|
1. |
Critical Review: Analytical Extraction of Additives FromPolymers |
|
Analyst,
Volume 122,
Issue 9,
1997,
Page 101-116
Harold J. Vandenburg,
Preview
|
|
摘要:
Critical Review Analytical Extraction of Additives From Polymers Harold J. Vandenburga, Anthony A. Clifford*a, Keith D. Bartlea, John Carrollb, Ian Newtonb, Louise M. Gardenb, John R. Deanc and Claire T. Costleyc a School of Chemistry, University of Leeds, Leeds, UK LS2 9JT b ICI Technology, Research and Technology Centre, P.O. Box 90, Wilton, Middlesbrough, Cleveland, UK TS90 8JE c Department of Chemical and Life Sciences, University of Northumbria at Newcastle, Ellison Building, Newcastle-upon-Tyne, Tyne and Wear, UK NE1 8ST Summary of Contents Introduction Conventional Extraction Techniques Dissolution of the Polymer Liquid–Solid Extractions New Developments in Extraction From Polymers SFE MAE ASE Supercritical Fluid Extraction Extraction Process Factors Affecting SFE From Polymers Effect of Supercritical CO2 on Polymers Effect of Temperature and Pressure Effect of Flow Rate Effect of Modifiers Nature of the Extractant Effect of Particle Size Summary of SFE Microwave Heating Ultrasonic Extraction ASE Other Extraction Methods Conclusions References Keywords: Review; polymer; supercritical fluid extraction; microwave-assisted extraction; liquid extraction Introduction Plastics contain many small molecules as well as the polymer itself.These include additives to alter the polymer properties or prolong the life of the polymer, such as plasticisers, antioxidants and ultraviolet (UV) light absorbers. There may also be processing aids, residual monomers, low molecular weight oligomers and inadvertent contaminants present.It is important for the manufacturer and regulators to know the level of these materials in the polymer to ensure the product is fit for its intended purpose. Food contact plastics are regulated by maximum concentrations allowable in the plastic, which applies to residual monomers and processing aids as well as additives. 1–3 There are some methods for determining concentrations of additives without extraction from the polymer, such as nuclear magnetic resonance spectrometry,4 UV spectrometry, 5 and UV desorption–mass spectrometry.6 However, in order to determine the levels in the polymer it is usually necessary to extract the compounds from the plastic quantitatively before analysis.The trade names and chemical names of many commonly used additives are given in Table 1. Conventional Extraction Techniques These can be divided into two categories: dissolution of the polymer and liquid–solid extraction methods.Many examples can be found in texts such as Crompton,7 Haslam et al.8 and Wheeler.9 A review of extraction methods to 1992 is included in Cotton.10 Dissolution of the Polymer A British Standard method11 describes the dissolution of polymers in refluxing toluene, with re-precipitation of the polymer by addition of ethanol. Decalin heated to 110 °C has been used as a solvent for poly(ethylene) (PE)12 and at 150 °C for poly(propylene) (PP).13 The high molecular weight polymer precipitates out as the solution is cooled and the supernatant solution is filtered and analysed. Formic acid is used to dissolve poly(amides) before further fractionation, for example by liquid–liquid extraction with toluene for 18 h to remove lubricants.14 Poly(ethylene terephthalate) (PET) has been dissolved in hexafluoropropan-2-ol–dichloromethane mixtures and the polymer precipitated by addition of acetone or methanol.15,16 Analysis of poly(vinyl chloride) (PVC) has traditionally been by sequential diethyl ether and methanol extractions to remove selected components, then dissolution of the polymer in tetrahydrofuran (THF). Centrifugation at two speeds would then produce two further fractions.Methanol would then be added to precipitate the polymer, and the methanol–THF would be evaporated and examined for non-extracted material.7 This Harold Vandenburg graduated from Manchester Polytechnic where he studied Applied Chemistry parttime whilst working in quality control in the pharmaceutical/ personal care sector. After working for a year in Australia and USA, he started six years of research on migration from polymers into foods at high temperatures at the Procter Department of Food Science at the University of Leeds.He obtained his PhD during this time. Still at the University of Leeds, but now at the School of Chemistry, he is undertaking research comparing the effectiveness of sample preparation methods for polymers and environmental samples.Analyst, September 1997, Vol. 122 (101R–115R) 101Rprocess can be simplified by dissolving the polymer in THF, followed by centrifugation at 20 000 rev min21 to remove mostly inorganic fillers. Addition of ethanol precipitates the polymer and polymeric additives. Polymeric plasticiser can be re-extracted from the precipitated polymer.14 Dissolution and re-precipitation therefore provides an effective method of extraction.The advantage is that there is no possibility of some analyte remaining bound in the polymer network, although inclusion of the analyte in the re-precipitated polymer can occur. There is often a considerable amount of ‘waxes’ in solution, which may need to be removed before further analysis. Some workers have considered this too time consuming17 and prefer liquid–solid extractions. Liquid–Solid Extractions Here the analyte is extracted from the solid medium by a liquid, which is separated by physical means, such as filtration. There are many methods for carrying out these extractions including Soxhlet, sonication and shake-flask extractions.Spell and Eddy17 studied the extraction of additives from PP at room temperature and found that required extraction time varied linearly with polymer density and decreased with increasing particle size. They also found a large variation in extraction time for different solvents and additives.By powdering the polymer to 50 mesh size, 98% extraction of 2,6-di-tert-butyl-4-methylphenol (BHT) was achieved by shaking at room temperature for 30 min with carbon disulfide. To achieve the same recovery with isooctane required 125 min, and 2000 min were required to recover Santonox with isooctane. The importance of small particles is further demonstrated by Newton.14 Refluxing ground PP with chloroform for 1 h gives complete extraction. For films, 3 h are required and for unground granules 3 h are sufficient to provide an extract for identification purposes only.Ethoxylated tertiary amines can be extracted from PP by refluxing the ground material with 1,2-dichloroethane for 1 h. Refluxing the granules for 3 h gives only 85% extraction. Soxhlet extraction has often been used, with a variety of solvents. Whilst this method eventually gives good extraction efficiencies, the extraction rate is slow. Times for extraction typically vary from 6 h18 to 48 h.19,20 Even with long extraction times recovery is not always good.Perlstein21 obtained recoveries of only 59% for extraction of Tinuvin 320 from unground PVC after 16 h Soxhlet extraction with diethyl ether. However, recoveries rose to 97% from ground polymer. The choice of solvent is significant for the duration of the extraction. Wims and Swarin22 found that talc filled PP needed 72 h extraction with chloroform, but only 24 h with THF. Thus, small particles are often essential to complete the extraction in reasonable times, and the solvents must be carefully selected to swell the polymer.Solid–liquid extraction has been shown to be effective, but often very slow and requires large amounts of solvents. This results in dilute solutions requiring further concentration before analysis, which is time consuming and may result in the loss of volatile compounds. The use of large volumes of often toxic solvents is environmentally unsound, as well as expensive in purchase and disposal costs.Some effort needs to be made to select the most appropriate solvent for the extraction. Therefore, there are great savings to be made if the extraction time and the solvent usage can be reduced. New Developments in Extraction From Polymers The principal objectives of any technique to replace ‘traditional’ extraction methods is to complete the extraction in less time, using less solvent and also have the possibility of automating the process.Recent articles23,24 describe some new techniques, such as supercritical fluid extraction (SFE), microwave-assisted extraction (MAE) and accelerated solvent extraction (ASE), and some applications are summarized. Although the purpose of this review is not to describe the techniques in detail, a brief description is given. SFE SFE uses fluids above their critical temperature and pressure. These supercritical fluids have densities and diffusivities between those of liquids and gasses.The solvating power is Table 1 Commonly used additives in polymers Trade name Chemical name BHT 2,6-Di-tert-butyl-4-methylphenol Chimassorb 81 2-Hydroxy-4-n-octoxybenzophenone Chimassorb 944 Poly(N-1,1,3,3-tetramethylbutyl-NA,NB-di(2,2,6,6-tetramethylpiperidinyl)-NA,NB- melaminoditrimethylene Cyasorb UV 531 2-Hydroxy-4-n-octoxybenzophenone DSTDP Distearyl 3,3A-thiodipropionate Erucamide 13-cis-Docosenamide Ionox 330 1,3,5-Trimethyl-2,4,6-tris(3,5-di-tert-butyl-4-hydroxybenzyl)benzene Irgafos 168 Tris(2,4-di-tert-butylphenyl) phosphite Irganox 1010 Pentaerythrityl-tetrakis(3,5-di-tert-butyl-4-hydroxyphenyl) propionate Irganox 1076 Octadecyl-3,5-di-tert-butyl-R-hydroxyhydrocinnamate Irganox 1098 N,NA-Bis{[3-(3A,5A-di-tert-butyl-4A-hydroxyphenyl)-(propionyl)]}- hexamethylenediamine Irganox 3114 1,3,5-Tris(3,5-di-tert-butyl-4-hydroxybenzyl)s-triazine 2,4,6-(1H, 3H,5H)trione Isonox 129 2,2A-Ethylidenebis(4,6-di-tert-butylphenol) Nauguard 524 Tris(2,4-di-tert-butylphenyl) phosphite Santonox 4,4-Thiobis(6-tert-butyl-m-cresol) Tinuvin 328 2-(2A-Hydroxy-3,3,5-di-tert-amylphenyl)benzotriazole Tinuvin 770 Bis(2,2,6,6-tetramethylpiperidin-4-yl) sebacate Tinuvin P 2-(2-Hydroxy-5-methylphenyl)-2H-benzotriazole Topanol CA 1,1,3-Tris(2-methyl-4-hydroxy-5-tert-butylphenyl)butane Ultranox 626 Bis(2,4-di-tert-butylphenyl)pentaerythritol diphosphite Weston 618 Distearylpentaerythritol diphosphite 102R Analyst, September 1997, Vol. 122related to the density, which in turn depends on pressure and temperature. Increasing pressure at constant temperature increases density and solvating power. However, increasing temperature at constant pressure decreases density and hence solvating power. CO2 is most widely used as solvent because of its convenient critical temperature (31.3 °C) and pressure (7.4 MPa), low cost, low toxicity and non-explosive character. Liquid solvents called modifiers are sometimes added to the CO2 in order to increase solubility or displace analytes from a matrix.MAE Sample and solvent are placed in a container and heated using microwave energy. The technique has evolved from closedvessel microwave acid digestions. The apparatus typically consists of closed vessels with temperature and pressure control, allowing the solvent to be heated under pressure above its normal boiling-point and remain liquid. The solvent must contain a component with a high relative permittivity to be heated by microwaves.Carousels of extraction vessels may be used, allowing for simultaneous extraction of up to 12 samples. ASE The sample is loaded into an extraction cell and solvent is pumped into the cell, which is heated in an oven. The temperature and pressure are programmed by the user. Pressure is applied to keep the solvent liquid above its normal boilingpoint. After the pre-set extraction time, more solvent is pumped through the cell into the collecting vessel and the remaining solvent is purged into the collecting vessel with nitrogen.The equipment is automated, allowing up to 24 sequential extractions to be programmed. In this paper, published accounts of laboratory methods for the extraction from polymers are reviewed. The new methods are here compared with each other and with traditional methods. Although there is a considerable literature on SFE, there are few reports on the other novel techniques.Supercritical Fluid Extraction SFE has been used for a great many matrices, including the extraction of environmental contaminants, natural products and food processing as well as polymers, which are discussed in several reviews.25–29 Extraction from polymers is one of the more recent uses of SFE. Early experiments established that SFE of polymers was a viable extraction method and could be much faster than traditional methods. Anton et al.30 extracted oligomers and an unidentified additive from an ethylene– propylene O-ring using CO2 at 25 °C and 6.6 MPa for 1 h.A second extraction produced more analyte, indicating that the extraction was not complete after the first hour. Hirata and Okamoto31 found that additives could be extracted at 90% recoveries from PE and PP films using CO2 at 25.4 MPa and 35 °C within 2 h, compared with 24 h for Soxhlet extraction. Cotton et al.32 showed that SFE of additives and oligomers from PP, nylon, PET and poly(ether-ether ketone) (PEEK) were possible.Irganox 1010, Irgafos 168 and erucamide were extracted from ground PE and Tinuvin 770 from PP at 92–95% recoveries in 15 min at 42.9 MPa and 60 °C. At 14.3 MPa the same recovery was achieved in 30 min.33 Following these successful extractions the optimum extraction conditions have been investigated. Extraction Process The two main factors in SFE are solubility of extractant in the fluid and rate of mass transfer out of the matrix.The mass transfer from the polymer is by diffusion from the bulk polymer to the surface, where dissolution in the supercritical fluid (SF) can occur. Bartle et al.34 described a simple model for extraction from spherical particles called the ‘hot ball’ model. This assumed that the concentration of extractant in the SF was effectively zero, and the only limiting step was transfer out of the matrix by a process which could be modelled as diffusion. This successfully predicted the characteristics of the extraction curve of ln(m/m0) versus time for extraction of BHT from PP (where m0 is the initial concentration in the plastic and m is the amount remaining in the plastic).These are that the curve falls steeply initially, as the extractant is extracted from the surface, and then becomes linear (Fig. 1). There were deviations from the predicted onset of the linear portion and of the extrapolated intercept of the linear portion with the m/m0 axis. These were explained in terms of non-spherical particles, solubility limitations and non-uniform distribution of the extractant in the polymer. The success of the model indicates that the basic processes can be modelled as diffusion.However, the deviations indicate that solubility and other factors are also significant. The linear part of the logarithmic plot can be extrapolated to determine the concentration of an extractant without complete extraction by using three extraction periods of equal duration.The amount m0 can be found using the equation m0 = m1 + [(m2)2/(m2 2 m3)]. The first extraction must be long enough such that the second and third extractions are on the linear portion of the curve. The extrapolation method successfully gave the extractable amount of BHT from PP. The