Experimental design in supercritical fluid extraction of cocaine from coca leaves

J. Biochem. Biophys. Methods 43 (2000) 353–366 www.elsevier.com / locate / jbbm

Experimental design in supercritical fluid extraction of cocaine from coca leaves a, a b Anne Brachet *, Philippe Christen , Jean-Yves Gauvrit , ´ Longeray b , Pierre Lanteri ´ b , Jean-Luc Veuthey a Remi a

University of Geneva, Laboratory of Pharmaceutical Analytical Chemistry, 20 bd. d’ Yvoy, CH-1211, Geneva 4, Switzerland b University of Lyon, ESCPE Lyon, Laboratory of Chemometrics, 43 bd. du 11 Novembre 1918, 69622, Villeurbanne, France Received 18 November 1999

Abstract An optimisation procedure for the supercritical fluid extraction (SFE) of cocaine from the leaves of Erythroxylum coca var. coca was investigated by means of experimental design. After preliminary experiments where the SFE rate-controlling mechanism was determined, a central composite design was applied to evaluate interactions between selected SFE factors such as pressure, temperature, nature and percentage of the polar modifier, as well as to optimise these factors. Predicted and experimental contents of cocaine were compared and robustness of the extraction method estimated by drawing response surfaces. The analysis of cocaine in crude extracts was carried out by capillary GC equipped with a flame ionisation detector (GC–FID), as well as by capillary GC coupled with a mass spectrometer (GC–MS) for peak identification.  2000 Elsevier Science B.V. All rights reserved. Keywords: Supercritical fluid extraction; Experimental design; Cocaine; Coca leaves; Plant extract

1. Introduction Supercritical fluid extraction (SFE) has become a method of choice for the extraction of solid matrices such as plant material [1]. However, to the best of our knowledge, a *Corresponding author. 0165-022X / 00 / $ – see front matter  2000 Elsevier Science B.V. All rights reserved. PII: S0165-022X( 00 )00062-2

354

A. Brachet et al. / J. Biochem. Biophys. Methods 43 (2000) 353 – 366

SFE procedure has never been reported for the extraction of cocaine from coca leaves which is usually performed by classical methods [2,3]. SFE is an alternative technique to liquid–solid extraction method because it offers several advantages such as shorter extraction time, smaller solvent consumption and reduced working temperature. SFE of polar compounds has not been extensively studied because of the low polarity of supercritical carbon dioxide. However, some natural products such as morphine alkaloids [4] and pyrrolizidine alkaloids [5] have already been extracted by SFE. In these cases, addition of modifiers was necessary to reduce matrix interactions and increase supercritical fluid polarity [6]. Analysis of cocaine by both capillary [7] and packed column [7,8] supercritical fluid chromatography (SFC), using neat CO 2 , showed a reasonable solubility which suggests that it can be extracted by SFE. In addition, cocaine has already been extracted by SFE in hair samples [7,9,10]. Furthermore, a recent study has been published on the extraction of hyoscyamine, scopolamine and cocaine from plant material [11]. Four recognised coca varieties deriving from two species of Erythroxylum contain significant levels of cocaine. Among them, Erythroxylum coca var. coca, is of particular interest as the leaf cocaine content in it is usually comprised between 0.5 and 0.8% dry weight [3,12,13]. In this paper, the feasibility of using a supercritical fluid to extract cocaine from coca leaves is investigated. The implications of the SFE rate-controlling mechanism are discussed regarding sample size and SFE flow parameters. Optimisation of the extraction conditions is usually assessed by systematic alteration of one variable while the others are maintained constant. However, this approach is unable to determine interactions between parameters and predict extraction conditions. In this respect, experimental designs are appropriate tools for this purpose. Furthermore, these designs allow efficient testing of method robustness [14]. Among these experimental designs, full factorial designs reveal the significance of the factors under investigation as well as the interactions between them. Second-degree designs such as central composite [15,16] or Doehlert designs [17] allow method modelling and determination of optimal conditions. A central composite design is used to optimise SFE experimental conditions for cocaine in coca leaves (E. coca var. coca), since it offers the possibility to evaluate the robustness of the method by drawing response surfaces.

