Experimental design approach for supercritical fluid extraction

Analytrca Chotuca Acta, 271(1993) 83-90 Elsevler Science Pubhshers B V , Amsterdam

Experimental design approach for supercritical fluid extraction M Kane, J R Dean and S M Hltchen Department of Chemrcal and Life Scrences, Umvers~tyof Northumbna at Newcastle, Elbson Buhimg, Newcastle upon Tyne NE1 8ST (UK)

C J Dowle ICI plc, WZton Research Centre, P 0 Box 90, Wilton, Mddlesbrough, Cleveland TS6 &JE (UK)

R L Tranter Glaxo Manufactunng Servrces, Harmwe Road, Barnard Cast&, County Durham DL12 8DT (UK)

(Received 1st June 1992, revised manuscript recewed 9th July 1992)

Abstract Expenmental design with multlhnear regression has been used to examme the relative contriiutlon of the main experimental vanables durmg supercntlcal fluid extraction SLXsteroidal compounds of vanous solubditles m supercntlcal CO, were considered The results indicate that the density of the supercntlcal flurd has the greatest effect on the solubdlsation and transfer of steroid from extraction cell to collectlon device The nununum number of cell volumes of supercntlcal CO, requued for effective extraction was expenmentally determmed Keywords Carbon dlomde, Extraction, Multdmear regression, Sterouis, SupercTltlcal fluid extraction

The status of current analytlcal mstrumentatlon IS such that the rehablhty and robustness of measurement IS not often an issue However, the analytlcal chemist must conader problems assoclated with sample preparation and the mterpretatlon of large amounts of expernnental data TINS paper examines how the use of supercrltlcal fluid extractlon [1,2] for sample preparation and experimental design m data generation can slmphfy both of these areas Supercrrtlcal fluid extraction (SFE) 1s a relaCorrespondence

to J R Dean, Department of Chemical and Life Sciences, Umverslty of Northumbrla at Newcastle, Elhson Buddmg, Newcastle upon Tyne NE1 8ST (UK)

twely new technique [3] for sample preparation based on the use of carbon dloxlde above Its crItIca temperature, 31 l”C, and crltrcal pressure, 73 8 bar The high drffuslvlty of the supercntlcal carbon dloxlde allows for rapld and effective extractions of mamly non-polar solutes [4] urlthout the need for the costly organic solvents frequently used m traditional sample clean-up procedures The main operating vanables m analytical SFE are density, temperature, flow-rate and time of extractlon and It IS expected that there are mteractions between these vanables Expernnental design techniques [5,6] may be used both to model and optmuse chemical mformatlon, where there are many potentially mteractmg vanables, and

OOO3-2670/93/$0600 0 1993 - Elsevler Science Publishers B V All nghts resewed

M Kane et al /Ad

are provmg to be valuable assets to the analytical chemist [7] Expenmental design allows a conslderatlon of the overall number of experunents, the arrangement of the experunents and the possible mteractlon effects between the variables A slmultaneous expernnental design based on multilinear regresslon has been used [8] for the optlmlzatlon of extractlon condltlons Thus has allowed both the prediction of optimal expenmental condltlons and Interaction effects between variables The combmatlon of these two powerful techniques enables the main operatmg vanables of SFE fully to be investigated

EXPERIMENTAL

Apparatus

Supercrltlcal extractions were undertaken on a computer driven system, the Hewlett-Packard (Avondale) Model 7680A SFE The instrument operates as an off-lme extraction umt wth a solvent washed chromatographlc trap as the collectlon device for all samples [9] Figure 1 lllustrates the mam components of the system All critical parameters are monitored and controlled by the personal computer (PC) software It should be noted that the vanable restrictor devxe allows for the mdependent control of both the flow-rate and the back pressure durmg the extraction Llqmd carbon dloxlde 1s pumped with constant flow by a twm piston reciprocating pump The low viscosity of carbon dloxlde requues that the pump head 1s cooled to avold leaks at the pump head