model was extended to cover polymer films and non-uniform distribution of the extractant using extraction of cyclic trimer in PET as an example.35 The temperature chosen (70 °C) was just above the glass transition temperature (Tg) (69 °C), but the amount extracted was considerably less than that extracted by Soxhlet extraction.Thus, much of the trimer is unavailable to SFE at this temperature. This is consistent with the findings of Ashby,36 who found that the overall migration from PET into olive oil was negligible until a temperature of 130 °C was reached. The effect of solubility on extraction was incorporated into the model by Bartle et al.37 using extraction of Irgafos 168 from PP to test the model.At 70 °C, the effect of increased pressure (solubility) on the extraction curve was to increase the slope of the initial steep ln(m/m0) curve and increase the slope of the linear portion of the graph, i.e., increased extraction rate. The Fig. 1 Example of ln(m/m0) plot for SFE from polymers using the ‘hot ball’ model. Analyst, September 1997, Vol. 122 103Reffects of pressure and flow rate on extraction are explained theoretically by Clifford et al.38 A method for direct observation of the extraction process in real-time is described by Howdle et al.39 using an organometallic complex.The IR absorption bands are sensitive to the environment, and therefore a shift in the spectrum occurs when the complex is extracted from the polymer. This allows monitoring of the extraction process in situ. SFE is therefore an effective method of extraction from polymers. There are several factors that can affect the success of SFE: temperature, pressure, time, addition of modifier, the matrix and the compound extracted.The interaction of these variables is particularly complex for extraction from polymers, partly because the solvent can interact with the polymer. Factors Affecting SFE From Polymers Effect of Supercritical CO2 on Polymers CO2 can be dissolved in polymers and effectively plasticise and swell them. A high pressure view cell with a sapphire window has been used to measure the swelling of soils and plant materials during treatment with supercritical CO2.40 A direct relationship was observed between the degree of swelling and the efficiency of extraction.Similar effects would be expected from polymers, with the size of the effect depending on the amount of CO2 absorbed. The amount of CO2 absorbed will depend on temperature, pressure and the polymer concerned. Shieh et al.41,42 studied the effects of supercritical CO2 on nine crystalline and 11 amorphous polymers.The appearance, mass changes, physical properties and solubility of CO2 were examined after removal from the fluid. The amorphous polymers were most affected, many exhibiting significant swelling, particularly poly(methyl methacrylate) (PMMA). Tg was measured for PMMA and glycol-modified poly(ethylene terephthalate) (PETG) and was found to be depressed after exposure to CO2. Crystalline polymers were plasticised less than the amorphous polymers.The Tg of PET was found to be depressed by 52 °C when CO2 was sorbed at 2 MPa.43 The greater effect on amorphous polymers is expected as it has been reported that CO2 is not soluble in the crystalline regions of polymers.44 The strength of the effect on PMMA was possibly explained by investigations using Fourier transform infrared spectrometry (FTIR), which determined that there were specific Lewis acid–base type interactions between the carbonyl group in PMMA and CO2.45 No such interaction was found in PE, and a weak interaction was found between CO2 and the p system of poly(styrene) (PS).Therefore, we can expect the greatest swelling and plasticisation in amorphous polymers with electron donating groups. Extraction at temperatures above Tg will generally be much faster than below Tg. Kalospiros and Paulaitis46 have developed a molecular thermodynamic model for predicting solvent-induced glass transitions as a function of sorbed gas in the polymer. However, CO2 can also induce crystallisation in polymers at lower temperatures,47 which would be expected to reduce the rate of diffusion.Condo and co-workers48,49 measured the depression of the Tg of PMMA, poly(ethyl methacrylate) (PEMA) and poly(methyl methacrylate- co-styrene) (SMMA60) by CO2 in situ, using creep compliance in a high pressure cell. The relationship with pressure and temperature was not simple, and four types of behaviour were predicted and found.The phenomenon of reverse vitrification was identified, where the polymer undergoes a glass to liquid transition by decreasing the temperature. This means that some polymers may be plasticised under much milder conditions of temperature and pressure than was previously believed. The interaction of CO2 with polymers can rapidly swell and plasticise the polymer, resulting in much faster diffusion and hence extraction. The plasticisation alters the Tg and softening points, and therefore the temperature and pressure selected for the extraction can be lower than would be expected from physical data available on the polymer.Effect of Temperature and Pressure Diffusion in polymers is a slow process with diffusion coefficients of the order of 10210 cm2 s21 at 40 °C. The rate of diffusion follows an exponential Arrhenius form with temperature, rate » A exp (2E/RT), where E is the activation energy, R is the gas constant and T the absolute temperature.This would indicate that increasing the temperature would exponentially increase the diffusion rate. If the solubility in the SF is not limiting, then higher temperature would lead to higher extraction rate. However, increase in temperature decreases the density of a SF at constant pressure, which reduces the solubility of the extractant in the SF. Therefore, it might be expected that a rise in temperature would lead to a rise in extraction rate, up to a point where solubility in the SF became a limiting factor. Increasing the temperature can also cause the polymer to undergo a transition from a glassy to a rubbery form at the glass transition temperature.The diffusion in the rubbery form is much faster than the crystalline glassy form. Therefore, a sharp rise in extraction rate would be expected at the Tg. Increasing the pressure increases the density of the SF and therefore the solubility of the extractant. At low temperatures, when the extraction is almost completely diffusion-limited, this should have little effect.The situation is further complicated because supercritical CO2 can plasticise a polymer and lower the Tg. The amount of lowering increases with amount of fluid in the polymer, which depends on the pressure and temperature. Supercritical CO2 has a much greater effect on amorphous polymers than on crystalline polymers. Therefore, the effect of pressure and temperature would be expected to be more pronounced in amorphous polymers.Increasing the pressure may increase the extraction rate in amorphous polymers, even when solubility does not limit the extraction as the plasticisation of the polymer increases with absorbed CO2. The effect of temperature on extraction of oligomers from PET was studied by Kuppers.50 Extraction at a constant CO2 density of 0.5 g cm23 (hence changing pressure) showed an increase in the rate of extraction of trimer from PET from 40 to 160 °C, but there was no jump at the Tg.Increasing the density at the same temperature gave even higher extraction rates, indicating that solubility is also limiting the extraction. However, it is possible that increased plasticisation of the polymer at higher pressures was responsible for the increase rather than the increased solubility. Plotting extraction at constant pressure (changing density) against T showed a sharp increase at Tg, then a flattening of the curve at high temperatures, presumably as solubility again became limiting.The lack of a jump in the constant density curve may indicate a change in the Tg at higher pressures of CO2. Cotton et al.51 extended the work on PET film to higher temperatures. They found an increase in extraction with temperature at constant pressure (40.6 MPa), but no discontinuity at Tg. Above 215 °C some melting took place, causing agglomeration which slowed extraction and blocked the restrictor, so 215 °C was the highest temperature used for extractions.The extrapolation method at this temperature gives quantitative recoveries using 3 3 30 min extractions, and 95% recoveries using 3 3 15 min extractions. The intercept of the extrapolated linear portion of the graph is lower than at 70 °C, indicating that solubility, as well as diffusion from the polymer, affects the extraction at the higher temperature. Schmidt et al.52 analysed carbonic acid diphenyl ester from poly(butylene terephthalate) (PBT) by SFE.At room temperature, six times less was extracted than at 55 °C. Hunt and Dowle53 found that extraction of diisooctyl phthalate (DIOP) and Topanol CA from 104R Analyst, September 1997, Vol. 122ground PVC increased with temperature up to 90 °C at 45 MPa, then levelled off, presumably as solubility became the limiting factor. The extraction rate increased on increasing the pressure from 35 to 40 MPa at 45 °C; increasing the pressure further to 45 MPa had no further effect.The extraction of dioctylphthalate (DOP) and dibutylphthalate (DBP) plasticisers from PVC was also measured for 25 min extractions by Marin et al.54 At a pressure of 52 MPa, the amount extracted by CO2 increased from 50 to 80 °C, at which temperature extraction was almost complete within the 25 min. At a constant temperature of 70 °C, increasing the pressure from 22 to 60 MPa increased the amount extracted from approximately 79 to 95%. At 100 °C, the same pressure increase led to a marginal increase in extraction from 95 to 98%.As the extraction is close to completion at the higher temperatures, it is not possible to see the full effect of the pressure increase. However, from solubility considerations, the effects should be greater at higher temperatures because the extraction then becomes solubility-limited. Extraction of tris(nonylphenyl) phosphite, Irganox 1076 and Weston 618 from low-density PE (LDPE) film was measured at 45.6 MPa at different temperatures.55 At 60 °C, only 40–60% was extractable.However, by heating to 150 °C, which is 10 °C above the melting-point, > 95% was recovered. This is against the general finding that heating above the melting-point reduces extraction rate as melting reduces the surface area of ground samples. Extraction of N,N-ethylenebisstearamide from PS increased from 77 to 94% by increasing the temperature from 60 to 150 °C. Extraction efficiency of BHT and Irganox 1010 from freeze-ground PP increased from 30 to 90 °C at a pressure of 30.4 MPa.56 At 90 °C, 60 min were sufficient for complete extraction of all additives except for Irganox 1010.During the extraction of flame retardants from poly(urethane) (PU), Mackay and Smith57 found that extraction at 30.4 MPa was rapid and quantitative within 10 min at 60 °C, whereas at 20.3 MPa recoveries were lower, and at 10.1 MPa some flame retardants were hardly extracted at all. The effect of pressure in selective extractions was pointed out by Engelhardt et al.58 At 15.2 MPa and 45 °C, erucamide was extracted from PE.By increasing the pressure to 20.3 MPa, an Irganox-type antioxidant was also extracted. They linked the SFE conditions needed to extract materials with their retention using reversed-phase (RP) HPLC. If 90% organic modifier is needed to elute a compound from an RP HPLC column, then an extraction pressure of 15.2–20.3 MPa is sufficient. If a compound elutes with 50–60% organic modifier by HPLC, then the pressure for SFE using CO2 needs to be 35.5 MPa.Cotton et al.59 measured the extraction of Irgafos 168, Irganox 1010 and Tinuvin 770 from PP pellets and ground PP. Increasing pressure gave increased extraction rates at 50 °C up to a limit of 30.4 MPa, after which solubility no longer limited the extraction. The rate increased sharply with increasing temperature, up to the melting-point of the polymer. Once melting occurs the particles coalesce and the surface area decreases.The diffusion coefficient was found to be two orders of magnitude higher than from previously published data. This was explained by the plasticisation and swelling of the polymer by sorbed CO2. The value of the extrapolation procedure was demonstrated by successfully determining the total concentrations of Irganox 1010 and Irgafos 168 in less time than was required to complete the extraction from ground PP. For the larger pellets, low results were obtained from the extrapolation because the ln(m/m0) versus time plot had not reached the linear portion after the first extraction period.Recovery of dialkyl organotin stabilisers from freeze-ground, unplasticised PVC at 17.7 MPa was found to increase with temperature up to 90 °C, then levelled off with further temperature increases, presumably as solubility became limiting. 60 Similarly, the recovery at 90 °C increased with increasing pressure up to 15.2 MPa and then levelled off as diffusion became limiting.The extraction was complete within 60 min at 90 °C and 17.7 MPa. No data are given for recoveries above 90 °C at pressures above 17.7 MPa, but unless melting had commenced at this temperature the extraction rate should be enhanced as solubility-limiting behaviour would be reached at a higher temperature. Extraction of ethylbenzene from PS was neither completely diffusion nor solubility limited, leading to unusual T and P relationships.61 The polymer was significantly swelled by the CO2, leading to a 106-fold increase in diffusivity of ethylbenzene in the polymer. Tg was lowered from 100 to 50 °C by only 6 MPa of CO2.Roston et al.62 used SFE to extract a sustained release drug covalently bonded to poly(butadiene). Optimum pressure for the extraction was 33.5 MPa. Increasing the pressure above this reduced the amount extracted. These workers suggest that this decrease at higher pressures was due to recovery increasing with linear velocity of extracting media through the cell up to a certain velocity, then decreasing.Decreased trapping efficiency at faster flow rates is also a possibility. The temperature could not be raised above 75 °C because the drug is not thermally stable. Significant swelling of the polymer was noted; 85% of the drug was recovered in 45 min with a further 5% recovered in the subsequent 30 min. Extraction of Irgafos 168, Irganox 1076 and Irganox 1010 from PE powder increased with increasing temperature at 30.4 MPa.63 However, at 15.2 MPa the amount extracted during the first 30 min initially increased with temperature as diffusion was faster, but then fell as solubility-limiting behaviour took over.The amount extracted at longer times was always higher at higher temperatures, as the concentration in the extracting fluid fell and diffusion control was re-established. At 50 °C, increasing pressure had little effect as the extraction was almost completely diffusion-controlled.At 80 °C, the extraction rate increased from 20.3 to 30.4 MPa. The recovery at 25.4 MPa and 80 °C was higher than by 36 h Soxhlet extraction with hexane. Lou et al.64 examined the extraction of caprolactam from nylon 6, and dimer and trimer from PBT. Extraction of caprolactam