinal purposes. Studies carried out in various agro-industries in Ethiopia showed the presence of contaminant pesticide residues in different samples collected from various localities of Ethiopia [6]. The obtained results revealed the presence of 4,4'-dichlorodiphenyltrichloroethane (4,4´-DDT) in

Results and Discussion
In this study, what is unique is that ultrahigh pressure SFE (UHPSFE) is used for the extraction of pesticides, atrazine, 2,4´-DDD, 4,4´-DDT and endrin, from a food sample, in this case onion. It is well-known that sc-CO2 of very high pressure and density enable extraordinary solubility or even complete miscibility of some lipids [29]. In our previous research, we have demonstrated that Moringa oleifera and Moringa peregrina seed lipids could be efficiently extracted using UHPSFE, giving a clean oil extract without any solvent residues [30]. Results in the current study show that UHPSFE followed by analysis with GC/MS is a green method using minimal amounts of organic solvents.

Multivariate Parameter Optimization
A Box-Behnken experimental design was used to study three main independent variables: density, temperature, and total volume of supercritical carbon dioxide used during the extraction, which all could potentially affect the extractability of pesticides from onion. In the design of experiment (DoE), density was used rather than pressure, since the latter does not have an impact on the extraction efficiency, while density does. If pressure and temperature are used in a DoE, the density values will be non-linearly distributed between the experimental points, which is problematic. The total volume of extraction solvent was used in the DoE instead of using extraction time at a constant flow rate, which is due to the fact that the flow rate was difficult to control at the high pressures used in this study. The experimental design chosen for the optimization provides a reduced number of experiments (15 as compared to 27 (3 3 ) experiments using a full factorial design) without loss of significant information [28]. The investigated range for each variable and the obtained responses in each experiment are given in Table 1 and Table 2, respectively. The investigated ranges were selected based on the limitation of the extraction instrument. Other parameters implicated in the extraction were kept constant, i.e. the amount of dried sample (1.0 g) and flow rate (3.0 mL min -1 ). Each experiment was done by spiking 1.0 g of onion sample at 500 µg kg -1 concentration level.
The detected peak area for each compound in the experiments (based on GC/MS analysis) was converted to recovery and treated as a response in the design. Figure S1 indicates the summary of model fit using Multinomial Logistic Regression (MLR). The high values for R 2 and Q 2 indicate that the quadratic equation well represents the system under the given experimental domain. All compounds gave valid models since all R 2 are greater than 0.9 and Q 2 are greater than 0.65. It is also evident from the fact that the linearity plot depicted in Figure S2 indicates a satisfactory correlation between the observed and predicted recovery of all target pesticides. The data points clustering around the diagonal line show a good fit of the model.
The coefficient plot in Figure 2 shows the effect of individual variables and their interactions on the extraction efficiency of pesticides from an onion sample. The figure clearly indicates that the density of sc-CO2 has a positive and significant effect on the extraction recovery of 2,4´-DDD and endrin. For both compounds, this could be explained by the fact that these two compounds are more

Results and Discussion
In this study, what is unique is that ultrahigh pressure SFE (UHPSFE) is used for the extraction of pesticides, atrazine, 2,4´-DDD, 4,4´-DDT and endrin, from a food sample, in this case onion. It is well-known that sc-CO 2 of very high pressure and density enable extraordinary solubility or even complete miscibility of some lipids [29]. In our previous research, we have demonstrated that Moringa oleifera and Moringa peregrina seed lipids could be efficiently extracted using UHPSFE, giving a clean oil extract without any solvent residues [30]. Results in the current study show that UHPSFE followed by analysis with GC/MS is a green method using minimal amounts of organic solvents.