2. Materials and methods

2.1. Materials Leaves of E. coca var. coca were harvested in Tingo Maria in Peru, authenticated by T. Plowman and kindly provided by Dr. L. Rivier from the Forensic Institute of Lausanne (Switzerland). CO 2 (99.99% purity) (Polygaz, Geneva, Switzerland) was used for supercritical fluid ¨ extraction. HPLC grade methanol was purchased from Romil (Kolliken, Switzerland). Ultrapure water was provided by a Milli-Q RG unit from Millipore (Bedford, USA).

A. Brachet et al. / J. Biochem. Biophys. Methods 43 (2000) 353 – 366

355

Cocaine hydrochloride and methadone hydrochloride were obtained from Siegfried ¨ Handel (Zofingen, Switzerland) and Hanseler (Herisau, Switzerland), respectively.

2.2. Supercritical fluid extraction The air-dried plant material was thoroughly ground in a domestic mixer in order to obtain a specific particle size distribution. 100 mg of this material was placed in an extraction cell (1 ml internal volume, 5.6 cm long 3 0.5 cm I.D.) supplied by Supelco (Bellefonte, USA). CO 2 and polar modifier were delivered by a Varian 2510 HPLC pump (Varian, Palo Alto, USA) and a SSI 220 B HPLC pump (State College, USA), respectively. A six-way switching valve (Rheodyne, USA) was used to by-pass the extraction cell which was placed in a home-made oven (20–1108C). Pressure in the system was monitored by a back-pressure regulator JASCO, model 880-81 (JASCO, Tokyo, Japan) heated at 708C to avoid dry ice formation. Each extraction was performed by filling the cell with the supercritical fluid before heating for 10 min at the temperature set by the experimental design. In a preliminary study, extraction kinetics were carried out to assess extraction time, flow rate and granulometry for complete cocaine extraction. SFE was conducted under the following experimental conditions: flow rate was maintained at 2 ml / min, extraction time set at 15 min and particle size distribution lied between 150 and 170 mm. After restriction, samples were collected by bubbling in 5 ml methanol, maintained at 58C in an ice bath to increase cocaine collection efficiency. After extraction, the system was flushed with methanol for 5 min. Extracts were evaporated to dryness under a gentle stream of nitrogen. The dry residue was redissolved in 2 ml of a methanolic solution of methadone (at a concentration of 45 mg / ml). The mixture was centrifuged at 1700 g for 5 min and analysed without any further purification by GC–FID and by GC–MS for peak confirmation.

2.3. GC–FID analysis GC–FID was performed with a Perkin Elmer Autosystem (Perkin Elmer, Rotkreuz, Switzerland). Helium was used as carrier gas at a pressure of 0.172 MPa. Injection was performed in the splitless mode and injection volume was 1 ml. Injection and detection temperatures were maintained at 250 and 3008C, respectively. The gas chromatograph was fitted with a 30 m 3 0.25 mm I.D. fused-silica capillary column coated with a phenyl-methyl silicone phase HP5-MS (Hewlett-Packard, Palo Alto, USA) with a film thickness of 0.25 mm. The oven temperature was as follows: initial temperature, 708C, initial hold, 2 min; temperature programme rate, 308C / min; final temperature, 2858C, final hold, 5 min. For FID, air and hydrogen flow rates were 400 and 40 ml / min, respectively. Integration was carried out with a HP 3395 integrator. Quantification of cocaine was performed by GC–FID using an internal standard calibration with a four-point linear calibration plot (concentrations between 45 and 450 mg / ml). The calibration curve was constructed by plotting the area ratio of cocaine and

356

A. Brachet et al. / J. Biochem. Biophys. Methods 43 (2000) 353 – 366

internal standard (methadone HCl) peaks vs. concentration ratio. Response factor and determination coefficient were assessed from the least square regression.

2.4. GC–MS analysis GC–MS was performed with a HP 5890 series II gas chromatograph coupled with a HP 5972 mass spectrometer. The mass selective detector was operated in the electron impact ionisation mode with a ionisation potential of 70 eV. Helium was used as carrier gas at a flow rate of 1 ml / min. Injection was performed in the splitless mode and the injection volume was 1 ml. Injection temperature was maintained at 2508C and detection temperature at 2808C. The gas chromatograph was fitted with a 30 m 3 0.25 mm I.D. fused-silica capillary column coated with a phenyl-methyl silicone phase HP5-MS with a film thickness of 0.25 mm. Operating conditions were similar to the previous GC–FID temperature programme.