during the first 30 and 120 min increased with increasing temperature from 50 to 170 °C. No jump at the Tg was apparent, implying that the CO2 had suppressed the Tg.However, during extraction from PBT, the amount of the dimer extracted after 120 min was highest at 150 °C and for the trimer at 110 °C. This indicates that the larger trimer is less soluble than the dimer, and extraction becomes solubility-limited at a lower temperature. The effect of temperature on the extraction of styrene dimers and trimers from PS was demonstrated by Jordan et al.65 A very low density of CO2 was used (0.184 g cm23), and kept constant at different temperatures.Extraction at 120 °C (10.8 MPa) produced eight times more dimers and trimers than extraction at 80 °C (8.8 MPa). The effect of the density of CO2 was examined by extracting at 120 °C at 20.3 MPa. The extraction time was reduced so that the same total amount of CO2 was used as in the lower pressure experiment. Exact comparisons of peak areas were not possible because the chromatographic system was overloaded, but at the higher pressures a sample size about 25% of that used at the lower pressure caused saturation. Therefore, the higher pressure extraction resulted in significantly greater extraction.The effects of pressure are more pronounced for extraction from an amorphous rubber. Burgess and Jackson66 found that extraction of carbon tetrachloride from chlorinated poly- (isoprene) could be completed within 40 min at 60 °C and 21 MPa, and not at higher temperatures or pressures. The normal Tg is 120 °C, but the softening point is lowered by the CO2 at high pressures. Therefore, increasing the pressure at high temperatures lowered the softening point and the polymer particles coalesced, reducing the surface area.Extractions on Analyst, September 1997, Vol. 122 105Rcrystalline polymers at low temperatures usually show little improvement with increasing pressure. However, in this case at 40 °C the extraction improves with increasing pressure. At this low temperature the extraction is unlikely to be solubilitylimited, and therefore the greater extraction is probably due to increased swelling of the polymer at higher pressures.The polymer matrix affects the extraction efficiency, as demonstrated by Juo et al.,67 who extracted Chimassorb 944, distearyl 3,3A-thiodipropionate (DSTDP), Irganox 1010, Irganox 1076, Tinuvin 144, Irganox 1098, Ionox 330 and oxidised Nauguard 524 from LDPE and high-density PE (HDPE). Both polymers were extracted at 60.8 MPa and 40 °C, for 30 min for LDPE and 5 h for HDPE.All additives were extracted from the LDPE, whereas some additives would not extract from HDPE even with 3% methanol modifier. Generally, diffusion from the polymer is the rate-limiting process, particularly during the later stages. Therefore, for optimum extraction, a high temperature should be used to maximise diffusion. High pressure should also be used for rapid extractions, both to increase solubility of the analyte and to plasticise the polymer. However, high pressures may lower the softening point of the polymer, and therefore the optimum conditions need to be experimentally determined.Effect of Flow Rate The flow rate will affect the extraction only if the extraction is solubility-limited. In this way changes to flow rate are similar to changes in pressure. The effects of pressure and flow rate on SFE from polymers have been modelled.38 For a solubilitylimited extraction, increasing the pressure will increase the rate of extraction up to the point at which diffusion becomes ratelimiting. At faster flow rates, the pressure limit will occur at lower pressure. Therefore, increasing the flow rate has a similar effect to increasing the pressure.This does not take into account the increased plasticisation of the polymer at higher pressures of CO2. Hawthorne et al.68 found that the flow rate had little effect on extraction of alkylbenzenes from PS beads except at a very low flow rate of 0.25 ml min21.For extraction of antioxidant from PE powder, there was little effect of flow rate at 50 °C, but some effect was noted at 80 °C as the faster diffusion at the higher temperature raised the concentration in the fluid.63 In contrast, Baner et al.69 found a marked reduction in the time to determine total extractables from Biopol (a biodegradable polymer) with increased flow rate. At a flow rate of 1.5 ml min21, the extraction took 40–50 min, reducing to about 15 min with a flow rate of 4.5 ml min21 and 8 min for 8.5 ml min21.As the total volume of CO2 is similar in each case, it seems that this extraction is almost completely solubilitylimited. Effect of Modifiers CO2 is characterised as a non-polar solvent with a solubility parameter similar to hexane. It does, however, have some affinity with slightly polar molecules because of its molecular quadrupole. For more polar molecules the addition of polar modifiers is used to increase the polarity of the supercritical phase and the solubility of polar compounds.In extraction from environmental matrices, binding of the extractant to active sites is very important, and the modifier can displace the extractant from the matrix, increasing extractability. This is less likely to be a factor in extraction of polymers. However, if they are similar in molecular character, modifiers can act by swelling the polymer, increasing diffusion and extraction rates.Modifiers used for this purpose may be non-polar or aromatic as well as polar. Therefore, modifiers may enhance extraction even when solubility is not a limiting factor. Hunt and Dowie53 found that only 50% of Topanol CA would extract from PVC at 45 MPa and 90 °C within 30 min. The addition of methanol modifier enhanced the extraction. The recoveries increased with increasing methanol concentration in the SF from 2 to 15%. The total present in the polymer was not accurately known; therefore, it was not known whether the recovery was quantitative.Mixed messages on modifiers come from the work of Kuppers.50 Dichloromethane (DCM) enhanced extraction of oligomers from PET, the higher oligomers being most affected. Using methanol or propan-2-ol completely prevented extraction. The reason for the latter finding is not explained. Recovery of oligomers and caprolactam from ground nylon 6 was poor using CO2 at 60 °C.70 Addition of methanol to the extraction vessel during a 5 min static extraction period enhanced extraction, which was almost complete for caprolactam in 15 min.For oligomers the recovery was lower, but was further improved by using 7.5% methanol during the dynamic extraction stage. Methanol enhanced the extraction of caprolactam from nylon for Jordan et al.65 Spiking the extraction cell with methanol before extraction with CO2 at 75 °C and 25.3 MPa extracted approximately three times as much caprolactam as the same extraction conditions without methanol.The same amount of caprolactam spiked onto Celite was completely recovered using CO2 only, indicating that solubility limitations were not the reason for the low recovery from the polymer. These workers suggest that the methanol was swelling the polymer and therefore enhancing the extraction rate. Lou et al.64 examined the effect of modifiers on extraction of caprolactam from nylon 6, and dimer and trimer from PBT. The modifiers (hexane, chloroform, methanol and benzene) were spiked into the extraction cell for an initial static extraction period.Methanol was the most effective modifier for nylon and chloroform for extraction from PBT. When compared with extraction using unmodified CO2, the modifiers had the most effect at lower temperatures. In fact, at higher temperature the amount extracted at longer times was lower than when using CO2 alone. This was thought to be due to a shrinking of the polymer after the modifier had been extracted from the cell.Improvements in extraction efficiencies could be achieved at both diffusion and solubility limiting conditions, indicating that the modifiers worked both by swelling the polymer and increasing solubility. Addition of benzene during a static extraction with CO2 for 30 min improved recoveries of antioxidant from PE powder.63 Benzene was selected as it is known to swell PE, and this was thought to be the main reason for the greater recoveries.However, for Irganox 1010 the greater solubility was also thought to be significant. Garde et al.71 optimised the extraction of antioxidants from PP. They found that the best methods used two static extractions of 30 min, one with hexane and one with methanol. Use of methanol as modifier in a 60 min dynamic extraction only slightly enhanced the recoveries, indicating that swelling of the polymer is the most important factor. Propane added to CO2 had little effect on the extraction of chlorofluorocarbons (CFCs) from PU foams.72 Using unground foam, the time required for 99% extraction was 1.5 h, whereas with addition of propane (6% m/m) 2 h were needed.However, using material ground to 0.2 cm particle size, 1 h was needed with CO2 alone and only 0.6 h with addition of propane. Compounds may be added to the supercritical phase as a reactant rather than as a simple modifier. Roston et al.62 used formic acid to hydrolyse the bond between the polymer and a drug.Using methanol as a modifier there was almost no extraction. Cedergren et al.73 extracted nicotine from PE patches using modified CO2. They found that 1 m triethylamine in methanol was the best modifier. The tertiary amine was added to prevent reaction with the CO2, which could lead to insoluble carbamates. Pre-treatment of the PE with concentrated triethylamine together with a static extraction stage gave the best recoveries. 106R Analyst, September 1997, Vol. 122Modifiers generally accelerate extraction when the modifier interacts with the polymer more than CO2. The modifier can then cause greater swelling than with CO2 alone. Therefore, methanol is useful when extracting from polar nylons, and aromatic modifiers from non-polar poly(olefins). The greater solubility in modified CO2 is most effective when extracting large or polar molecules. Nature of the Extractant The nature of the extractant affects extraction in both solubility and diffusion limiting cases.The larger the molecule the slower the diffusion in the polymer, and hence the slower the extraction. High molecular weight compounds tend to be less soluble in supercritical CO2 and hence solubility will also limit the extraction more for larger molecules. The extraction of DIOP from PVC at 45 MPa and 90 °C was almost complete after 20 min, whereas the polar Topanol CA was only 50% extracted.53 Irganox 1010 has proved difficult to extract in a number of cases when smaller compounds were extracted.56,63 Higher nylon oligomers also proved difficult to extract.70 Extraction of flame retardants at 30.4 MPa for 10 min at 60 °C was complete for molecules of molecular weights 286, 388 and 472 Da, but the largest molecule with a molecular weight of 571 Da was only 77% extracted.57 There are some reports of smaller molecules extracting slower than larger molecules. The plasticiser DOP extracted faster from a 100 + 50 PVC–plasticiser blend than the smaller DBP.54 This was attributed to DBP being more strongly bonded to matrix sorption sites than DOP.However, it should be noted that DOP may be a more effective plasticiser than DBP, hence leading to faster extraction rates. Effect of Particle Size As one of the limiting steps in extraction is diffusion to the surface of the polymer, the particle size or film thickness is extremely important.52,53,59,60,70,74 The diffusion coefficient of additives in polymers at 40 °C is typically about 10210 cm2 s21.The rate of diffusion (s21) is proportional to D/L2, where L is the length of the shortest dimension. As a first approximation, therefore, for an extraction time of 1000 s (17 min) a particle diameter of 0.3 mm is required. Therefore, grinding of the polymer is often an essential step in the analysis. An exception to this is the extraction of thin films and foams, for which the shortest dimension is small.Garde et al.71 could extract no more than 50% of antioxidants from PP pellets, but could achieve 90% recoveries from the same polymer extruded into film. Loss of volatile additives is possible owing to the heat generated by grinding polymers. Therefore, the polymer must be frozen, usually with liquid nitrogen, before grinding. Hexabromocyclododecane was extracted from PS styrofoam and Irganox 1010 from PE ethafoam within 30 min at 150 °C at 45.6 MPa.55 Flame retardants were extracted from PU foam within 5 min at 30.4 MPa, 60 °C.57 A different approach to avoiding the need to grind the sample was taken by Mackay and Smith.75 Examination of polymer samples (unplasticised PVC) after SFE revealed that the CO2 had not penetrated into the core of the sample, and recoveries were low. The polymer was dissolved in THF and an internal standard added.The solvent was evaporated and the resulting polymer samples were extracted with CO2 at 50 °C, 35.5 MPa for 10 min.Extraction was far from complete, but the internal standard extracted to the same extent as the analyte and the correct analysis result was obtained. Extraction of additives from liquid polymers has been achieved by bonding them onto silica.76 Additives from poly(alkylene glycol) and sorbitan ester were extracted by SFE with CO2 at 45 °C. Oligomers were co-extracted, but most of these could be removed by a silica guard column. Non-quantitative extraction of additives from a range of polymers was demonstrated by Braybrook and Mackay77 for biocompatability testing.Solvent extraction is problematical for this purpose because solvent residues interfere with the biocompatability test. Mackay and Smith78 showed the value of on-line SFE–SFC–MS by identifying a range of additives from PU. Some of these additives could not be analysed by GC–MS. Summary of SFE SFE can provide a method for extracting additives from polymers. It is much faster than Soxhlet extraction.Grinding of the sample is usually necessary except for thin films and foams. Extractions are likely to be fastest from amorphous polymers, but the temperature must be carefully chosen to be lower than the softening point under experimental conditions. This softening point is likely to be lower than that of the polymer at atmospheric pressure, and the conditions of temperature and pressure must be carefully selected. For crystalline polymers, the temperature should be as high as possible without onset of melting, and the pressure also should be as high as possible to ensure no solubility limitations and maximum plasticisation of the polymer.Most difficulty is likely to be experienced with extraction of high molecular weight and/or polar compounds. The use of a modifier is likely to enhance the extraction in these cases. The modifier should be one that is known to swell the polymer. For extractions that still take a long time, an extrapolation procedure can be used to determine concentration in the plastic without complete extraction.Microwave Heating Microwave-assisted sample preparation techniques are widely used in analytical laboratories, largely in the field of digestion of samples. Zlotorzynski79 reviewed the use of microwave radiation in analysis and suggested that microwave extraction is in its infancy. Reviews on the applications of microwaveassisted sample preparation have recently been published.80,81 There are several publications on MAE in environmental analysis, but few on extraction from polymers.The advantages of MAE are that samples can be rapidly heated and several samples can be extracted simultaneously. The sample can be contained in a pressure-resistant vessel with safety valves. The solvent can therefore be heated above its normal boiling-point. At 1.2 MPa, the temperature reached in a pressure vessel with acetone is 164 °C, with dichloromethane 140 °C, and with acetonitrile 194 °C.81 The high temperatures will accelerate diffusion through the polymer and hence improve extraction rates. The high temperature may increase the swelling of the polymer owing to greater solvent–polymer interaction at higher temperatures.Another advantage compared with Soxhlet extraction is that any composition of solvent mixtures can be used. During Soxhlet extraction the solvent is the vapour condensate, which will only have the same composition as a mixture of solvents if an azeotropic mixture is used.The solvent selected must have a high relative permittivity to be heated by microwaves; therefore, pure hydrocarbons cannot be used. Freitag and John82 used acetone–heptane (1 + 1) to extract additives from LDPE, HDPE and PP. After grinding to 20 mesh, 91–97% of Irgafos 168, Chimassorb 81 and Irganox 1010 were extracted within 6 min from HDPE and within 3 min from PP and LDPE. Molecular weight had a significant effect, with Irganox 1010 being the slowest to extract. 