Chm Acta 271 (1993) 83-90

A second carbon dloxlde cylmder provides cryogemc cooling to the system Sample cells are made of stainless steel with re-usable hand tlghtened PEEK caps at each end The cells are available m 15- and 7 O-ml volumes Both cell volumes have been studled m thrs mvestlgatlon, although the malorlty of the results were achieved using the smaller cells Temperature condltlons are mamtamed accurately at each stage of the extractlon by the use of focused cryogemcs m conJunctIon with a series of electrothermal heatmg blocks A two mmute penod of static extraction, where there IS no fluid flow, was used at the beginning of each extraction sequence to allow the analyte to be fully solublhsed The mass transfer occurs when the instrument IS operating m the dynamic flow mode and this 1s taken as the time of extraction After the depressurlsatlon stage, the final rmse step washes an appropriate solvent (methanol m the experunent) throughout the entire solvent hne, restrictor and trapping column mto 2-ml sample vials contamed m a fraction collector The trap, 7 cm m length and 5 mm 1 d, 1s loosely packed with Hypersll ODS (particle size 30 pm) and has a vahd volume of 650 ~1 Volatile analyte loss IS avoided by cryogenic coohng of the trap Analysis of the extracts was performed on a Shlmadzu (Tokyo) Model UV-160 double beam UV-vlslble spectrophotometer usmg lo-mm sdrca cells (Thermal Syndicate) for all measurements Quantltatlon was achieved by dllutmg all extracted samples to 10 ml usmg Techmco class B graduated flasks and measurmg absorbance agamst a four pomt linear cahbratlon graph None of the compounds studied demonstrated any mstability m methanol when monitored over a week Reagents

FIB 1 Block diagram of the HP 7fi8OAextractor

Supercrltlcal fluid grade carbon &oxlde (A1r Products, Sunderland) of cetied purity 99 995% was used for extraction, while mdustnal grade carbon dloxlde was used for cryogenic coolmg All SIXsteroidal compounds (Fig 2) studied were analytical workmg standards of certlfled purity supplied by Glaxo, Barnard Castle Methanol (BDH, Poole) of AnalaR grade was used throughout the expenmentatlon

M Kane et aL/Anal

Chm. Acta 271 (1993) 83-W

Sample uatroductwn All steroids were studled as pure compounds thus mmnmsmg potential matrix effects To achieve this each sample was mtroduced mto the extractlon cell as a 50-pg pure standard deposlted on the mtemal surface of a glass capillary tube This was done by measurmg a precise 50-~1 volume of a 1000 pg ml-’ stock solution prepared III methanol onto the glass and allowing the solvent to evaporate By reducmg the complex@ of the extractlon solublhty effects, the mass transfer due to flow-rate and cell geometry can be studied The mteractlons of the sample with the matrix and their resultmg effect on extraction efficiency cannot be studied by thrs method of extraction Spllung onto glass allows the visual confumatlon of successful extraction

85

critical condltlons the flow-rate wdl depend on the density achieved At Hugh densities (as the supercrltlcal flwd tends towards more hqmd-hke condltlons) the flow-rate through the extraction vessel wdl remam smular to the hquld pump-rate Lower densities (as the supercritical fled tends towards more gas-hke condltlons) result m greater flow-rate values All quoted values of flow-rate are the hqurd carbon dloxlde pump-rate