2.5. Software Coefficients for regression models and optimised conditions were calculated using (LPRAI, Marseille, France) and LUMIERE (SIER, Enghien les Bains, France) software packages, respectively. Response surfaces were drawn with Microsoft Excel (version 7.0). NEMROD

3. Results and discussion

3.1. Selection of relevant factors Preliminary experiments were carried out in order to select relevant factors as well as their experimental domain to obtain the highest cocaine extraction recovery. As neat carbon dioxide was not sufficient to extract cocaine, a polar modifier was added. Methanol and water with or without triethylamine were investigated. Addition of water in methanol had a significant influence on cocaine recovery while addition of triethylamine in methanol did not improve extraction yields. Therefore, a mixture of methanol and water was selected for subsequent investigations. The effects of supercritical flow rate on extraction kinetics of cocaine from coca leaves were also investigated to determine the rate-limiting step of the supercritical fluid extraction [18]. Extraction kinetics were conducted on coca leaves of various granulometry at different flow rates with CO 2 –MeOH–H 2 O (90:9:1, v / v / v) at 20 MPa and 408C (cf. Fig. 1). Extracts were collected at timed intervals over a period of 60 min. Cocaine recovery was given in percentage of the final cocaine amount extracted after 60 min. The effect of supercritical flow rate on cocaine extraction rate is shown in Fig. 1a. Results demonstrate that cocaine extraction is mainly governed by the solubility step rather than by the desorption / kinetic step. At 1 ml / min, cocaine extraction rate was significantly slower. Increasing the flow rate from 1 to 2 ml / min yielded faster cocaine

A. Brachet et al. / J. Biochem. Biophys. Methods 43 (2000) 353 – 366

357

Fig. 1. Extraction kinetic curves (a) at different flow rates for a particle size distribution between 150 and 170 mm and (b) as a function of particle size at a constant flow rate (2 ml / min).

358

A. Brachet et al. / J. Biochem. Biophys. Methods 43 (2000) 353 – 366

recovery. Above 2 ml / min, flow rates did not have an appreciable effect on cocaine extraction rates. The lower apparent cocaine recovery observed at 3 ml / min was verified to be insignificantly different to that obtained at 2 ml / min. According to these results, the flow rate was set at 2 ml / min for all the following experiments. The effect of particle size on extraction was investigated at 2 ml / min. As shown in Fig. 1b, extraction kinetics were not affected by particle size. It should be noted that cocaine diffusion in the sample matrix was not the major limiting step since grinding the samples did not increase the extraction rate. However, when the particle size distribution was between 220 and 470 mm, supercritical fluid could not yield quantitative recovery of cocaine even after 60 min of extraction. Thus, the granulometry was set at 150–170 mm and the extraction time at 15 min. Four factors were selected for the experimental design: pressure, temperature, percentage of polar modifier (mixture of MeOH / H 2 O) and percentage of water in methanol. A full two-level factorial design with four factors was then implemented; the range of these factors is reported in Table 1. In the experimental domain, extraction was carried out in the supercritical or subcritical state. However, the nature of the fluid was not taken into account for the optimisation because it did not affect the mathematical model.

3.2. Full Factorial design In order to determine the influencing factors as well as their interactions, a 2 4 full factorial design was carried out. Therefore, sixteen extractions were required to cover all possible combinations of factor levels (experiments 1–16 in Table 2). Experiments at the centre (n 5 6) were conducted for estimation of the experimental error (experiments 17– 22 in Table 2). All experiments were randomly performed without replication. The measured response was defined as cocaine content in % dry weight in coca leaves. Multiple regression enables a mathematical relationship between the responses obtained experimentally and the different independent system variables. The synergist model is shown below: * X 1* X4 cocaine content (% dry weight) 5 b 0 1 b *1 X1 1 b *2 X2 1 b 3* X3 1 b 4* X4 1 b *12 X 1* X2 1 b *13 X *1 X3 1 b 14 * X *2 X3 1 b 24 * X *2 X4 1 b *34 X *3 X4 1 b 123 * X *1 X 2* X3 1 b *124 X *1 X *2 X4 1 b *134 X *1 X *3 X4 1 b *234 X *2 X *3 X4 1 b *1234 X *1 X *2 X *3 X4 . 1 b 23