1,1,1-Trichloroethane gave slightly faster extractions, but is environmentally less desirable. The order of extraction of Irganox 1010 from the polymers was PP > LDPE > HDPE, although the scatter of the LDPE results was fairly high. Dissolution of larger particles in toluene–1,2-dichlorobenzene was effected in 5 min. However, Analyst, September 1997, Vol. 122 107Rthe dissolution method gave only 85% recovery of the additives from the PE samples, although 95% recovery was achieved from the PP.Extractions from larger pellets were less efficient. The vessels were pressurised, but the temperatures and pressures reached were not reported. Nielson83 compared microwave extraction with sonication for HDPE, PP and LDPE. BHT, Irganox 1010 and Irganox 1076 could be extracted at > 90% recoveries from ground HDPE in 20 min at 50% power. Two solvent systems were used: propan- 2-ol–cyclohexane (1 + 1) and DCM–propan-2-ol (98 + 2).In each case the propan-2-ol was present to absorb the microwave energy and the other solvent to swell the polymer. The sample was stirred at 5 min intervals. The polymer was ground to 20 mesh, and 5 g of plastic were used to 50 ml of solvent. This is a much larger sample than the 50–100 mg typically used in SFE. Butylated hydroxy ethylbenzene (BHEB), Isonox 129 and erucamide slip agent were extracted from LDPE using the same conditions. From PP, Irganox 3114, Irganox 1010, Irganox 1076, Irgafos 168, Tinuvin 328, Ultranox 626, Cyasorb UV 531, BHT and AM 340 could be extracted from the ground polymer (20 mesh) [5 g of plastic to 50 ml of solvent, DCM–propan-2-ol (98 + 2)] in 20 min at 20% power, with stirring required every 5 min.Only the Irganox 3114 had a low recovery of 79%. When larger pellets were used the recoveries were also high, except for Irganox 1010, for which only 50% recovery was possible without grinding.The vessels were not pressurised and the temperature using the DCM–propan-2-ol mixture did not exceed 50 °C. Costley et al.84 report on extraction of cyclic trimer from PET using a variety of solvents heated to 120 °C in pressure vessels. The polymer fused at temperatures above 120 °C with DCM; therefore, higher temperatures were not investigated. Extraction for 2 h with DCM at 120 °C gave the same extraction as 24 h Soxhlet extraction with xylene as solvent. MAE using hexane– acetone (1 + 1), water, acetone and acetone–DCM (1 + 1) all gave much lower recoveries than DCM at 120 °C.From these examples, MAE appears to be a rapid and effective technique for polymer extractions. The solvent can be selected to swell the polymer, provided that some microwave absorbing solvent is also present. The polymer needs to be ground for efficient extraction. However, there are too few reports for firm conclusions to be drawn. Ultrasonic Extraction Ultrasonic extraction works principally by agitating the solution and producing cavitation in the liquid.This would be expected to enhance the rate of transfer across the polymer/liquid boundary layer, but not to increase the diffusion of compounds within the polymer. There are several reports of ultrasonic extraction from polymers. Brandt85 extracted tri(nonylphenyl) phosphite (TNPP) from a styrene–butadiene polymer using 2 3 20 min extractions with isooctane as solvent. This compares with 2 3 1 h extractions for boiling under reflux.Nielson83 compared ultrasonic extraction with MAE for extraction of a variety of analytes from PP, LDPE and HDPE (see under Microwave Heating). For all samples, the ultrasonic extraction could be achieved within 1 h, provided that the samples were stirred every 10 min. For LDPE and PP most compounds were extracted within 10 min. The exception was Irganox 1010, which required 1 h for > 90% extraction. Further experiments by Nielson86 on extraction from HDPE using the same regime confirmed these results. However, where phosphite antioxidants are present the use of DCM–cyclohexane was preferred as it prevented hydrolysis of the phosphite by the alcohol.Caceres et al.87 used the same solvent mixtures as Nielson83 to extract Tinuvin 770 and Chimassorb 944 from HDPE. The additives could only be extracted at less than 20% recoveries from pellets using ultrasonic extractions of up to 5 h. The size of the pellets is not given, but the fact that the sample was not ground may be the reason for the difference in results.Extraction of Chimassorb 81 from LDPE and ethylene-vinyl acetate polymer (EVA) was achieved with 6 h standing of the sample under DCM (maceration) followed by 3 3 20 min sonication.88 The initial maceration time allowed swelling of the polymer. The extraction time using maceration alone was 48 h. In this case it was important to avoid high temperatures as these could degrade the analyte.Ultrasonic extraction from polymers has given some reasonably fast extractions, but the advantages over shaking the sample have not been widely demonstrated. ASE Lou et al.89 extracted monomers and oligomers from nylon and PBT using hexane as extraction solvent in a home-made ASE. They investigated the effect of temperature, pressure and flow rate with 20 min static followed by 30 min dynamic extractions. Pressure was found to have no effect other than to keep the solvents liquid at high temperature and flow rate had little effect between 0.4 and 2 ml min21.Extraction efficiencies increased in all cases as the temperature was raised from 50 to 170 °C, which was attributed to faster diffusion rates. These workers observed that solvents which are good swelling agents, and hence give fastest extractions during Soxhlet extraction, tend to dissolve the polymer at the high temperatures used during ASE. Dissolved polymer re-precipitates on cooling and can block transfer lines in the instrument.Solvents therefore cannot be selected on the basis of those used for atmospheric pressure extractions. Hexane was used in the extractions even though it gives poor recoveries during Soxhlet extraction. Lou et al. point out that selection of a suitable extraction solvent is probably the most difficult step in optimising ASE, as there are few data on the solubility of polymers in solvents at high temperatures.These workers had previously analysed the same polymers using SFE with pure and modified CO2 64 and compared the result with that obtained using pure CO2 at 170 °C and 30.7 MPa. The recoveries for ASE for caprolactam from nylon and the dimer and trimer from PBT are 1.1, 6.5 and 37.6 times higher, respectively, than those obtained with SFE. However, at these conditions the SFE was not optimum, particularly for the dimer and trimer, where the peak extraction after 30 min occurred at 110 and 90 0C, respectively.This extraction peak at low temperatures clearly indicates solubility-limited extractions. Addition of modifier (methanol for nylon and chloroform for PBT) during the static extraction stage further increased recoveries from SFE, particularly for the dimer and trimer, but recoveries were still higher with ASE by approximately 1.5 times for the dimer and trimer. No experiments were performed with modified CO2 during the dynamic extraction. From these results, it appears that ASE offers significant advantages over SFE with CO2 alone for extraction of compounds with a low solubility in CO2. Other Extraction Methods Some new variations on the dissolution theme have recently been published.Staal et al.90 dissolved polycarbonate and polysulfone in THF and precipitated them onto a C18 guard column. A gradient elution from 50 + 50 water–THF as nonsolvent to 100% THF successively eluted the additives, then the oligomers and finally the polymer itself.No quantitative work was reported The method could be adapted by altering the solvent programme to separate the compounds of interest. Another way to separate the additives from the polymer after dissolution was explored by Nerin et al.91 They linked a highperformance size-exclusion chromatography (SEC) column with a normal-phase HPLC column, via a three-way switching 108R Analyst, September 1997, Vol. 122valve. The polymer elutes from the SEC column first and is drained through the valve.The valve is then switched to allow the additives onto the analytical HPLC column. One problem with the dissolution method is that high boiling solvents are usually required to dissolve the polymer. The solvent is therefore difficult to remove after precipitation of the polymer. This was addressed by Macko and co-workers92,93 who used an autoclave to dissolve HDPE in heptane at 160–170 °C, well above the normal boiling-point.The polymer was precipitated by cooling, and after filtration the additives could be determined by direct analysis of the resulting solution by normal-phase HPLC. Alternatively, the solvents were relatively easy to remove by evaporation. The dissolution took 1 h, and the complete analysis time was 3 h. Caceres et al.87 compared several methods for extraction of Tinuvin 770 and Chimassorb 944 from HDPE pellets. Room temperature diffusion into chloroform and ultrasonication gave less than 20% extraction.Soxtec extraction with DCM for 4 h resulted in only 50% extraction. Dissolution of the polymer in dichlorobenzene at 160 °C for 1 h followed by re-precipitation of the polymer with propan-2-ol gave 65–70% recovery. The most successful method was boiling under reflux with toluene at 160 °C for 2–4 h, which extracted 95% of both additives. The relatively poor performance of the Soxtec extraction compared with the reflux extraction is probably due to the large difference in temperature between the boiling solvents.The pellets were not ground and the size was not specified. Conclusions There are several methods that can be used for the extraction of low molecular weight material from polymers. The principal points of each are given in Table 2. Table 3 shows a summary of the extraction papers discussed here. There is a choice to be made between inexpensive, simple equipment giving long extraction times and more expensive but rapid techniques.Soxhlet extraction will generally extract all additives, but extraction times can be as long as 48 h. During Table 2 Summary of extraction techniques Soxhlet SFE Microwave Sonication ASE Solvent Any CO2 (possibly Must contain a Any Any modified) microwave absorbing component Typical sample size 1–5 g 10–100 mg 1–5 g 1–5 g 1–10 g Analysis time 6–24 h 20 min–2 h 30–60 min 40–60 min 15 min Solvent usage 50–100 ml 10 ml 30 ml 30–50 ml 30–50 ml Advantages Inexpensive, Low solvent Fast, low Economical Low solvent widely accepted use, fast, can be solvent use, can use automated, automated extract multiple any solvent solvents possible simultaneously Disadvantages Slow, large solvent Expense, may take Expense, not Not always effective Expense, not use time to optimise sufficient body sufficient body method of evidence for of evidence for extraction from extraction from plastics plastics Table 3 Polymer extraction summary Extraction Polymer Additive technique Comments Ref.Solid–liquid extraction— Poly (ethylene terephthalate) Cyclic trimer and other oligomers up to the heptamer Dissolution followed by precipitation 0.1 g of polymer was dissolved in dichloromethane and hexafluoropropan-2-ol (7 + 3 v/v 10 ml) and then acetone was added to precipitate the high molecular mass polymer. The sample was filtered, concentrated to dryness and the residue dissolved in dimethylacetamide. Positive ion atmospheric-pressure chemical ionisation (APCI) was used to analyse the extracts 15 Poly (ethylene terephthalate) Butyric acid (1), Malathion (2), Diazinon (3).All are recycling byproducts Dissolution/ precipitation (1) PET was dissolved in a large volume of hexafluoropropanol and dichloromethane. Large oligomers were then removed from the solution by polymer precipitation using acetone. Additives (2) and (3) as before but precipitated using methanol. Average recoveries from spiked samples were 80–98% 16 Poly (vinyl chloride) Dioctyl phthalate Dissolution/ precipitation Sample dissolved in THF and centrifuged at 20 000 rev min21.Polymer is then precipitated using ethanol and isolated by filtration. The filtrate is evaporated to dryness and analysed 94 Poly(vinyl chloride) and poly(vinyl chloride)–vinyl acetate copolymer Tinuvin 320; Cyasorb UV 9; Uvinul N- 539 Soxhlet extraction PVC film or finely ground PVC particles (1 g). Soxhlet extraction with diethyl ether for 16 h; evaporation to dryness; precipitate dissolved in THF (10 ml); filtration through Millipore Teflon filter of 0.5 mm pore size.Recoveries ranged from 65% (unground PVC, diameter 2–10 mm) to 94% (ground, i.e., < 0.5 mm) for Tinuvin 320; 59% (unground PVC, diameter 2–10 mm) to 97% (ground, i.e., < 0.5 mm) for Cyasorb UV-9; and 89% (film, thickness 150 mm) to 95% (ground, i.e., < 0.5 mm) for Uvinul N-539 95 Table continued on next page Analyst, September 1997, Vol. 122 109RTable 3 continued Extraction Polymer Additive technique Comments Ref. Poly(propylene) Chimassorb 944 Dissolution/precipitation Polymer was dissolved in decalin at 150 °C and re-precipitated by cooling 13 Polyamides Lubricants, e.g., stearic acid and ethylene bis-stearamide Dissolution followed by liquid– liquid extraction Dissolve 5 g of polymer in 30 ml of formic acid and extract with 150 ml of toluene for 18 h 14 Polycarbonate, polysulfone Additives, monomers and oligomers Dissolution THF dissolution of polymer samples. Preliminary note on the on-line extraction of polymers by multiple solvents on packed HPLC columns. 