Multivariate Parameter Optimization
A Box-Behnken experimental design was used to study three main independent variables: density, temperature, and total volume of supercritical carbon dioxide used during the extraction, which all could potentially affect the extractability of pesticides from onion. In the design of experiment (DoE), density was used rather than pressure, since the latter does not have an impact on the extraction efficiency, while density does. If pressure and temperature are used in a DoE, the density values will be non-linearly distributed between the experimental points, which is problematic. The total volume of extraction solvent was used in the DoE instead of using extraction time at a constant flow rate, which is due to the fact that the flow rate was difficult to control at the high pressures used in this study. The experimental design chosen for the optimization provides a reduced number of experiments (15 as compared to 27 (3 3 ) experiments using a full factorial design) without loss of significant information [28]. The investigated range for each variable and the obtained responses in each experiment are given in Tables 1 and 2, respectively. The investigated ranges were selected based on the limitation of the extraction instrument. Other parameters implicated in the extraction were kept constant, i.e., the amount of dried sample (1.0 g) and flow rate (3.0 mL min −1 ). Each experiment was done by spiking 1.0 g of onion sample at 500 µg kg −1 concentration level.
The detected peak area for each compound in the experiments (based on GC/MS analysis) was converted to recovery and treated as a response in the design. Figure S1 indicates the summary of model fit using Multinomial Logistic Regression (MLR). The high values for R 2 and Q 2 indicate that the quadratic equation well represents the system under the given experimental domain. All compounds gave valid models since all R 2 are greater than 0.9 and Q 2 are greater than 0.65. It is also evident from the fact that the linearity plot depicted in Figure S2 indicates a satisfactory correlation between the observed and predicted recovery of all target pesticides. The data points clustering around the diagonal line show a good fit of the model.
The coefficient plot in Figure 2 shows the effect of individual variables and their interactions on the extraction efficiency of pesticides from an onion sample. The figure clearly indicates that the density of sc-CO 2 has a positive and significant effect on the extraction recovery of 2,4´-DDD and endrin. For both compounds, this could be explained by the fact that these two compounds are more polar than the others (see Figure 1) and that the polarity of sc-CO 2 increases slightly with density. Furthermore, the extraction temperature has a negative and significant effect upon the extraction efficiency of atrazine and 4,4´-DDT. A likely explanation is that these two compounds have the lowest boiling point of the four compounds investigated here, which means that they are more prone to losses during the collection step of UHPSFE. It has been previously shown that collection in SFE may introduce negative bias [31].  polar than the others (see Figure 1) and that the polarity of sc-CO2 increases slightly with density. Furthermore, the extraction temperature has a negative and significant effect upon the extraction efficiency of atrazine and 4,4´-DDT. A likely explanation is that these two compounds have the lowest boiling point of the four compounds investigated here, which means that they are more prone to losses during the collection step of UHPSFE. It has been previously shown that collection in SFE may introduce negative bias [31].   The regression models obtained were used to calculate the response surfaces as contour plots for each variable separately (Figure 3 and Figure S3). Figure 3 shows response contour plots for the analyte recoveries, and the plots given were used for interpreting the variations of recovery as a function of each pair of the independent variables. The regression models obtained were used to calculate the response surfaces as contour plots for each variable separately (Figure 3 and Figure S3). Figure 3 shows response contour plots for the analyte recoveries, and the plots given were used for interpreting the variations of recovery as a function of each pair of the independent variables.  From the graph in Figure 3, the interaction of temperature and density can be seen; the extractability of endrin and to some extent also 2,4'-DDD is assisted by an increasing temperature, while an increasing density has a positive effect for the other analytes. This is logic, considering the higher boiling point of endrin, followed by 2,4'-DDD, and then 4,4´-DDT and atrazine. An optimizer function based on a simplex algorithm with a non-linear desirability function was applied to find the conditions where the highest responses are existing, giving equal weigh for all analytes. It turned out that sc-CO2 density of 0.90 g mL -1 at 53 °C (372 bars of pressure) and 29 mL of sc-CO2 was the optimum extraction condition for maximum recoveries.