Coefficients b i of this model, were calculated and graphically represented with their ] confidence interval in Fig. 2. The confidence interval was calculated as (t*S.D. exp ) /ŒN Table 1 Experimental domain Factors Pressure (MPa) % Polar modifier in CO 2 % Water in methanol Temperature (8C)

X1 X2 X3 X4

Low level (2)

Centre level (0)

High level (1)

15 5 10 40

20 10 20 70

25 15 30 100

A. Brachet et al. / J. Biochem. Biophys. Methods 43 (2000) 353 – 366

359

Table 2 Four-factor central composite design with corresponding response Trial

Experimental factors

Cocaine (% dry weight)

X1

X2

X3

X4

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

2 1 2 1 2 1 2 1 2 1 2 1 2 1 2 1

2 2 1 1 2 2 1 1 2 2 1 1 2 2 1 1

2 2 2 2 1 1 1 1 2 2 2 2 1 1 1 1

2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1

0.17 0.38 0.43 0.21 0.53 0.53 0.54 0.52 0.24 0.53 0.54 0.45 0.45 0.43 0.55 0.55

17 18 19 20 21 22

0 0 0 0 0 0

0 0 0 0 0 0

0 0 0 0 0 0

0 0 0 0 0 0

0.52 0.53 0.51 0.57 0.58 0.59

23 24 25 26 27 28 29 30

2 1 0 0 0 0 0 0

0 0 2 1 0 0 0 0

0 0 0 0 2 1 0 0

0 0 0 0 0 0 2 1

0.59 0.58 0.55 0.58 0.57 0.60 0.60 0.60

31 32 33

0 0 0

0 0 0

0 0 0

0 0 0

0.50 0.58 0.59

where N is the number of experiments of the full factorial design, S.D. exp the experimental standard deviation calculated with the six centre points and t the Student variable at 95% confidence level. The relative standard deviation (R.S.D.55.9%) for replicate extractions was satisfactory considering that it represents the total analysis variation (supercritical fluid extraction and GC-analysis). From these results, six coefficients were statistically significant. Fig. 2 indicates that the water content in methanol (X 3 ) had the greatest effect (b 3 50.0709) on cocaine recovery. It is noteworthy that pressure (X 1 ) had no effect on cocaine recovery, while this factor associated to the

360

A. Brachet et al. / J. Biochem. Biophys. Methods 43 (2000) 353 – 366

Fig. 2. Synergist model coefficients.

percentage of modifier in CO 2 (X 2 ) showed a strong influence (b 12 5 20.0506). Thus, attention must be paid not only on main effects but also on interaction factors. Graphical representations of the two most important first-order interactions (b 12 , b 34 ) are reported in Fig. 3. In Fig. 3a, the percentage of modifier in CO 2 (X 2 ) at high pressure level had no influence on cocaine recovery (0.47 and 0.43%), while it was necessary to work with a high modifier percentage at low pressure level (0.35 to 0.52%). However, it is surprising that under these latter conditions, cocaine recovery was higher than with a high pressure. Increasing the water content in methanol (X 3 ) considerably enhanced cocaine extraction at low temperature (cf. Fig. 3b). At a low water concentration, temperature had much more influence on cocaine recovery (0.30–0.44%), whereas with a high percentage of water in methanol, working at a high or a low temperature nearly comes to the same (0.53 and 0.49%). Furthermore, a second-order interaction b 123 between pressure (X 1 ), modifier percentage in CO 2 (X 2 ) and water percentage in methanol (X 3 ) revealed to be important (cf. Fig. 2). Thus, this coefficient b 123 cannot be ignored in the response modelisation.

3.3. Optimisation of the extraction Centre point experiments were used to evaluate the extraction procedure repeatability and test the response curvature. Comparison between the mean response of the six centre point experiments (b 0exp 50.55) and the estimated response of the model (b 0theo 50.44) revealed a probable curvature. Therefore, quadratic terms had to be added to improve the response modelling. In order to obtain optimal conditions for cocaine extraction from

A. Brachet et al. / J. Biochem. Biophys. Methods 43 (2000) 353 – 366

361

Fig. 3. Graphical representations of interactive effects between (a) pressure (X 1 ) and % of modifier in CO 2 (X 2 ) and between (b) % of water in methanol (X 3 ) and temperature (X 4 ).