90 Poly(ethylene) Cyasorb 531 Soxhlet extraction Powdered sample (100 g) Soxhlet-extracted with 500 ml hexane or chloroform for 12 h. This was repeated three times. Then the extract was filtered and filtrate evaporated to dryness. Soluble fractions of low molecular weight polymers were removed from the individual extracts using methanol followed by evaporation to dryness and dissolution in chloroform.Chromatographic identification only 96 Poly(ethylene) Irganox 1010, Irgafos 168, a-tocopherol Disolution with hot heptane under pressure Polymer (1 g, cut into slices) was dissolved in heptane at 160–170 °C under elevated pressure in an autoclave. Polymer precipitated by cooling and supernatant analysed by HPLC. 2 h per sample needed for complete analysis. 92, 93 Poly(ethylene) (high-density) BHT, Irganox anti-oxidants, Isonox, Cyasorb, Am 340, MD 1024, Irgafos 168 Ultrasonic extraction Use of cyclohexane–dichloromethane as extraction solvent reduces risk of hydrolysis of Irgafos 168. 30–40 min of ultrasonic extraction needed for complete extraction. 86 Poly(ethylene) (high-density) Tinuvin 770, Chimassorb 944 Soxtec, ultrasonic extraction, room temperature diffusion, dissolution, reflux Boiling HDPE pellets under reflux with toluene was the most successful method 87 Poly(ethylene) (low-density) Ionol, Santanox, oleamide Flask-shaker Pelletised (7-mesh) LDPE (5 g) was shaken with solvents (10 ml).Carbon disulfide extracted Ionol and Santonox in about 2 h, much faster than isooctane. Extraction with powdered (50-mesh) LDPE was much faster 17 Poly(ethylene) (low-density) Chimassorb 81 Ultrasonic extraction and maceration Ground polymer (0.25 g) was placed in dichloromethane (3 ml), shaken and kept in the dark for 6 h, followed by 3 3 15 min ultrasonic extractions.This gave same extraction as 48 h of standing in the dark 88 Poly(ethylene) (low-density) DSTDP; Irganox 1035; Santonox R; peroxide initiator Vulcup Soxhlet extraction Poly(ethylene) film (5 g) Soxhlet-extracted with 100 ml THF, extract concentration to 5 ml. Chromatographic identification only 97 Polymer DLTDP; DSTDP; TNPP; Goodrite 3114; Weston 618; Topanol CA, Irganox 1076; Cyasorb UV 531 Solid–liquid extraction from polymers by soaking in boiling solvents Reflux pressed foil samples with boiling CH2Cl2 for 2 h; evaporate to dryness; solution in 5 ml THF; filtration through 0.5 mm Millipore Teflon filter 20 Poly(propylene) Irganox 1010, Irganox 1330, Irgafos 168, Irganox 3114, Atmer 163 and Tinuvin 326 Soxhlet extraction Soxhlet extraction using chloroform (50 ml).Optimum extraction obtained in 1 h using sieved ( < 1.18 mm) freeze-ground samples (3 g). For film and granular samples, 3 h extraction necessary 14 Poly(propylene) Tinuvin 770; Hostavin TMN 20; Tinuvin 144 Soxhlet extraction Sample (10 g) in powder or pellet form was extracted.Kumagawa extraction with CHCl3 for 16 h, extract concentration to 20 ml under a flow of nitrogen; then 80 ml of acetone added to precipitate oligomers. Sample filtered and washed with hot acetone; filtrate concentrated under a flow of nitrogen and finally made up to volume (10 ml) with CHCl3. Results for poly(propylene) pellets were 96.2% for Tinuvin 770 and 95.6% for Hostavin TMN 20 95 Poly(propylene) DLTDP; DSTDP; TNPP; BHT; Goodrite 3114; Weston 618; Topanol CA; Irganox 1076; Cyasorb UV 531; oleamide; erucamide; Ethyl 330; stearamide; Irganox 1010 Soxhlet extraction Sample (50 g) in pellet form.Soxhlet extraction with 250 ml CH2Cl2 for 48 h; evaporate to dryness; re-dissolving in 5 ml THF; filtration through Millipore Teflon filter of 0.5 mm pore size. Good results (88–120% recovery) reported for two out of three samples 20 Poly(propylene) Irganox 1010; Irgafos 168; Tinuvin 770; erucamide; Irganox 3114; Tinuvin 440; Tinuvin P Soxhlet extraction Sample (10 g) in pellet form.Soxhlet extraction with diethyl ether for 15 h; evaporation to dryness; wax precipitation by refluxing with 5 ml ethanol; cooling; filtration. Qualitative data only 98 Table continued on next page 110R Analyst, September 1997, Vol. 122Table 3 continued Extraction Polymer Additive technique Comments Ref. Poly(propylene) BHT, Topanol CA, Irganox 1076 Cold liquid solvent extraction Sample of beads or shavings (1 g).Overnight extraction with 5 ml of acetonitrile at ambient temperature in a sealed amber-coloured vial with constant stirring. Qualitative only 99 Poly(propylene) DSTDP, BHT, Topanol CA, Santowhite powder Cold liquid solvent extraction Sample (4 g of 8-mesh pellets) shaken in a 50 ml screw-capped, darkened glass vial with (a) 20 ml THF or (b) CH2Cl2 for 24 h at room temperature 22 Poly(propylene) copolymer Irganox 1330 Solid–liquid extraction from polymers by soaking in boiling solvents Sample frozen and pulverised (0.5–5.0 g).Refluxing for 40 min under nitrogen purging using 25–100 ml of decalin, hexane, CHCl3 and THF. After cooling, samples filtered through Whatman GF/A microfibre filter. For quantitative work a 100-fold molar excess of BHT (0.5–1.5 mg l21) was added to the extraction solvent to protect Irganox 1330 from oxidation during extraction. Results indicate that all solvents gave good recoveries provided that the sample was milled to a particle size of < 1.0 mm 100 Poly(styrene) (styrenic polymer) External lubricants Dissolution followed by evaporation Wash 20 g of poly(styrene) resin with ethanol.Collect washings and evaporate to dryness with nitrogen. Dissolve residue in chloroform. Analyse using FTIR 101 Poly(styrene) (styrenic polymer) Internal lubricants Dissolution followed by evaporation Dissolve 20 g of resin in 150 ml of dichloromethane.Add 100 ml of ethanol dropwise to precipitate polymer. Analyse using FTIR 101 PVC, PE BHT, tinuvin 326, Tinuvin 327, Irganox 1076, Cyasorb UV 9, Cyasorb UV 1084 Dissolution, separation of polymer by size-exclusion chromatography Polymer dissolved and separated by SEC. Polymeric fraction diverted to waste and fraction containing analytes directed to silica HPLC column. Qualitative data only 88 Styrene–butadiene rubber Tris(nonylated phenyl) phosphite Ultrasonic extraction Ultrasonic extraction with isooctane for 2 3 20 min gave same extraction as 2 3 1 h for boiling under reflux 85 Supercritical fluid extraction— Biodegradable lactide- co-glycoside Oligomers, stearic acid, anthraquinone- based dye SFE Screening method for in vitro cytotoxicity testing.SFE extracts not contaminated with solvent residues 77 Biopol Triacetin, total extractables SFE 40 °C, 35 MPa, 10 min; CO2 only. Pieces were cut from a bottle. Faster flow rates gave faster extraction with almost complete recovery after 8 min at a flow rate of 8.5 ml min21 69 Nylon Caprolactam SFE SFE–SFC with CO2 only on milled polymer at 35 MPa and 75 °C showed extractable caprolactam.Addition of methanol modifier to the extraction cell significantly increased the amount of extracted caprolactam. The methanol was thought to act by swelling the polymer rather than through enhanced solubility. 65 Nylon 6 Caprolactam and oligomers SFE SFE conditions: temperature, 60–80 °C; CO2 density, 0.85 g ml21.Extraction effeciencies higher than methanol Soxhlet extractions. Use of 7.5% methanol modifier and additional modifier in extraction cell necessary for high extraction efficiencies 70 Nylon 6 Caprolactam SFE and Soxhlet extraction Study of modifier addition and temperature variation. Conditions for SFE: modified supercritical CO2; temperature, 50–170 °C; pressure, 30 MPa; extraction time, 20 min static and 30 min dynamic.Comparable results obtained by both techniques. Chloroform, benzene and methanol as modifiers had a large effect at low temperatures, the effect decreased at higher temperatures 64 Nylon pellets Cyclic trimer and ethyl bis-stearamide SFE On-line SFE–SFC. Conditions for SFE: CO2 only; temperature, 70 °C; pressure, 5–40 MPa sample size: 5–10 mg. Qualitative only 32 Poly(butylene terephthalate) Dimer/trimer SFE and Soxhlet extraction Study of modifier addition and temperature variation.Conditions for SFE: chloroform-modified supercritical CO2; temperature, 50 °C; pressure, 30 MPa; extraction time, 20 min static and 30 min dynamic. Comparable results obtained for the dimer extraction; lower recovery by SFE for the trimer (73%). Chloroform was the most effective modifier, followed by benzene and methanol 64 Poly(butylene terephthalate) polymers Volatiles SFE On-line–GC. Conditions for SFE: CO2 only; temperature, 55 °C; pressure, 20 MPa; extraction time, 10 min.Average recovery from a 25 mm sample thickness is 98% 52 Poly(ethylene terephthalate) Cyclic trimer SFE and Soxhlet extraction On- and off-line SFE–SFC. SFE conditions: CO2 only; temperature, 70 °C; pressure, 40 MPa; extraction time, 13 consecutive 30 min extractions, i.e., over a period of 6.5 h. Poor recovery obtained, as compared with Soxhlet, by SFE 35 Poly(ethylene terephthalate) Cyclic trimer SFE and liquid solvent extraction CO2 only; temperature, 90–215 °C; pressure, 40 MPa; extraction time, 30 min.Sample size: 0.02–3 g of film. Soxhlet conditions: Sample size, 12–13 g of film; solvent, xylene; extraction time, 24 h. Good results obtained by SFE at elevated temperature 51 Poly(ethylene terephthalate) Cyclic trimer SFE On-line SFE–SFC: conditions for SFE: CO2 only; temperature, 70 °C; pressure, 5–40 MPa. Sample size: 5–10 mg. Qualitative only Table continued on next page Analyst, September 1997, Vol. 122 111RTable 3 continued Extraction Polymer Additive technique Comments Ref.Poly(ethylene terephthalate) Cyclic trimer and other oligomers SFE Conditions: supercritical CO2 with and without modifiers used (methanol, isopropyl alcohol, dichloromethane and acetone); temperature, 40–150 °C; density, 0.5–0.9 g ml21. Sample size: 0.75 g of ground PET chips or 1 g of PET fibres. The use of isopropyl alcohol and methanol as modifiers prevented extraction; a 7% dichloromethane modified supercritical CO2 was beneficial.A three-stage SFE procedure was recommended using supercritical CO2 only 50 Poly(vinyl chloride) Diisooctyl phthalate (DIOP) as plasticizer, chlorinated polyethylene wax, Topanol SFE and liquid solvent extraction Conditions for SFE: CO2 only; temperature, 45–115 °C; pressure, 7–45 MPa; extraction time, 0.5–435 min. Sample size: 0.2–0.4 g. Analysis using off-line packed SFC. Results for SFE compare favourably with those obtained using liquid extraction and the actual formulation value 53 Poly(vinyl chloride) Stabilizers SFE and dissolution/ precipitation Screening method for food contact plastics, non-quantitative. SFE extracted same stabilizers as precipitation method 5 Poly(vinyl chloride) Tinuvin P SFE Polymer was dissolved in THF and internal standard added. The polymer was re-cast and extracted with CO2 at 50 °C and a pressure of 30 MPa for 10 min.Extraction was far from complete, but the internal standard compensated and good analyses were obtained 75 Poly(vinyl chloride) Dibutyl phthalate, dioctyl phthalate SFE and Soxhlet extraction Best SFE conditions: temperature,95 °C; pressure, 48 MPa; extraction time, 25 min; unmodified CO2. Extraction efficiency was 98% compared with 5 h Soxhlet extraction with cyclohexane 54 Poly(vinyl chloride) Organotin stabiliser SFE SFE conditions: pressure, 18 MPa; formic acid modifier; temperature optimum at 90 °C.Extraction complete within 60 min 60 Poly(vinyl chloride) Diisooctyl phthalate, diethylhexyl phthalate, BHT, Tinuvin P, tributyltin chloride, vintyl chloride monomer SFE Screening method for in vitro cytotoxicity testing; therefore, not quantitative. SFE conditions: pressure, 42 MPa; temperature, 50–150 °C.Main advantage over solvent extraction is that the extracts are not contaminated with toxic solvent residues 77 Poly(alkylene glycol (PAG) Additives SFE SFE conditions: temperature, 45 °C; pressure, 30 MPa; CO2 only.PAG was adsorbed into silica and extracted with CO2. An in-line silica column was used to remove co-extracted oligomers 76 Poly(isoprene) Carbon tetrachloride SFE Optimum extraction conditions 60 °C at 21 MPa, CO2 only. Increasing the temperature caused softening and agglomeration of rubber particles. Increasing the pressure lowered the softening temperature of the rubber and therefore also caused the particles to agglomerate 66 Poly(styrene) Ethylbenzene SFE CO2 only, temperature from 50 to 100 °C, pressure: 7–12 MPa.Supercritical CO2 swelled the polymer, lowered Tg and resulted in a 106-fold increase in diffusivity of ethylbenzene in PS. A model used to estimate diffusivity in the swelled polymer is presented 61 Poly(styrene) Alkylbenzenes SFE CO2 only. SFE was limited by diffusion within the polymer rather than solubility in the CO2. Fastest extractions therefore employ small particle size and either static or dynamic extraction 68 Poly(styrene) Styrene dimers and trimers SFE CO2 only. 9–11 MPa, 80–120 °C. On-line SFE–SFC was effective at identifying additives and oligomers. The rate of extraction increased with temperature 65 Poly(urethane) foams CFCs SFE SFE conditions: 50 °C at 11 MPa. Supercritical CO2 extracted much more than liquid CO2 and nitrogen. Addition of propane modifier seems to increase extraction rate slightly. Ground foam extracted much faster than un-ground foam 72 Poly(ethylene) Erucamide acid amide SFE On-line SFE-capillary SFC.CO2 only; temperature, 45 °C; pressure, 15 MPa; extraction time, 15 min. Sample size: 2.7 mg. Example chromatogram presented 58 Poly(ethylene) Irganox SFE On-line SFE–capillary SFC. CO2 only; temperature, 45 °C; pressure, 20 MPa; extraction time, 30 min. Sample size: 9 mg. Example chromatogram presented 58 Poly(ethylene) Chimassorb, Tinuvin 144, Irganox 1098, oxidised Nauguard 524, Irganox 1010, DSTDP, Irganox 1076 and Ionox 330 SFE and Soxhlet extraction SFE conditions: CO2 only; temperature, 40 °C; pressure, 60 MPa; extraction time, 30 min and 5 h for low- and high-density poly- (ethylene) samples, respectively.Sample size: 0.2 g. Soxhlet: 1 g sample extracted with 100 ml of toluene for 3 h. After cooling, 20 ml of ethanol added to precipitate low molecular weight polymer. Extract solutions filtered, dried and reconstituted in CH2Cl2 prior to analysis. Similar mass spectra obtained by both extraction techniques 67 Poly(ethylene) Additives SFE Conditions for SFE: CO2 only; temperature, 65 °C; pressure, 15 MPa; extraction time, 10 min.Sample size: 3 mg of film 102 Poly(ethylene) Irganox 1010, Irganox 1076, Irgafos 168 SFE and Soxhlet extraction Conditions for SFE: temperature, 50–80 °C; pressure, 15–30 MPa, CO2 and benzene-modified CO2. Results at 80 °C and 30 MPa generally fastest, with amount extracted sometimes exceeding that from Soxhlet extraction. Benzene modifier increased extraction rates, particularly at lower temperatures 63 Poly(ethylene) Nicotine SFE SFE conditions: pressure, 21 MPa; temperature, 60 °C; CO2 modified with triethylamine; extraction time, 20 min 73 Table continued on next page 112R Analyst, September 1997, Vol. 122Table 3 continued Extraction Polymer Additive technique Comments Ref. Poly(ethylene) (low-density) SFE On-line SFE–SFC analysis. SFE conditions: temperature, 100 °C; pressure, 45 MPa. Typical extraction time 20 min for ground polymer samples (30–50 mesh) 74 Poly(ethylene) (low-density) Paraffins and olefins SFE Screening method for in vitro cytotoxicity testing 77 Poly(ethylene) and poly(propylene) Irganox 1010, BHT, erucamide, Tinuvin 770, Irgafos 168, Isonox 129 and DLTDP SFE On-line SFE–SFC.SFE conditions: CO2 only; temperature, 50 °C; pressure, 14 and 43 MPa; extraction time, 15 and 30 min. Accurate quantification and high extraction efficiency reported. 