Analytical Performance of the New Extraction Method
The proposed method was evaluated in terms of linear range, LODs, precision, and accuracy using onion samples spiked with the analytes under study. The obtained results are summarized in Table 3. Matrix-matched calibration curves were established using onion samples spiked at five different concentration levels (7.8, 125, 250, 500, and 2000 µg kg -1 ) for all analytes except endrin (31.2, 125, 250, 500, and 2000 µg kg -1 ) and treated following the developed sc-CO2 extraction procedure.
The results show that the proposed method is linear with R 2 ranging between 0.994-0.999. LODs ranging between 0.2 and 2.0 ng g -1 were achieved. Precision studies were carried out in order to evaluate the repeatability (intra-day precision) and reproducibility (inter-day precision) of the proposed sc-CO2 extraction with the GC-MS detection method. Repeatability and reproducibility were assayed by analyzing spiked samples at 500 µg kg -1 concentration levels. Results expressed as relative standard deviation of response (%RSDs) are shown in Table 3, demonstrating values lower than 11%. From the graph in Figure 3, the interaction of temperature and density can be seen; the extractability of endrin and to some extent also 2,4'-DDD is assisted by an increasing temperature, while an increasing density has a positive effect for the other analytes. This is logic, considering the higher boiling point of endrin, followed by 2,4'-DDD, and then 4,4´-DDT and atrazine. An optimizer function based on a simplex algorithm with a non-linear desirability function was applied to find the conditions where the highest responses are existing, giving equal weigh for all analytes. It turned out that sc-CO 2 density of 0.90 g mL −1 at 53 • C (372 bars of pressure) and 29 mL of sc-CO 2 was the optimum extraction condition for maximum recoveries.

Analytical Performance of the New Extraction Method
The proposed method was evaluated in terms of linear range, LODs, precision, and accuracy using onion samples spiked with the analytes under study. The obtained results are summarized in Table 3. Matrix-matched calibration curves were established using onion samples spiked at five different concentration levels (7.8, 125, 250, 500, and 2000 µg kg −1 ) for all analytes except endrin (31.2, 125, 250, 500, and 2000 µg kg −1 ) and treated following the developed sc-CO 2 extraction procedure. The results show that the proposed method is linear with R 2 ranging between 0.994-0.999. LODs ranging between 0.2 and 2.0 ng g −1 were achieved. Precision studies were carried out in order to evaluate the repeatability (intra-day precision) and reproducibility (inter-day precision) of the proposed sc-CO 2 extraction with the GC-MS detection method. Repeatability and reproducibility were assayed by analyzing spiked samples at 500 µg kg −1 concentration levels. Results expressed as relative standard deviation of response (%RSDs) are shown in Table 3, demonstrating values lower than 11%.
In order to check the trueness of the proposed method, recovery experiments were performed using spiked onion sample at 500 µg kg −1 concentration level. Recoveries were estimated by the comparison of the obtained signal for each analyte with the signal obtained for a blank sample spiked after the sample treatment and prior to its analysis. As can be seen, recoveries range from 80 to 103%, demonstrating the convenience of the proposed sample pre-treatment for quantitative and qualitative analysis of pesticide residues studied in onion samples [32]. Moreover, a blank onion sample was analyzed and there was no co-elution observed for the pesticides under investigation (Figure 4). These results could further be used as a basis to draw the conclusion that the matrices of the onion samples do not have significant effects on the proposed method for extraction of the analytes. The experimental findings reveal that the onion sample tested is either free of the pesticides investigated, or the levels are below the detection limit of the proposed analytical method. Table 3. Analytical performances of the proposed SFE method with GC-MS analysis for onion samples spiked with atrazine, 2,4'-DDD, endrin, and 4,4'-DDT. In order to check the trueness of the proposed method, recovery experiments were performed using spiked onion sample at 500 µg kg -1 concentration level. Recoveries were estimated by the comparison of the obtained signal for each analyte with the signal obtained for a blank sample spiked after the sample treatment and prior to its analysis. As can be seen, recoveries range from 80 to 103%, demonstrating the convenience of the proposed sample pre-treatment for quantitative and qualitative analysis of pesticide residues studied in onion samples [32]. Moreover, a blank onion sample was analyzed and there was no co-elution observed for the pesticides under investigation (Figure 4). These results could further be used as a basis to draw the conclusion that the matrices of the onion samples do not have significant effects on the proposed method for extraction of the analytes. The experimental findings reveal that the onion sample tested is either free of the pesticides investigated, or the levels are below the detection limit of the proposed analytical method.