362

A. Brachet et al. / J. Biochem. Biophys. Methods 43 (2000) 353 – 366

coca leaves, a face-centred central composite design was chosen. The latter is composed by the initial full factorial design and the star point design (experiments 23–30 in Table 2). The centre point was repeated thrice to test the block effect (experiments 31–33 in Table 2). The mathematical model obtained by multiple regression of a quadratic-degree expression, from which the cubic term b 123 was added, has the following expression: cocaine content (% dry weight)50.578810.0073*X 1 10.0310*X 2 10.0642*X 3 10.0237*X 4 20.0282*X 12 20.0520*X 22 20.0328*X 32 20.0197*X 24 20.0506*X *1 X 2 20.0150*X *1 X 3 10.0130*X *1 X 4 20.0059*X *2 X 3 10.0211*X *2 X 4 20.0444*X *3 X 4 10.0514*X 1* X 2* X 3

This model allows to establish response surfaces and predict any value within the experimental domain. After having verified that there is no block effect, all data was tested statistically. Coefficient of determination (R 2 ), adjusted coefficient of determination (R 2adj ) and residual standard deviation S.D. res , were satisfactory, as reported in Table 3. Residual error values were contained within the range of 62S.D. exp , as illustrated in Fig. 4. Even if two excessive residual error values were observed, the response was sufficiently explained by the regression model. From the mathematical model, optimal conditions for cocaine extraction were calculated and were: 19.4 MPa (X 1 ), 11.33% of polar modifier in CO 2 (X 2 ), 29% of water in methanol (X 3 ) and 708C (X 4 ). Optimal conditions were applied to the extraction of cocaine from E. coca var. coca. The cocaine content was 0.60%, which is in complete agreement with the predicted value (0.61%) as well as with the value reported in the literature [3]. Extraction kinetic was carried out under these optimal conditions. Fig. 5 shows that 7.5 min is sufficient to extract cocaine from coca leaves completely at a flow rate of 2 ml / min. As shown in Fig. 6, the total-ion current (TIC) GC–MS pattern of a SFE extract (without any other purification step) is sufficiently clean to identify and quantify cocaine. It can be noted that other tropane alkaloids are coextracted with cocaine as well as fatty acids (broad peak between 7.4 and 8.4 min) but these compounds do not interfere. Hygrine, ecgonidine methyl ester, ecgonine methyl ester, cis-cinnamoylcocaine and trans-cinnamoylcocaine are eluted at 4.7, 6.4, 6.8, 10.1 and 10.7 min, respectively. These compounds were identified according to their MS spectra but were not quantified in this study. Thus, the selectivity of the method (SFE and GC analysis) allows to use GC–FID for cocaine quantification. Table 3 Statistical data S.D. exp a S.D. res b R2 c R 2adj d a

0.03 0.05 0.8950 0.8024

Experimental standard deviation in % (n59). Residual standard deviation in % (degree of freedom517). c Coefficient of determination. d Adjusted coefficient of determination. b

A. Brachet et al. / J. Biochem. Biophys. Methods 43 (2000) 353 – 366

Fig. 4. Residual error.

Fig. 5. Extraction kinetic curve at optimal conditions.

363

364

A. Brachet et al. / J. Biochem. Biophys. Methods 43 (2000) 353 – 366

Fig. 6. Total-ion current (TIC) GC–MS chromatogram of major coca alkaloids in E. coca var. coca. Peaks: 15hygrine; 25ecgonidine methyl ester; 35ecgonine methyl ester; 45methadone (internal standard); 55 cocaine; 65cis-cinnamoylcocaine; 75trans-cinnamoylcocaine.

3.4. Evaluation of response surfaces Response surfaces can be visualised as three-dimensional plots by presenting the response in function of two factors and keeping the others constant. We chose to evaluate the influence of pressure and percentage of modifier in CO 2 on cocaine recovery because the full factorial study showed an important interaction between them [11]. As depicted in Fig. 7, the response surface showed a fairly strong degree of curvature, where the optimum can be readily determined. Pressure had a weak influence on cocaine recovery which confirms the little importance of this factor, while the percentage of modifier in CO 2 had a major effect. The optimal zone (17–22 MPa for pressure and 9.5–13% for the percentage of modifier in CO 2 ) was perfectly defined in the centre of the experimental domain and showed that the extraction method had a good robustness.