33 Poly(ethylene), poly(styrene) BHT, BHEB, Isonox 129, Irganox 1076, Irganox 1010, Irgafos 168, Cyasorb 3346, Cyanox 1790, stearyl stearamide, HBCD and erucamide SFE On-line SFE–SFE.CO2 only; temperature, 150 °C; pressure, 46 MPa; extraction time, 30 min. Typical recoveries ranged from 86 to 108% 55 Polymer Misoprostol SFE Formic acid modified CO2 used to extract misoprostol steroid drug from polymer to which it is covalently bonded. Formic acid hydrolyses covalent bond 62 Poly(propylene) BHT, Tinuvin 326, Seenox DM and Irganox 1010 SFE and liquid solvent extraction On-line SFE–capillary SFC.CO2 only; temperature, 30–90 °C; pressure, 30 MPa; extraction time, 30 min. Qualitative results only. Selectivity of extraction investigated 56 Poly(propylene) Apple aroma compounds SFE and liquid solvent extraction Conditions for SFE: CO2 only; temperature, 10–70 °C; pressure, 6–12 MPa; extraction time, 5–20 min. Sample size: 10 mg of film. Analysis using on-line GC. Results for SFE compare favourably (96–105% recovery) with those obtained using liquid solvent (dichloromethane) extraction 94 Poly(propylene) Irganox 1010, Irganox 1076, total extractables SFE 40 °C, 35 MPa, 10 min, CO2 only, 30 mm film.SFE extracted more material than 4 h Soxhlet extraction with dichloromethane 69 Poly(propylene) Irganox 1076, Irgafos 168, Hostanox SE- 02 SFE SFE conditions: temperature, 60 °C; CO2 density, 0.85 g ml21. Optimum extraction used two static extractions with hexane and methanol modifiers 71 Poly(propylene), poly(ethylene) Cyasorb UV 531, Topanol OC, Irganox 1330, Irganox 1010, Irganox 1076 SFE SFE conditions: 25.6 MPa, 35 °C, CO2 only. 2 h extraction was required for complete extraction 31 Poly(urethane) foams Amgard TCEP, Amgard TMCP, Amgard V6 and Thermolin 101 SFE On-line SFE–SFC. SFE conditions: CO2 only; temperature, 60 °C; pressure, 30 MPa; extraction time, 5 min. Good recoveries obtained compared with solvent extraction 75 Poly(urethane) Oligomers, TPP, BHT, Irganox 1010, Irganox 1076 SFE Screening method for in vitro cytotoxicity testing 77 Poly(urethanes) [Pellethane, estane, and 90-N poly(urethane) Ellastolan] BHT, residual ethers, plasticisers (either an adipate acid or a phthalate ester) SFE On-line SFE–SFC–MS.Conditions for SFE: CO2 only; temperature, 60 °C; pressure, 30 MPa; extraction time, 10 min. Feasibility study 78 Microwave-assisted extraction (MAE) and ASE— Poly(propylene) and poly(ethylene) (powdered) Irganox 1010, Irgafos 168, Chimassorb 81 MAE Hexane–acetone (1 + 1)—max. 6 min. 1,1,1-Trichloroethane—max. 3 min. Microwave power, 70%; solvent volume, 30 ml for both 82 HDPE, LDPE, PP Irganox 1010, Irgafox 168, Cyasorb UV 531, BHT Microwave heating, atmospheric pressure Cyclohexane–propan-2-ol (1 + 1) found to be the best solvent for extraction from poly(olefins). 20 min extraction time typical 83 Poly(ethylene terephthalate) Cyclic trimer MAE and Soxhlet extraction MAE conditions: sample size, 8 g of film; solvent, xylene + water or dichloromethane (40 ml); extraction time, 30–120 min; temperature, 70–120 °C.Soxhlet conditions: sample size, 15–20 g of film; solvent, xylene (190 ml); extraction time, 24 h. Good agreement between both extraction methods 84 PBT Dimers and trimers ASE Freeze-ground samples extracted with hexane at temperature from 50 to 70 °C. Extraction rate increased with temperature. Pressure had no effect other than to keep solvent liquid.Extraction faster than SFE with CO2 alone 89 Nylon-6 Caprolactam ASE Freeze-ground samples extracted with hexane at temperatures from 50 to 170 °C. Extraction rate increased with temperature. Pressure had no effect other than to keep solvent liquid. Extraction faster than with SFE with CO2 alone, but very similar to CO2 with methanol modifier 89 Analyst, September 1997, Vol. 122 113Rsuch long extractions and subsequent concentration steps there is a possibility of losses of volatile or thermally labile components.The selection of extraction solvent can make a large difference to the extraction time. However, the equipment is inexpensive, and once set up requires little ‘hands-on’ attention. SFE has been shown to provide a much faster extraction method than Soxhlet extraction. In most cases rapid and quantitative extraction has been achieved using ground samples or thin films. The method needs to be optimised for pressure, temperature and modifier.Equipment exists which can extract several samples simultaneously and which can be programmed to extract up to 28 samples consecutively. Microwave-assisted extraction offers a rapid method for the extraction of up to 12 samples simultaneously. The equipment is also much more expensive than that for Soxhlet extraction, and there are relatively few publications concerning this method. ASE has been found to be effective in extraction from environmental samples and has great potential for extraction from polymers.The main problem is likely to be selection of extraction solvent which does not dissolve the polymer at high temperature. Although ASE uses liquid solvents, the total solvent usage may not be higher than through the use of modifiers with SFE. In conclusion, Soxhlet extraction is inexpensive and well established, but time consuming. Of the newer methods, only SFE has been available long enough for large amounts of published evidence to become available.Amongst the applications of analytical SFE, extractions of polymers have been the most successful, although a few cases are problematical. Other newer methods show promise, but cannot yet be assessed from a wide range of published data. References 1 EEC 89/109, Off. J. Eur. Comm., 1989, L40, 38. 2 EEC 90/128, Off. J. Eur. Comm., 1990, L349, 26. 3 EEC 92/39, Off. J. Eur. Comm., 1992 , L168, 21. 4 Schilling, F. C., and Kuck, V. J., Polym. Degradation Stab., 1991, 31, 141. 5 Brauer, B., Funke, T., and Schulenbergschell, H., Dtsch. Lebensm. Rundsch., 1995, 91, 381. 6 Wright, S. J., Dale, M. J., Langridge-Smith, P. R. R., Zhan, Q., and Zenobi, R., Anal. Chem., 1966, 68, 3585. 7 Crompton, T. R., The Analysis of Plastics, Pergamon Press, Oxford, 1984. 8 Haslam, J., Willis, H. A., and Squirrel, D. C. M., Identification and Analysis of Plastics, Iliffe Books, London, 2nd edn., 1977. 9 Wheeler, D. A., Talanta, 1968, 15, 1315. 10 Cotton, N.J., PhD Thesis, University of Leeds, 1992 11 British Standard 2782, Part 4, Method 405D, 1965. 12 Schabron, J. F., and Fenska, L. E., Anal. Chem., 1980, 52, 1411. 13 Freitag, W., Fresenius’ Z. Anal. Chem., 1983, 316, 495. 14 Newton, I. D., in Polymer Characterisation, ed. Hunt, B. J., and James, M. I., Blackie, Glasgow, 1993, pp. 8–36. 15 Barns, K. A., Damant, A. P., Startin, J. R., and Castle, L., J. Chromatogr. A, 1995, 712, 191. 16 Komolprasert, V., Lawson, A. R., and Hargraves, W.A., J. Agric. Food Chem., 1995, 43, 1963. 17 Spell, H. L., and Eddy, R. D., Anal. Chem., 1960, 32, 1811. 18 Crompton, T. R., Eur. Polym. J., 1968, 4, 473. 19 Majors, R. E., J. Chromatog. Sci., 1970, 8, 339. 20 Haney, M. A., and Dark, W. A., J. Chromatogr. Sci., 1980, 18, 655. 21 Perlstein, P., Anal. Chim. Acta, 1983, 149, 21 22 Wims, A. M., and Swarin, S. J., J. Appl. Polym. Sci., 1975, 19, 1243. 23 Majors, R. E., LC–GC, 1995, 13, 82. 24 Majors, R. E., LC–GC, 1996, 14, 88. 25 Chester, T. L., Pinkston, J. D., and Raynie, D. E., Anal. Chem., 1992, 64, 153R. 26 Chester, T. L., Pinkston, J. D., and Raynie, D. E., Anal. Chem., 1994, 66, 106R. 27 Levy, J. M., HRC–J. High Resolut. Chromatogr., 1994, 17, 212. 28 Camel, V., Tambute, A., and Caude, M., Analusis, 1992, 20, 503. 29 Smith, C. G., Smith, P. B., Pasztor, A. J., Mckelvy, M. L., Meunier, D. M., Froelicher, S. W., and Ellaboudy, A. S., Anal. Chem., 1993, 65, R 217. 30 Anton, K., Menes, R., and Widmer, H.M., Chromatographia, 1988, 26, 221. 31 Hirata, Y., and Okamoto, Y., J. Microcol. Sep., 1989, 1, 46. 32 Cotton, N. J., Bartle, K. D., Clifford, A. A., Ashraf, S., Moulder, R., and Dowle, C. J., HRC–J. High Resolut. Chromatogr., 1991, 14, 165. 33 Ryan, T. W., Yocklovich, S. G., Watkins, J. C., and Levy, E. J., J. Chromatogr., 1990, 505, 273. 34 Bartle, K. D., Clifford, A. A., Hawthorne, S. B., Lagenfield, J. J., Miller, D. J., and Robinson, R., J. Supercrit. Fluids, 1990, 3, 143. 35 Bartle, K. D., Boddington, T., Clifford, A. A., Cotton, N. J., and Dowle, C. J., Anal. Chem., 1991, 63, 2371. 36 Ashby, R., Food Addit. Contam., 1988, 5, 485. 37 Bartle, K. D., Boddington, T., Clifford, A. A., and Hawthorne, S. B., J. Supercrit. Fluids, 1992, 5, 207. 38 Clifford, A. A., Bartle, K. D., and Zhu, S. A., Anal. Proc., 1995, 32, 227. 39 Howdle, S. M., Ramsay, J. M., and Cooper, A. I., J. Polym. Sci. B: Polym. Phys., 1994, 32, 541. 40 Fahmy, T. M., Paulitis, M.E., Johnson, D. M., and McNally, M. E. P., Anal. Chem., 1993, 65, 1462. 41 Shieh, Y. T., Su, J. H., Manivannan, G., Lee, P. H. C., Sawan, S. P., and Spall, W. D., J. Appl. Polym. Sci., 1996, 59, 695. 42 Shieh, Y. T., Su, J. H., Manivannan, G., Lee, P. H. C., Sawan, S. P., and Spall, W. D., J. Appl. Polym. Sci., 1996, 59, 707. 43 Chiou, J. S., Barlow, J. W., and Paul, D. R., J. Appl. Polym. Sci., 1985, 30, 3911. 44 Michaels, A. S., and Bixler, H. J., J. Polym. Sci., 1961, L, 393. 45 Kazarian, S. G., Vincent, M. F., Bright, F. V., Liotta, C. L., and Eckert, C. A., J. Am. Chem. Soc., 1996, 118, 1729. 46 Kalospiros, N. S., and Paulaitis, M. E., Chem. Eng. Sci., 1994, 49, 659. 47 Handa, Y. P., Roovers, J., and Wang, F., Macromolecules, 1994, 27, 5511. 48 Condo, P. D., and Johnston, K. P., J. Polym. Sci. B: Polym. Phys., 1994, 32, 523. 49 Condo, P. D., Paul, D. R., and Johnston, K. P., Macromolecules, 1994, 27, 365. 50 Kuppers, S., Chromatographia, 1992, 33, 434. 51 Cotton, N. J., Bartle, K. D., Clifford, A. A., and Dowle, C. J., J. Chromatogr. Sci., 1993, 31, 157. 52 Schmidt, S., Blomberg, L., and Wannnman, T., Chromatographia, 1989, 28, 400. 53 Hunt, T. P., and Dowle, C. J., Analyst, 1991, 116, 1299. 54 Marin, M. L., Jiminez, A., Lopez, J., and Vilaplana, J., J. Chromatogr. A, 1996, 750, 183. 55 Ashraf-Khorassani, M., Boyer, D. S., and Levy, J. M., J. Chromatogr. Sci., 1991, 29, 517. 56 Daimon, H., and Hirata, Y., Chromatographia, 1991, 32, 549. 57 Mackay, G. A., and Smith, R. M., Analyst, 1993, 118, 741. 58 Engelhardt, H., Zapp, J., and Kolla, P., Chromatographia, 1991, 32, 527. 59 Cotton, N. J., Bartle, K. D., Clifford, A. A., and Dowle, C. J., J. Appl. Polym. Sci., 1993, 48, 1607. 60 Oudsema, J. W., and Poole, C. F., HRC–J. High Resolut. Chromatogr., 1993, 16, 198. 61 Dooley, K. M., Launey, D., Becnel, J. M., and Caines, T., ACS Symp. Ser., 1995, 608, 269. 62 Roston, D. A., Sun, J. J., Collins, P.W., Perkins, W. E., and Tremont, S. J., J. Pharm. Biomed. Anal., 1995, 13, 1513. 63 Lou, X. W., Janssen, H. G., and Cramers, C. A., J. Microcol. Sep., 1995, 7, 303. 64 Lou, X. W., Janssen, H. G., and Cramers, C. A., J. Chromatogr. Sci., 1996, 34, 282. 65 Jordan, S. L., Taylor, L. T., Seemuth, P. D., and Miller, R. J., Text. Chem. Color., 1997, 29, 25. 66 Burgess, A. N., and Jackson, K., J. Appl. Polym. Sci., 1992, 46, 1395. 67 Juo, C. G., Chen, S. W., and Her, G. R., Anal. Chim.Acta, 1995, 311, 153. 68 Hawthorne, S. B., Galy, A. B., Schmitt, V. O., and Miller, D. J., Anal. Chem., 1995, 67, 2723. 114R Analyst, September 1997, Vol. 12269 Baner, L., Bucherl, T., Ewender, J., and Franz, R., J. Supercrit. Fluids, 1992, 5, 213. 70 Venema, A., Vandeven, H. J. F. M., David, F., and Sandra, P., HRC– J. High Resolut. Chromatogr., 1993, 16, 522. 71 Garde, J. A., Galotto, J., Catala, R., and Gavara, R., presented at the ILSI Symposium on Food Packaging: Ensuring the Quality and Safety of Foods, September 11–13, 1996, Budapest, Hungary. 72 Filardo, G., Galia, A., Gambino, S., Silvestri, G., and Poidomani, M., J. Supercrit. Fluids, 1996, 9, 234. 73 Cedergren, L., Fors, S., and Jonn, S., presented at the 3rd European Symposium on Analytical SFC and SFE organized in conjunction with the 6th International Symposium on SFC and SFE, September 6–8, 1995, Uppsala, Sweden. 74 Ashraf-Khorassani, M., and Levy, J. M., HRC–J. High Resolut. Chromatogr., 1990, 13, 742. 75 Mackay, G. A., and Smith, R. M., HRC–J. High Resolut. Chromatogr., 1995, 18, 607. 76 Hunt, T. P., Dowle, C. J., and Greenway, G., Analyst, 1993, 118, 17. 77 Braybrook, J. H., and Mackay, G. A., Polym. Int., 1992, 27, 157. 78 Mackay, G. A., and Smith, R. M., J. Chromatogr. Sci., 1994, 32, 455. 79 Zlotorzynski, A., Crit. Rev. Anal. Chem., 1995, 25, 43. 80 Smith, F. E., and Arsenault, E. A., Talanta, 1996, 43, 1207. 81 Renoe, B. W., Am. Lab., 1994, August, 34. 82 Freitag, W., and John, O., Angew. Makromol. Chem., 1990, 175, 181. 83 Nielson, R. C., J. Liq. Chromatogr., 1991, 14, 503. 84 Costley, C. T., Dean, J. R., Newton, I., and Carroll, J., Anal. Commun., 1997, 34, 89. 85 Brandt, H. J., Anal. Chem., 1961, 33, 1390. 86 Nielson, R. C., J. Liq. Chromatogr., 1993, 16, 1625. 87 Caceres, A., Ysambert, F., Lopez, J., and Marquez, N., Sep. Sci. Technol., 1996, 31, 2287. 88 Nerin, C., Salafranca, J., and Cacho, J., Food Addit. Contam., 1996, 13, 243. 89 Lou, X., Janssen, H., and Cramers, C. A., Anal. Chem., 1997, 69, 1598. 90 Staal, W. J., Cools, P., Vanherk, A. M., and German, A. L., Chromatographia, 1993, 37, 218. 91 Nerin, C., Salafranca, J., Cacho, J., and Rubio, C., J. Chromatogr. A, 1995, 690, 230. 92 Macko, T., Siegl, R., and Lederer, K., Angew. Makromol. Chem., 1995, 227, 179. 93 Macko, T., Furtner, B., and Lederer, K., J. Appl. Polym. Sci., 1996, 62, 2201. 94 Nielson, T. J., Jagerstad, I. M., Oste, R. E., and Sivik, B.T. G., J. Agric. Food Chem., 1991, 39, 1234. 95 Sevini, F., and Marcato, B., J. Chromatogr. A, 1983, 260, 507. 96 Lehotay, J., Danecek, J., Liska, O., Lesko, O. J., and Brandsteterova, E., J. Appl. Polym. Sci., 1980, 25, 1943. 97 Majors, R. E., and Johnson, E. L., J. Chromatogr. Sci., 1978, 167, 17. 98 Raynor, M. W., Bartle, K. D., Davies, I. L., Williams, A., Clifford, A. A., Chalmers, J. M., and Cook, B. W., Anal. Chem., 1988, 60, 427. 99 Vargo, J. D., and Olson, K.L., Anal. Chem., 1985, 57, 672. 100 Gasslander, U., and Jaegfeldt, H., Anal. Chim. Acta, 1984, 166, 243. 101 Kumur, T., Analyst, 1990, 115, 1319. 102 Hirata, Y., Nakata, F., and Horihata, M., J. High Resolut. Chromatogr., 1988, 11, 81. Paper 7/04052K Received June 10, 1997 Accepted July 7, 1997 Analyst, September 1997, Vol. 122 115R Critical Review Analytical Extraction of Additives From Polymers Harold J. Vandenburga, Anthony A. Clifford*a, Keith D. Bartlea, John Carrollb, Ian Newtonb, Louise M.Gardenb, John R. Deanc and Claire T. Costleyc a School of Chemistry, University of Leeds, Leeds, UK LS2 9JT b ICI Technology, Research and Technology Centre, P.O. Box 90, Wilton, Middlesbrough, Cleveland, UK TS90 8JE c Department of Chemical and Life Sciences, University of Northumbria at Newcastle, Ellison Building, Newcastle-upon-Tyne, Tyne and Wear, UK NE1 8ST Summary of Contents Introduction Conventional Extraction Techniques Dissolution of the Polymer Liquid–Solid Extractions New Developments in Extraction From Polymers SFE MAE ASE Supercritical Fluid Extraction Extraction Process Factors Affecting SFE From Polymers Effect of Supercritical CO2 on Polymers Effect of Temperature and Pressure Effect of Flow Rate Effect of Modifiers Nature of the Extractant Effect of Particle Size Summary of SFE Microwave Heating Ultrasonic Extraction ASE Other Extraction Methods Conclusions References Keywords: Review; polymer; supercritical fluid extraction; microwave-assisted extraction; liquid extraction Introduction Plastics contain many small molecules as well as the polymer itself.These include additives to alter the polymer properties or prolong the life of the polymer, such as plasticisers, antioxidants and ultraviolet (UV) light absorbers. There may also be processing aids, residual monomers, low molecular weight oligomers and inadvertent contaminants present. It is important for the manufacturer and regulators to know the level of these materials in the polymer to ensure the product is fit for its intended purpose. Food contact plastics are regulated by maximum concentrations allowable in the plastic, which applies to residual monomers and processing aids as well as additives. 