Comparison of the New SFE Method with Other Extraction Methods in the Literature
The proposed extraction method, SFE using neat ultrahigh pressure sc-CO2 as a solvent followed by GC-MS analysis, was compared with solid phase extraction followed by gas chromatography with electron capture detector (SPE with GC-ECD) [33], a quick, easy, cheap, effective, rugged, and safe extraction method followed by GC with pulsed flame photometric detector (QuEChERS with GC-FPD) [34], QuEChERS with multi-walled carbon nanotubes (MWCNTs) with GC-MS analysis [35], and QuEChERS with GC-MS analysis [36]. The summary of the results obtained is given in Table 4.

Comparison of the New SFE Method with Other Extraction Methods in the Literature
The proposed extraction method, SFE using neat ultrahigh pressure sc-CO 2 as a solvent followed by GC-MS analysis, was compared with solid phase extraction followed by gas chromatography with electron capture detector (SPE with GC-ECD) [33], a quick, easy, cheap, effective, rugged, and safe extraction method followed by GC with pulsed flame photometric detector (QuEChERS with GC-FPD) [34], QuEChERS with multi-walled carbon nanotubes (MWCNTs) with GC-MS analysis [35], and QuEChERS with GC-MS analysis [36]. The summary of the results obtained is given in Table 4. Compared to these methods, the proposed method provided among the highest recoveries and lowest LOD, as well as a better or comparable coefficient of determination. In addition, the new method is likely to be greener compared to the other analytical methods because it uses only sc-CO 2 as an extraction solvent. Hence, the proposed method is a promising extraction method for pesticides having relatively similar or slightly different physical and chemical properties.

Chemicals and Reagents
Ultrapure CO 2 in cylinders with a dip tube was provided by AGA industrial gases (Lidingö, Sweden). Methanol and n-heptane of chromatographic grade were purchased from VWR BDH Chemicals (Gdańsk, Poland) and Fisher Chemicals (Loughborough, UK), respectively. Analytical grade standards including atrazine, 2,4´-DDD, 4,4´-DDT and endrin were purchased from Sigma-Aldrich (Buchis, Switzerland). The stock solution was prepared by weighing 2.5 mg of each standard and dissolving in a small amount of ethyl acetate, and the resulting solution was diluted with methanol in a 25 mL volumetric flask and stored at −20 • C. Mixed working standard solutions, 10 mg L −1 , were prepared by proper dilution of the stock solution with methanol once per week. This working solution was used during method development and validation. A 1.0 g onion sample was spiked at 500 µg kg −1 for each pesticide and air dried at room temperature for 24 h before extraction in order to achieve as much adsorption of the pesticides to the sample matrix as possible.

Sample Site and Sampling
The onion sample was collected randomly from a local market in Addis Ababa, Ethiopia. The geographical location of the sampling site is 8 • 58'50.17" N and longitude 38 • 48'27.94" E with elevation of 2296.1 m above sea level. The onion was cut into small pieces using an iron knife and was thereafter air dried. It was further dried in oven at 92 • C (boiling point of water in Addis Ababa) until constant mass was obtained. Then, the sample was ground with an electric mill, sieved through a 0.25 mm pore size, wrapped in a methanol-rinsed aluminum foil, and kept in a polyethylene plastic bag.

Multivariate Experimental Design
A Box-Behnken response surface experimental design was used to optimize independent quantitative variables including density (0.70-1.0 g mL −1 ), volume (10-40 mL) and temperature (40-70 • C). The design was created using MODDE 10.1 software package (Sartorius Stedim Biotech, Malmö, Sweden) and the responses for each experiment were calculated based on the recovery. The density of sc-CO 2 was calculated using NIST Chemistry WebBook, SRD 69 (Gaithersburg, MD, USA) and given in Table S1. The adequacy of the models was evaluated by the R 2 and Q 2 values (where R 2 shows the model fit and Q 2 shows an estimate of the future prediction precision), predicted vs. observed plot and coefficient plots. Q 2 should be greater than 0.1 for a significant model and greater than 0.5 for a good model. The optimum processing conditions for the maximum peak area for each target compound were obtained by using graphical and numerical analysis based on the criteria of the desirability function and the response surface plots.