Expemntal &szgn If the correct conclusions are to be drawn from an experunent, the various factors which affect the result must be identified and, if possible, controlled [lo] In this case the solubdlty of the analytes m carbon dloxlde has been assumed to depend on a number of expemnental factors e g density (g ml-‘), temperature PC), tune of exFlow-rate traction (mm) and flow-rate (ml mm-‘) Any The carbon dloxlde 1s pumped from the cylmdesign of the experrmentatlon should involve a der at approxnnately 53 bar at 15°C Carbon consideration of these vanables 111order to obtain dloxlde 1s hqmd under these conditions and the the best solubdlty parameters and hence the best flow-rate 1s measured at this stage Under superexpernnental extraction condltlons Fractional factorial expenmental design [5,6] allows for the detection and estlmatlon of any interactions between the expenmental factors Classical umvanate optnmsation fads to give any mformation about factor mteractlons, and as a result can produce unsatisfactory optmuun comhtlons Factonal design also needs fewer measurements than the classical approach to achieve the same precl-slon, thus reducing the method development time C&OH As four vanables (three considered as quadrat~ c=o ics) were considered a large number of possible HO OH combmatlons of expernnents could be selected ua As the length of each mdlvldual experunent can / have serious effects on the method development 0 tune it was important to reduce the number of experunents to a mmunum without serious confoundmg [ll], random or systematic errors being yw ~H=cH=ococH3 introduced This was established as 18 expenu* c=o u# c=o 0 OH ments (Table 1) The quadratic terms were stud80 CH,OCOCli, ue Yda led at low, intermediate and high values temperue / ue / ature was studied at 40, 60 and 8O”C, carbon P P / / 0 dloxlde density at 0 40, 0 65 and 0 80 g ml- ’ and 0 flow-rate at 2 0, 3 0 and 4 0 ml mm- ’ However, -17 21 rhpvFhm the tune of extraction was considered at only two Fig 2 Structural formulae of the steroldal compounds levels, 2 and 10 mm It should be noted that the

dP

Li!F cc!P

86

M fine

et al /Arm! Cht.m Acta 271(1993) 83-W

TABLE 1 The experunental design wtth the UV-vwble Temperature

Denslty (g mm - ‘1

Tune

FIOWrate (ml mm-‘)

Steroid (% recovery) a

Cm3

No 1

No 2

No 3

No 4

No 5

No 6

040 065 040 090 065 090 040 040 0 65 040 090 040 0 65 090 090 090 040 090

20 20 10 0 20 10 0 10 0 20 100 20 20 20 100 100 10 0 100 20 20 20

20 20 20 20 20 20 30 30 40 40 40 40 40 40 20 30 40 20

93 98 91 99 103 18 22 44 95 6 99 0 102 95 103 99 0 98

0 22 0 88 100 97 59 0 73 16 33 0 107 108 98 97 0 23

0 16 20 32 20 26 0 0 12 1 22 1 12 19 54 41 1 26

13 26 12 109 5 108 4 20 43 10 64 5 18 123 99 108 5 51

4 2 11 24 4 17 0 2 4 0 4 0 0 56 49 34 0 7

43 86 12 101 107 109 35 79 87 2 106 45 106 101 0 86 42 101

(“0 80 40 40 80 60 40 60 80 80 40 40 60 40 80 80 80 80 40

percentage recoveries observed for the SIXsteroids usmg 15 ml cell volume

B No 1 = megestrol acetate, No 2 = cortisone acetate, No 6 = betamethasone-17,21-dlproplonate

3 = clobetasol,

No

4 = clobetasone,

No

5 = hydmcortwne,

RESULTS AND DISCUSSION

chosen values should be wlthm the workmg hmlts of the mstrumentatlon and that extrapolation of the response curves beyond the constraints of the mltlal design 1s to be avoided The experiments were carried out m a randomlsed order for each steroid to mnnmlse the effect of bias due to, for example, sample carry over

Six steroids were selected for analysis (Fig 2) The sectlon of the steroids was based on a common structural backbone with a range of functlonal groups It was determmed experlmentally that the SE steroids had some degree of solublhty

TABLE 2 Coeffuxents of regresslon and standard error of linear Eqn 1 for cortisone acetate Variable

Descnptor

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

“1 “2 “3 “4 “l”l “l”2 “l”3 “l”4 “2”t “2”3 v2v4 “3”4 “4”4

No

Beta

Standard error

b

Standard error

t(4)

P level

2 701 3 171 0 088 4 771 -2501 1001 -1029 -0603 -3.512 1.237 -0057 0006 -4219

1419 1236 0 473 1473 1358 0 392 0311 0 391 0964 0 294 0 354 0 297 1378

6 377 618.821 0 953 225390 -0049 2260 -0 167 -0324 -525 04 17 874 -3037 0 021 -33 18

3 350 241 176 5 116 69 552 0 027 0 885 0051 0209 144 152 4 241 18 717 0 987 10 828