1–3 There are some methods for determining concentrations of additives without extraction from the polymer, such as nuclear magnetic resonance spectrometry,4 UV spectrometry, 5 and UV desorption–mass spectrometry.6 However, in order to determine the levels in the polymer it is usually necessary to extract the compounds from the plastic quantitatively before analysis.The trade names and chemical names of many commonly used additives are given in Table 1. Conventional Extraction Techniques These can be divided into two categories: dissolution of the polymer and liquid–solid extraction methods. Many examples can be found in texts such as Crompton,7 Haslam et al.8 and Wheeler.9 A review of extraction methods to 1992 is included in Cotton.10 Dissolution of the Polymer A British Standard method11 describes the dissolution of polymers in refluxing toluene, with re-precipitation of the polymer by addition of ethanol.Decalin heated to 110 °C has been used as a solvent for poly(ethylene) (PE)12 and at 150 °C for poly(propylene) (PP).13 The high molecular weight polymer precipitates out as the solution is cooled and the supernatant solution is filtered and analysed.Formic acid is used to dissolve poly(amides) before further fractionation, for example by liquid–liquid extraction with toluene for 18 h to remove lubricants.14 Poly(ethylene terephthalate) (PET) has been dissolved in hexafluoropropan-2-ol–dichloromethane mixtures and the polymer precipitated by addition of acetone or methanol.15,16 Analysis of poly(vinyl chloride) (PVC) has traditionally been by sequential diethyl ether and methanol extractions to remove selected components, then dissolution of the polymer in tetrahydrofuran (THF).Centrifugation at two speeds would then produce two further fractions. Methanol would then be added to precipitate the polymer, and the methanol–THF would be evaporated and examined for non-extracted material.7 This Harold Vandenburg graduated from Manchester Polytechnic where he studied Applied Chemistry parttime whilst working in quality control in the pharmaceutical/ personal care sector.After working for a year in Australia and USA, he started six years of research on migration from polymers into foods at high temperatures at the Procter Department of Food Science at the University of Leeds. He obtained his PhD during this time. Still at the University of Leeds, but now at the School of Chemistry, he is undertaking research comparing the effectiveness of sample preparation methods for polymers and environmental samples. Analyst, September 1997, Vol. 122 (101R–115R) 101Rprocess can be simplified by dissolving the polymer in THF, followed by centrifugation at 20 000 rev min21 to remove mostly inorganic fillers.Addition of ethanol precipitates the polymer and polymeric additives. Polymeric plasticiser can be re-extracted from the precipitated polymer.14 Dissolution and re-precipitation therefore provides an effective method of extraction. The advantage is that there is no possibility of some analyte remaining bound in the polymer network, although inclusion of the analyte in the re-precipitated polymer can occur.There is often a considerable amount of ‘waxes’ in solution, which may need to be removed before further analysis. Some workers have considered this too time consuming17 and prefer liquid–solid extractions. Liquid–Solid Extractions Here the analyte is extracted from the solid medium by a liquid, which is separated by physical means, such as filtration. There are many methods for carrying out these extractions including Soxhlet, sonication and shake-flask extractions.Spell and Eddy17 studied the extraction of additives from PP at room temperature and found that required extraction time varied linearly with polymer density and decreased with increasing particle size. They also found a large variation in extraction time for different solvents and additives. By powdering the polymer to 50 mesh size, 98% extraction of 2,6-di-tert-butyl-4-methylphenol (BHT) was achieved by shaking at room temperature for 30 min with carbon disulfide.To achieve the same recovery with isooctane required 125 min, and 2000 min were required to recover Santonox with isooctane. The importance of small particles is further demonstrated by Newton.14 Refluxing ground PP with chloroform for 1 h gives complete extraction. For films, 3 h are required and for unground granules 3 h are sufficient to provide an extract for identification purposes only. Ethoxylated tertiary amines can be extracted from PP by refluxing the ground material with 1,2-dichloroethane for 1 h.Refluxing the granules for 3 h gives only 85% extraction. Soxhlet extraction has often been used, with a variety of solvents. Whilst this method eventually gives good extraction efficiencies, the extraction rate is slow. Times for extraction typically vary from 6 h18 to 48 h.19,20 Even with long extraction times recovery is not always good. Perlstein21 obtained recoveries of only 59% for extraction of Tinuvin 320 from unground PVC after 16 h Soxhlet extraction with diethyl ether.However, recoveries rose to 97% from ground polymer. The choice of solvent is significant for the duration of the extraction. Wims and Swarin22 found that talc filled PP needed 72 h extraction with chloroform, but only 24 h with THF. Thus, small particles are often essential to complete the extraction in reasonable times, and the solvents must be carefully selected to swell the polymer.Solid–liquid extraction has been shown to be effective, but often very slow and requires large amounts of solvents. This results in dilute solutions requiring further concentration before analysis, which is time consuming and may result in the loss of volatile compounds. The use of large volumes of often toxic solvents is environmentally unsound, as well as expensive in purchase and disposal costs. Some effort needs to be made to select the most appropriate solvent for the extraction.Therefore, there are great savings to be made if the extraction time and the solvent usage can be reduced. New Developments in Extraction From Polymers The principal objectives of any technique to replace ‘traditional’ extraction methods is to complete the extraction in less time, using less solvent and also have the possibility of automating the process. Recent articles23,24 describe some new techniques, such as supercritical fluid extraction (SFE), microwave-assisted extraction (MAE) and accelerated solvent extraction (ASE), and some applications are summarized.Although the purpose of this review is not to describe the techniques in detail, a brief description is given. SFE SFE uses fluids above their critical temperature and pressure. These supercritical fluids have densities and diffusivities between those of liquids and gasses. The solvating power is Table 1 Commonly used additives in polymers Trade name Chemical name BHT 2,6-Di-tert-butyl-4-methylphenol Chimassorb 81 2-Hydroxy-4-n-octoxybenzophenone Chimassorb 944 Poly(N-1,1,3,3-tetramethylbutyl-NA,NB-di(2,2,6,6-tetramethylpiperidinyl)-NA,NB- melaminoditrimethylene Cyasorb UV 531 2-Hydroxy-4-n-octoxybenzophenone DSTDP Distearyl 3,3A-thiodipropionate Erucamide 13-cis-Docosenamide Ionox 330 1,3,5-Trimethyl-2,4,6-tris(3,5-di-tert-butyl-4-hydroxybenzyl)benzene Irgafos 168 Tris(2,4-di-tert-butylphenyl) phosphite Irganox 1010 Pentaerythrityl-tetrakis(3,5-di-tert-butyl-4-hydroxyphenyl) propionate Irganox 1076 Octadecyl-3,5-di-tert-butyl-R-hydroxyhydrocinnamate Irganox 1098 N,NA-Bis{[3-(3A,5A-di-tert-butyl-4A-hydroxyphenyl)-(propionyl)]}- hexamethylenediamine Irganox 3114 1,3,5-Tris(3,5-di-tert-butyl-4-hydroxybenzyl)s-triazine 2,4,6-(1H, 3H,5H)trione Isonox 129 2,2A-Ethylidenebis(4,6-di-tert-butylphenol) Nauguard 524 Tris(2,4-di-tert-butylphenyl) phosphite Santonox 4,4-Thiobis(6-tert-butyl-m-cresol) Tinuvin 328 2-(2A-Hydroxy-3,3,5-di-tert-amylphenyl)benzotriazole Tinuvin 770 Bis(2,2,6,6-tetramethylpiperidin-4-yl) sebacate Tinuvin P 2-(2-Hydroxy-5-methylphenyl)-2H-benzotriazole Topanol CA 1,1,3-Tris(2-methyl-4-hydroxy-5-tert-butylphenyl)butane Ultranox 626 Bis(2,4-di-tert-butylphenyl)pentaerythritol diphosphite Weston 618 Distearylpentaerythritol diphosphite 102R Analyst, September 1997, Vol. 122related to the density, which in turn depends on pressure and temperature. Increasing pressure at constant temperature increases density and solvating power.However, increasing temperature at constant pressure decreases density and hence solvating power. CO2 is most widely used as solvent because of its convenient critical temperature (31.3 °C) and pressure (7.4 MPa), low cost, low toxicity and non-explosive character. Liquid solvents called modifiers are sometimes added to the CO2 in order to increase solubility or displace analytes from a matrix. MAE Sample and solvent are placed in a container and heated using microwave energy.The technique has evolved from closedvessel microwave acid digestions. The apparatus typically consists of closed vessels with temperature and pressure control, allowing the solvent to be heated under pressure above its normal boiling-point and remain liquid. The solvent must contain a component with a high relative permittivity to be heated by microwaves. Carousels of extraction vessels may be used, allowing for simultaneous extraction of up to 12 samples.ASE The sample is loaded into an extraction cell and solvent is pumped into the cell, which is heated in an oven. The temperature and pressure are programmed by the user. Pressure is applied to keep the solvent liquid above its normal boilingpoint. After the pre-set extraction time, more solvent is pumped through the cell into the collecting vessel and the remaining solvent is purged into the collecting vessel with nitrogen. The equipment is automated, allowing up to 24 sequential extractions to be programmed.In this paper, published accounts of laboratory methods for the extraction from polymers are reviewed. The new methods are here compared with each other and with traditional methods. Although there is a considerable literature on SFE, there are few reports on the other novel techniques. Supercritical Fluid Extraction SFE has been used for a great many matrices, including the extraction of environmental contaminants, natural products and food processing as well as polymers, which are discussed in several reviews.25–29 Extraction from polymers is one of the more recent uses of SFE.Early experiments established that SFE of polymers was a viable extraction method and could be much faster than traditional methods. Anton et al.30 extracted oligomers and an unidentified additive from an ethylene– propylene O-ring using CO2 at 25 °C and 6.6 MPa for 1 h. A second extraction produced more analyte, indicating that the extraction was not complete after the first hour.Hirata and Okamoto31 found that additives could be extracted at 90% recoveries from PE and PP films using CO2 at 25.4 MPa and 35 °C within 2 h, compared with 24 h for Soxhlet extraction. Cotton et al.32 showed that SFE of additives and oligomers from PP, nylon, PET and poly(ether-ether ketone) (PEEK) were possible. Irganox 1010, Irgafos 168 and erucamide were extracted from ground PE and Tinuvin 770 from PP at 92–95% recoveries in 15 min at 42.9 MPa and 60 °C.At 14.3 MPa the same recovery was achieved in 30 min.33 Following these successful extractions the optimum extraction conditions have been investigated. Extraction Process The two main factors in SFE are solubility of extractant in the fluid and rate of mass transfer out of the matrix. The mass transfer from the polymer is by diffusion from the bulk polymer to the surface, where dissolution in the supercritical fluid (SF) can occur.Bartle et al.34 described a simple model for extraction from spherical particles called the ‘hot ball’ model. This assumed that the concentration of extractant in the SF was effectively zero, and the only limiting step was transfer out of the matrix by a process which could be modelled as diffusion. This successfully predicted the characteristics of the extraction curve of ln(m/m0) versus time for extraction of BHT from PP (where m0 is the initial concentration in the plastic and m is the amount remaining in the plastic).These are that the curve falls steeply initially, as the extractant is extracted from the surface, and then becomes linear (Fig. 1). There were deviations from the predicted onset of the linear portion and of the extrapolated intercept of the linear portion with the m/m0 axis. These were explained in terms of non-spherical particles, solubility limitations and non-uniform distribution of the extractant in the polymer.The success of the model indicates that the basic processes can be modelled as diffusion. However, the deviations indicate that solubility and other factors are also significant. The linear part of the logarithmic plot can be extrapolated to determine the concentration of an extractant without complete extraction by using three extraction periods of equal duration. The amount m0 can be found using the equation m0 = m1 + [(m2)2/(m2 2 m3)]. The first extraction must be long enough such that the second and third extractions are on the linear portion of the curve.The extrapolation method successfully gave the extractable amount of BHT from PP. The model was extended to cover polymer films and non-uniform distribution of the extractant using extraction of cyclic trimer in PET as an example.35 The temperature chosen (70 °C) was just above the glass transition temperature (Tg) (69 °C), but the amount extracted was considerably less than that extracted by Soxhlet extraction.Thus, much of the trimer is unavailable to SFE at this temperature. This is consistent with the findings of Ashby,36 who found that the overall migration from PET into olive oil was negligible until a temperature of 130 °C was reached. The effect of solubility on extraction was incorporated into the model by Bartle et al.37 using extraction of Irgafos 168 from PP to test the model. At 70 °C, the effect of increased pressure (solubility) on the extraction curve was to increase the slope of the initial steep ln(m/m0) curve and increase the slope of the linear portion of the graph, i.e., increased extraction rate.The Fig. 1 Example of ln(m/m0) plot for SFE from polymers using the ‘hot ball’ model. Analyst, September 1997, Vol. 122 103Reffects of pressure and flow rate on extraction are explained