Supercritical Fluid Extraction Procedure
Supercritical fluid extraction (SFE) of pesticides was carried out by using a home-built extraction system consisting of an ISCO 65D syringe-pump (Teledyne Isco, Thousand Oaks, CA, USA), a GC oven in which the extraction vessel was placed, and a needle valve connected to a restrictor (a long stainless-steel tube with internal diameter 0.0036 inch) to control the flow rate. A 1.0 g onion sample was spiked by adding an appropriate volume of a standard working solution and mixed with 1.0 g of glass beads to reduce the volume into a 5 mL stainless-steel extraction vessel in sandwich mode, using 3 mm glass beads at both the bottom and the top of the cell to protect the cell sealing. Then, it was heated in the GC oven for 5 min at 53 • C and extraction was continued in a dynamic mode using 29 mL of CO 2 at 3 mL min −1 . The density of sc-CO 2 was kept at 0.90 g mL −1 . Then the extraction was flushed with 2 mL of n-heptane and dried under a gentle N 2 stream. Finally, the extract was reconstituted with 100 µL of n-heptane, and 2 µL of this was injected to GC-MS for analysis. A calibration curve was achieved by extracting onion samples spiked at five different concentration levels (7.8, 125, 250, 500, and 2000 µg kg −1 ) for all analytes except endrin (31.2, 125, 250, 500, and 2000 µg kg −1 ). Repeatability and reproducibility were studied by extracting onion spiked with a mixture of all analytes at 500 µg kg −1 following the optimized extraction parameters.

GC-MS Analysis
A HEWLETT G2614A gas chromatography (GC) equipped with G1099A mass spectrometry (MS) (Hewlett Packard, Palo Alto, CA, USA) was used for analysis and quantification. A HP-5 ms (USA) ultra-inert capillary column (30 m × 250 µm and 0.25 µm i.d.) was used for separations. Helium gas of purity 99.999% was used as carrier gas at a flow of 1.0 mL min −1 . The oven temperature program used during analysis was set as follows: 82 • C for 1 min; increased to 185 • C at 25 • C min −1 ramp rate and held at this temperature for 1 min; then increased to 250 • C at 9 • C min −1 ramp rate and maintained at this temperature for 1 min. The GC injection port temperature was kept at 250 • C. Splitless injection mode was used during the whole analysis. Peaks were identified by their retention time and mass spectra after acquisition of the total ion chromatogram. In order to confirm the retention times of all analytes, scan mode was carried over the range 50-550 m/z. Selective ion monitoring (SIM) of each pesticide and two characteristic fragment qualifier ions were selected to identity of pesticides. The m/z selected for SIM mode detection was: atrazine (173, 200, 215), 2,4´-DDD (199, 235, 237), 4,4´-DDT (212, 235, 239), and endrin (245, 263, 281). Peak area was utilized as an instrumental response, which was converted later to recovery. Recovery for each pesticide was calculated using the concentration obtained after extracting spiked onion following the optimum extraction concentration.

Method Validation
In order to validate the proposed analytical method, linearity, precisions, and limits of detection (LOD) were evaluated. The linearity of the method was done at five different concentration levels, 7.8, 125, 250, 500, and 2000 µg kg −1 for all analytes except endrin, which was calibrated at 31.2, 125, 250, 500, and 2000 µg kg −1 . Each SFE experiment was repeated in triplicate following the optimized procedure and injected in into GC-MS in triplicate. Repeatability (intra-day precision) and reproducibility (inter-day precision) of the method was evaluated using onion samples spiked with each pesticide at 500 µg kg −1 . Recovery analysis was done for each analyte following the same procedure and the same concentration. The limit of detection (LODs) was determined as the lowest concentration yielding a signal to noise (S/N) ratio of 3 [37].

Conclusions
An UHPSFE method with GC-MS analysis was developed for the determination of four pesticide residues in onion samples. The extraction variables density, temperature and volume of sc-CO 2 were optimized by a Box-Behnken design, achieving high analyte recoveries, low LODs and good precision.
The results also indicate that the variables have a marked influence on the extraction recoveries, but are different for the different analytes. For instance, a higher density of the extraction solvent (sc-CO 2 ) was beneficial for the relatively polar pesticides. The optimized method was successfully applied to the analysis of atrazine, 4,4'-DDE, endrin and 2,4'-DDT residues in spiked onion samples. In comparison to extraction methods in the literature, the new proposed method provides a solvent-free green alternative.