1903 2.8g9 0 186 3.239 -1842 2 555 -3309 -1547 -3642 4 214 -0 161 0 021 -3062

0 129 0044 0861 0039 0 139 0 063 0029 0 197 0022 0 013 0 879 0984 0037

87

M Kane et aL/AnaL Chun Acta 271 (1993) 83-90

ture, density, tnne, and flow-rate, respectively, b1 b13 are the parametric coefficients, b, 1s the mtercept Multlllnear least squares regression was used to calculate the coefficients, b, m Eqn 1 Because of the qurte large dd’ferences between the magmtude of the squares of temperature and density (VT and ~122m Eqn 11, the coefficients were calculated also wrth the data for each variable standardrzed to its mean of zero and a standard deviation of unity The coefflclents of cortlsone acetate are reported m Table 2 as b, and Beta, for the un-standardlsed and standardlsed vanables, respectively As this data 1s well behaved there 1s no difference m the conclusions drawn by the two approaches A plctonal representation of the coefficients for the standardlsed data 1s given m Fig 3 Slgmficance 1s determmed by a t-test with a confidence level of 0 05 (95% confidence hmlt) at 4 degrees of freedom In Table 2 a coefficient mth a probablhty (P) level of less than 0 05 or a t-test value of greater than 2 78 wJ1 be connd-

m supercnttlcal carbon dloxlde, rangmg from the totally soluble megestrol acetate and betamethasane-17,21-&proplonate to the very slightly soluble clobetasol and hydrocotisone Cortisone acetate 1s used m the followmg discussion as an example of the optmusatlon process The model assumes that nothmg 1s known about possible mterations, therefore the parametric equation used, Eqn 1, contams fourteen terms Each of the terms represents either the variable or its mteractlon wth another vanable The squared terms are quadratics and allow for curvature of the experimental response The time variable IS not consldered as a quadratic as the response (percentage extraction) cannot reduce with mcreasmg tune Y = b, + b,u, + b,u, + b,v, + b4v4 + b,uf + b,u,u, + b,u,u, + b,u,u, + b,u,2 + b,,w,

+ b,,w,

+ b,,w,

+ b&

(1)

where, Y 1s the response (percent extracted), ul, $3 v3, u4 are the four mam variables temperaRegression Weights

s*grUficant Vanables

l

-5

F

I

I

I

I

“AR2

I

I

VA14

6

I

“AI.

I

0

“fiRI

I

“AR,@

I

I

“ill‘,

VhRI 1

“fi”.

“OR,

UAR5

“AR3

“RR1

UARlZ

Fig 3 Standardmd regresson weights of vanables for cmtlsone acetate * = Slgnlficant vanables

-I

88

M Kime et al /And

Chm Acta 271f1993) 83-W

TABLE 3 Significant vanable descnptors determmed by multlhnear regresslon Steroid

Slgmfkant vanable

Megestrol acetate Cortisone acetate Clobetasol Clobetasone Hydrocortwone Betamethasone-17,21dlproplonate

2 2 2--9 -

4 -

7 _

9 9 _

-

-

-

10 --

11 13 -

2--9_-

ered slgmflcant The slgmflcant coefficients are hlghhghted m bold m Table 2 Table 3 1s a summary of the srgmflcant vanables determined for the SIXsteroids by multilinear regression Two of the steroids (clobetasol and hydrocortlsone) show no significant variables and this may be attributed to their low solubdlty m carbon droxlde Of the remaining four steroids, the two common terms are the density (variable 2) and the density squared term (variable 9) suggesting that density IS the single major factor to affect the extraction of the steroids Flow-rate and temperature were also ldentlfled as slgmficant for cortisone acetate with mteractlons mvolvmg tune However, it IS apparen from the response curves that density 1s the mos slgruflcant variable m achieving optlmlmum ex traction condltlons Example surfaces of cortl sone acetate are given m Figs 4 and 5 for 2 lo-mm extraction time These show the optumur extraction conditions are approximately WC, 1 ml mm-’ and 0 80 g ml-‘, and that flow-rate ha6 a greater effect on extractlon than the tempera ture The response surfaces of the density and flow-rate based on the model were calculated fox the other compounds studied Clobetasol and hy drocortlsone, both of which failed to produce any slgmflcant variables (Table 3), succeeded m only showing a straight density gradient with no dependence on flow-rate Betamathasone-17,21-drproplonate and clobetasone both demonstrated some dependency on the flow-rate at lower denslties