4. Conclusion Results showed that supercritical fluid extraction is a valuable alternative technique to classical extraction methods of cocaine from coca leaves. Extraction kinetic curves revealed that cocaine solubility was the rate-limiting step. By applying a full factorial

A. Brachet et al. / J. Biochem. Biophys. Methods 43 (2000) 353 – 366

365

Fig. 7. Response surface at 708C with 29% of water in methanol [11].

design, the effects of four SFE parameters on cocaine recovery were measured. It was revealed that the content of water in methanol was the most dominant factor. A remarkable second-order interaction was demonstrated between pressure, percentage of modifier in CO 2 and percentage of water in methanol. Optimisation of cocaine extraction conditions was assessed by using a central composite design. The ensuring mathematical model enabled to predict cocaine content at any point in the experimental domain as well as to determine optimal extraction conditions. The quality of the model was verified by the good agreement between experimental and predicted responses. Under optimal conditions, cocaine content was 0.60% in air-dried leaf which is in good agreement with the value reported in the literature. Response surfaces were drawn from the regression model and allowed to illustrate interactions between selected SFE factors and to evaluate the extraction method robustness.

References ´ ´ [1] Castioni P, Christen P, Veuthey J-L. L’extraction en phase supercritique des substances d’origine vegetale. Analusis 1995;23:95–106.

366

A. Brachet et al. / J. Biochem. Biophys. Methods 43 (2000) 353 – 366

[2] Moore JM, Casale JF, Klein RF, Cooper DA, Lydon J. Determination and in-depth chromatographic analyses of alkaloids in South American and greenhouse-cultivated coca leaves. J Chromatogr A 1994;659:163–75. [3] Turner CE, Ma CY, Elsohly MA. Constituents in Erythroxylum coca I: gas chromatographic analysis of cocaine from three locations in Peru. Bull Narcot 1979;31:71–6. [4] Janicot JL, Caude M, Rosset R, Veuthey J-L. Extraction of major alkaloids from poppy straw with near-critical mixtures of carbon dioxide and polar modifiers. J Chromatogr 1990;505:247–56. [5] Bicchi C, Rubiolo P, Frattini C. Off-line supercritical fluid extraction and capillary gas chromatography of pyrrolizidine alkaloids in Senecio species. J Nat Prod 1991;54:941–5. [6] Westwood SA. Supercritical fluid extraction and its use in chromatographic sample preparation, Glasgow: Blackie Academic and Professional, 1993. [7] Later DW, Richter BE, Knowles DE, Andersen MR. Analysis of various classes of drugs by capillary supercritical fluid chromatography. J Chromatogr Sci 1986;24:249–53. [8] Crowther JB, Henion JD. Supercritical fluid chromatography of polar drugs using small-particle packed columns with mass spectrometric detection. Anal Chem 1985;57:2711–6. [9] Morrison JF, Chesler SN, Yoo WJ, Selavka CM. Matrix and modifier effects in the supercritical fluid extraction of cocaine and benzoylecgonine from human hair. Anal Chem 1998;70:163–72. ´ [10] Veuthey J-L, Edder P, Staub C. Depistage des toxicomanies au moyen de l’extraction en phase ´ a` la GC–MS: applications a` l’urine et aux cheveux. Analusis 1995;23:258–65. supercritique couplee ´ P, Veuthey J-L. Application of central [11] Brachet A, Mateus L, Cherkaoui S, Christen P, Gauvrit JY. Lanteri composite designs in the supercritical fluid extraction of tropane alkaloids in plant extract Analusis 1999;27:772–8. [12] US Drug Enforcement Administration. 1993 (unpublished results). [13] Plowman T, Rivier L. Cocaine and cinnamoylcocaine content of Erythroxylum species. Ann Bot 1983;51:641–59. ´ P, Longeray R. Chimiometrie: ´ ´ [14] Lanteri le mariage reussi entre les sciences analytiques et l’informatique. Analusis 1996;24:M17–27. [15] Box GEP, Hunter WG, Hunter JS. Statistics for experimenters: an introduction to design, data analysis and model building, New York: Wiley, 1978. [16] Box GEP, Draper NR. Empirical model-building and response surfaces, New York: Wiley, 1987. [17] Doelhert DH. Uniform Shell Designs. Appl Stat 1970:231–239. [18] Hawthorne SB, Galy AB, Schmitt VO, Miller DJ. Effect of SFE flow rate on extraction rates: classifying sample extraction behavior. Anal Chem 1995;67:2723–32.