Fig 4 Response surface for cortwone acetate studymg the effect of density and flow-rate on the extractton effuzlency at 55°C and 10 mm extraction time

Effects of cell geometry The effects of cell geometry 112-141 have been studled using the commercially avallable 15d

and 7 O-ml extractlon cells for one steroid under the same condltlons Each cell 1s constructed of stamless steel and has an internal diameter of 10 cm The 15ml cell has a length of 4 0 cm m contrast to the 7 O-ml cell which has a length of

Fig 5 Response surface for cortwne acetate studymg the effects of density and temperature on the extractlon efkency at 3 0 ml mm-’ and 10 mm extractlon time

M Kane et al /Anal

89

Chm Acta 271 (1993) 83-90

and high density combmatlons (the darker shaded regon) Increasing the flow-rate for the 15-ml cell apparently reduces the interaction between the supercrrtlcal fluid and the steroid The 7 O-ml (Fig 7) cell has the effect of makmg the flow-rate much less critical to the optmmm extraction recovery It 1s probable that the supercritical fluid flow wlthm the sample cell has some effect on the extraction efficiency m SFE It IS thought that turbulent flow 1s more desirable than lammar flow as the resulting eddies produced Hrlthm the sample cell wdl aid diffusion within the sample matrvr The Reynolds number [15] (Eqn 3) Fw 6 Response surface for megestrol acetate studymg the effects of density and flow-rate on the extractlon efficiency usmg 15ml cell at 55°C and 10 mm extractlon tune

9 5 cm The results for megestrol acetate, a steroid totally soluble m carbon dlxoxlde, are shown m Figs 6 and 7 The response surfaces show the contrastmg effects of flow-rate and density when only the volume of the cell 1s changed With the 15-ml cell there IS quite clearly a drop m response with an mcreasmg flow rate at an mtermedlate density (Fig 6) At higher densities the drop m response 1s not as pronounced, although the optnnum response region 1s restncted to low flow

Fig 7 Response surface for megestrol acetate studymg the effects of density and flow-rate on the extractlon effklency using the 7 O-ml Cell at 55°C and 10 mm extractlon time

pvd Re = CL

(3)

(where p = density of carbon dloxlde, v = velocity of carbon dloxlde flow, d = internal diameter of cell, ~1= vlscoslty of carbon dloxlde) was used to investigate the type of flow which exists within the sample cells both of which have the same internal diameter (0 01 m) The calculation 1s dependent on high density conditions (0 9 g ml-‘), a flow-rate of 3 ml mm-’ and an extraction tnne of 10 mm The vlscoslty of carbon dloxlde was taken as the liquid value at 15 x 10e4 N s mm2 The small amount of mass flow-rate, calculated to be45~10-~kgs-‘atadensltyof092gm1-’, and the internal diameter for the sample cells means that the Reynolds number will be considerably lower than the 2000 threshold value for the transition between lammar and turbulent flow, even assummg low (gas-hke) vlscos~ty conditions for the carbon dloxlde The calculated values for the Reynolds number varied between 40 and 400 according to the vlscoslty of carbon dloxlde value taken Thus lammar flow 1s the most hkely type of supercntlcal fluid flow wlthm the sample cell This situation may be beneflclally altered when the extractlon cell is packed with sample where a disruption of the supercntlcal fluid flow may produce sufficent eddies to create turbulent flow However, this was not the situation m this expenment as the bulk of the extraction cell was empty Therefore, changes m supercntlcal fluid fluid-flow were ruled out as a reason for the observed effects when cell srze was changed

90

M Kane et al /Ad

TABLE 4 Comparison of cell volumes swept usmg 15-ml and 7 O-ml cell volumes Density (g ml-l)

Cell volume swept (ml) 15-ml cell

7 O-ml cell

04 06 09

41 19 14

7 4 3

One factor which has undoubtedly altered by usmg dtierent sued cells 1s the number of cell volumes swept with supercrItIcal carbon dloxlde The number of cell volumes swept can be calculated usmg Eqn 4 Cell volumes swept Mass of carbon dloxtde (g) = Density (g ml-‘) X cell volume (ml)

(4)

where mass of carbon dloxlde = density of liquid carbon dloxlde (0 92 g ml-l) x flow-rate (ml mm-‘) X time of extraction (mm) The effect on the mass transfer of the analytes by reducing the number of cell volumes swept 1s likely to be detrimental to the extraction process This 1s supported by the fact that the percentage recoveries for the 7 O-ml cell were lower by approxtmately 20-40% than those of the 15-ml cell quoted m Table 1, under exactly the same conditions With mcreasmg density of supercntlcal carbon dloxlde the number of cell volumes swept for a given time, will be reduced (Table 4) When the density approaches 0 9 g ml- ’ the number of cell volumes swept for the 7 O-ml cell 1s only sufficient for partial mass transfer to occur Thrs 1s shown m Fig 7 as the decrease m extraction efficiency that occurs at a density of 0 8 to 0 9 g ml- ’ and flow-rates of 2-2 5 ml mm-’ Increasing the flow-rate will compensate for this effect as the mass of carbon dlo=de used 1s then increased It IS therefore not recommended to extract when the number of cell volumes swept of carbon dloxlde 1s below 4 coiKlzwo?ls

It 1s apparent from the results that density 1s the major variable for the extraction mto supercritical carbon dloxlde of the steroid compounds

Chun Acta 271 (1993) 83-W

studied Although the expemental optmusatlon can be applied only to each mdlvldual compound, the overall trend pes a basis on which to approach unknown steroids It must be noted that mdlvldual sample matrices will affect the relative contrlbutlons of the extraction parameters The mmlmum number of cell volumes swept for effective extraction has been determmed experunentally to be four Below this there 1s msufflclent contact between the sample and supercrrtlcal carbon dloade This 1s particularly unportant when using cell sues of different dunenslons The financial support of ICI plc, Wilton Materials Research Centre and Glaxo Manufacturing Services 1s gratefully acknowledged, and partlcularly the cooperation of Dr W Campbell (ICI) and Mr K Lelper (Glaxo) Finally, we acknowledge the loan of the HP7680A SFE from Hewlett-Packard (UK) Ltd

REFERENCES S B Hawthorne, Anal Chem ,62 (1990) 633 R E Majors, hq Chromatogr - Gas Chromatogr , 4 (1991) 10 M L Lee and K E Marlades, Analytical Supercntlcal Fhud Chromatography and ExtractIon, Chromatography Conferences, Inc , Provo, UT, 1990, Chap 5 J W Kmg, J Chromatogr Scl , 27 (1989) 355 R L Mason, RF Gunst and J L Hess, Statistical Design and Analysis of Expenments, Wiley, &Chester, 1989 R G Brereton, Chemometncs Apphcatlons of MathematICSand Statitzs to Laboratory Systems, Ellis Honvood, &Chester, 1990 7 P L Goldsnuth, CA E D ICI Frbres, Harrogate, 1987, personal communication 8 CSS Statistica, Release 3 OF, Statsoft U K ,Letchworth 9 M Kane, JR Dean, S M Hit&en, C J Dowle and R L Tranter, Anal Proc , 29 (1992) 31 10 J C Miller and J N Mdler Statistics for Analytical Chemistry, Ellis Horwood, Chichester, 1984 11 E P Box, W G Hunter and J S Hunter, Statistics for Experimenters, Wiley, New York, 1978 12 KG Furton and J Rem, Anal Chun Acta, 248 (1991) 263 13 J Rem, C M Cork and KG Furton, J Chromatogr , 545 (1991) 149 14 KG Furton, and J Rem, Chromatographla, 31 (1991) 297 1.5 J M Coulson and J F Richardson, Chemical Engmeermg, 4th edn , Pergamon Press, Oxford, 1990