theoretically by Clifford et al.38 A method for direct observation of the extraction process in real-time is described by Howdle et al.39 using an organometallic complex.The IR absorption bands are sensitive to the environment, and therefore a shift in the spectrum occurs when the complex is extracted from the polymer. This allows monitoring of the extraction process in situ. SFE is therefore an effective method of extraction from polymers. There are several factors that can affect the success of SFE: temperature, pressure, time, addition of modifier, the matrix and the compound extracted. The interaction of these variables is particularly complex for extraction from polymers, partly because the solvent can interact with the polymer.Factors Affecting SFE From Polymers Effect of Supercritical CO2 on Polymers CO2 can be dissolved in polymers and effectively plasticise and swell them. A high pressure view cell with a sapphire window has been used to measure the swelling of soils and plant materials during treatment with supercritical CO2.40 A direct relationship was observed between the degree of swelling and the efficiency of extraction.Similar effects would be expected from polymers, with the size of the effect depending on the amount of CO2 absorbed. The amount of CO2 absorbed will depend on temperature, pressure and the polymer concerned. Shieh et al.41,42 studied the effects of supercritical CO2 on nine crystalline and 11 amorphous polymers. The appearance, mass changes, physical properties and solubility of CO2 were examined after removal from the fluid.The amorphous polymers were most affected, many exhibiting significant swelling, particularly poly(methyl methacrylate) (PMMA). Tg was measured for PMMA and glycol-modified poly(ethylene terephthalate) (PETG) and was found to be depressed after exposure to CO2. Crystalline polymers were plasticised less than the amorphous polymers. The Tg of PET was found to be depressed by 52 °C when CO2 was sorbed at 2 MPa.43 The greater effect on amorphous polymers is expected as it has been reported that CO2 is not soluble in the crystalline regions of polymers.44 The strength of the effect on PMMA was possibly explained by investigations using Fourier transform infrared spectrometry (FTIR), which determined that there were specific Lewis acid–base type interactions between the carbonyl group in PMMA and CO2.45 No such interaction was found in PE, and a weak interaction was found between CO2 and the p system of poly(styrene) (PS).Therefore, we can expect the greatest swelling and plasticisation in amorphous polymers with electron donating groups.Extraction at temperatures above Tg will generally be much faster than below Tg. Kalospiros and Paulaitis46 have developed a molecular thermodynamic model for predicting solvent-induced glass transitions as a function of sorbed gas in the polymer. However, CO2 can also induce crystallisation in polymers at lower temperatures,47 which would be expected to reduce the rate of diffusion.Condo and co-workers48,49 measured the depression of the Tg of PMMA, poly(ethyl methacrylate) (PEMA) and poly(methyl methacrylate- co-styrene) (SMMA60) by CO2 in situ, using creep compliance in a high pressure cell. The relationship with pressure and temperature was not simple, and four types of behaviour were predicted and found. The phenomenon of reverse vitrification was identified, where the polymer undergoes a glass to liquid transition by decreasing the temperature. This means that some polymers may be plasticised under much milder conditions of temperature and pressure than was previously believed.The interaction of CO2 with polymers can rapidly swell and plasticise the polymer, resulting in much faster diffusion and hence extraction. The plasticisation alters the Tg and softening points, and therefore the temperature and pressure selected for the extraction can be lower than would be expected from physical data available on the polymer.Effect of Temperature and Pressure Diffusion in polymers is a slow process with diffusion coefficients of the order of 10210 cm2 s21 at 40 °C. The rate of diffusion follows an exponential Arrhenius form with temperature, rate » A exp (2E/RT), where E is the activation energy, R is the gas constant and T the absolute temperature. This would indicate that increasing the temperature would exponentially increase the diffusion rate. If the solubility in the SF is not limiting, then higher temperature would lead to higher extraction rate.However, increase in temperature decreases the density of a SF at constant pressure, which reduces the solubility of the extractant in the SF. Therefore, it might be expected that a rise in temperature would lead to a rise in extraction rate, up to a point where solubility in the SF became a limiting factor. Increasing the temperature can also cause the polymer to undergo a transition from a glassy to a rubbery form at the glass transition temperature. The diffusion in the rubbery form is much faster than the crystalline glassy form.Therefore, a sharp rise in extraction rate would be expected at the Tg. Increasing the pressure increases the density of the SF and therefore the solubility of the extractant. At low temperatures, when the extraction is almost completely diffusion-limited, this should have little effect. The situation is further complicated because supercritical CO2 can plasticise a polymer and lower the Tg.The amount of lowering increases with amount of fluid in the polymer, which depends on the pressure and temperature. Supercritical CO2 has a much greater effect on amorphous polymers than on crystalline polymers. Therefore, the effect of pressure and temperature would be expected to be more pronounced in amorphous polymers. Increasing the pressure may increase the extraction rate in amorphous polymers, even when solubility does not limit the extraction as the plasticisation of the polymer increases with absorbed CO2.The effect of temperature on extraction of oligomers from PET was studied by Kuppers.50 Extraction at a constant CO2 density of 0.5 g cm23 (hence changing pressure) showed an increase in the rate of extraction of trimer from PET from 40 to 160 °C, but there was no jump at the Tg. Increasing the density at the same temperature gave even higher extraction rates, indicating that solubility is also limiting the extraction.However, it is possible that increased plasticisation of the polymer at higher pressures was responsible for the increase rather than the increased solubility. Plotting extraction at constant pressure (changing density) against T showed a sharp increase at Tg, then a flattening of the curve at high temperatures, presumably as solubility again became limiting. The lack of a jump in the constant density curve may indicate a change in the Tg at higher pressures of CO2.Cotton et al.51 extended the work on PET film to higher temperatures. They found an increase in extraction with temperature at constant pressure (40.6 MPa), but no discontinuity at Tg. Above 215 °C some melting took place, causing agglomeration which slowed extraction and blocked the restrictor, so 215 °C was the highest temperature used for extractions. The extrapolation method at this temperature gives quantitative recoveries using 3 3 30 min extractions, and 95% recoveries using 3 3 15 min extractions.The intercept of the extrapolated linear portion of the graph is lower than at 70 °C, indicating that solubility, as well as diffusion from the polymer, affects the extraction at the higher temperature. Schmidt et al.52 analysed carbonic acid diphenyl ester from poly(butylene terephthalate) (PBT) by SFE. At room temperature, six times less was extracted than at 55 °C. Hunt and Dowle53 found that extraction of diisooctyl phthalate (DIOP) and Topanol CA from 104R Analyst, September 1997, Vol. 122ground PVC increased with temperature up to 90 °C at 45 MPa, then levelled off, presumably as solubility became the limiting factor. The extraction rate increased on increasing the pressure from 35 to 40 MPa at 45 °C; increasing the pressure further to 45 MPa had no further effect. The extraction of dioctylphthalate (DOP) and dibutylphthalate (DBP) plasticisers from PVC was also measured for 25 min extractions by Marin et al.54 At a pressure of 52 MPa, the amount extracted by CO2 increased from 50 to 80 °C, at which temperature extraction was almost complete within the 25 min.At a constant temperature of 70 °C, increasing the pressure from 22 to 60 MPa increased the amount extracted from approximately 79 to 95%. At 100 °C, the same pressure increase led to a marginal increase in extraction from 95 to 98%. As the extraction is close to completion at the higher temperatures, it is not possible to see the full effect of the pressure increase.However, from solubility considerations, the effects should be greater at higher temperatures because the extraction then becomes solubility-limited. Extraction of tris(nonylphenyl) phosphite, Irganox 1076 and Weston 618 from low-density PE (LDPE) film was measured at 45.6 MPa at different temperatures.55 At 60 °C, only 40–60% was extractable. However, by heating to 150 °C, which is 10 °C above the melting-point, > 95% was recovered.This is against the general finding that heating above the melting-point reduces extraction rate as melting reduces the surface area of ground samples. Extraction of N,N-ethylenebisstearamide from PS increased from 77 to 94% by increasing the temperature from 60 to 150 °C. Extraction efficiency of BHT and Irganox 1010 from freeze-ground PP increased from 30 to 90 °C at a pressure of 30.4 MPa.56 At 90 °C, 60 min were sufficient for complete extraction of all additives except for Irganox 1010.During the extraction of flame retardants from poly(urethane) (PU), Mackay and Smith57 found that extraction at 30.4 MPa was rapid and quantitative within 10 min at 60 °C, whereas at 20.3 MPa recoveries were lower, and at 10.1 MPa some flame retardants were hardly extracted at all. The effect of pressure in selective extractions was pointed out by Engelhardt et al.58 At 15.2 MPa and 45 °C, erucamide was extracted from PE. By increasing the pressure to 20.3 MPa, an Irganox-type antioxidant was also extracted.They linked the SFE conditions needed to extract materials with their retention using reversed-phase (RP) HPLC. If 90% organic modifier is needed to elute a compound from an RP HPLC column, then an extraction pressure of 15.2–20.3 MPa is sufficient. If a compound elutes with 50–60% organic modifier by HPLC, then the pressure for SFE using CO2 needs to be 35.5 MPa. Cotton et al.59 measured the extraction of Irgafos 168, Irganox 1010 and Tinuvin 770 from PP pellets and ground PP.Increasing pressure gave increased extraction rates at 50 °C up to a limit of 30.4 MPa, after which solubility no longer limited the extraction. The rate increased sharply with increasing temperature, up to the melting-point of the polymer. Once melting occurs the particles coalesce and the surface area decreases. The diffusion coefficient was found to be two orders of magnitude higher than from previously published data.This was explained by the plasticisation and swelling of the polymer by sorbed CO2. The value of the extrapolation procedure was demonstrated by successfully determining the total concentrations of Irganox 1010 and Irgafos 168 in less time than was required to complete the extraction from ground PP. For the larger pellets, low results were obtained from the extrapolation because the ln(m/m0) versus time plot had not reached the linear portion after the first extraction period. Recovery of dialkyl organotin stabilisers from freeze-ground, unplasticised PVC at 17.7 MPa was found to increase with temperature up to 90 °C, then levelled off with further temperature increases, presumably as solubility became limiting. 60 Similarly, the recovery at 90 °C increased with increasing pressure up to 15.2 MPa and then levelled off as diffusion became limiting.The extraction was complete within 60 min at 90 °C and 17.7 MPa. No data are given for recoveries above 90 °C at pressures above 17.7 MPa, but unless melting had commenced at this temperature the extraction rate should be enhanced as solubility-limiting behaviour would be reached at a higher temperature.Extraction of ethylbenzene from PS was neither completely diffusion nor solubility limited, leading to unusual T and P relationships.61 The polymer was significantly swelled by the CO2, leading to a 106-fold increase in diffusivity of ethylbenzene in the polymer.Tg was lowered from 100 to 50 °C by only 6 MPa of CO2. Roston et al.62 used SFE to extract a sustained release drug covalently bonded to poly(butadiene). Optimum pressure for the extraction was 33.5 MPa. Increasing the pressure above this reduced the amount extracted. These workers suggest that this decrease at higher pressures was due to recovery increasing with linear velocity of extracting media through the cell up to a certain velocity, then decreasing. Decreased trapping efficiency at faster flow rates is also a possibility. The temperature could not be raised above 75 °C because the drug is not thermally stable.Significant swelling of the polymer was noted; 85% of the drug was recovered in 45 min with a further 5% recovered in the subsequent 30 min. Extraction of Irgafos 168, Irganox 1076 and Irganox 1010 from PE powder increased with increasing temperature at 30.4 MPa.63 However, at 15.2 MPa the amount extracted during the first 30 min initially increased with temperature as diffusion was faster, but then fell as solubility-limiting behaviour took over. The amount extracted at longer times was always higher at higher temperatures, as the concentration in the extracting fluid fell and diffusion control was re-established. At 50 °C, increasing pressure had little effect as the extraction was almost completely diffusion-controlled.At 80 °C, the extraction rate increased from 20.3 to 30.4 MPa. The recovery at 25.4 MPa and 80 °C was higher than by 36 h Soxhlet extraction with hexane.Lou et al.64 examined the extraction of caprolactam from nylon 6, and dimer and trimer from PBT. Extraction of caprolactam during the first 30 and 120 min increased with increasing temperature from 50 to 170 °C. No jump at the Tg was apparent, implying that the CO2 had suppressed the Tg. However, during extraction from PBT, the amount of the dimer extracted after 120 min was highest at 150 °C and for the trimer at 110 °C. This indicates that the larger trimer is less soluble than the dimer, and extraction becomes solubility-limited at a lower temperature.The effect of temperature on the extraction of styrene dimers and trimers from PS was demonstrated by Jordan et al.65 A very low density of CO2 was used (0.184 g cm23), and kept constant at different temperatures. Extraction at 120 °C (10.8 MPa) produced eight times more dimers and trimers than extraction at 80 °C (8.8 MPa).The effect of the density of CO2 was examined by extracting at 120 °C at 20.3 MPa. The extraction time was reduced so that the same total amount of CO2 was used as in the lower pressure experiment. Exact comparisons of peak areas were not possible because the chromatographic system was overloaded, but at the higher pressures a sample size about 25% of that used at the lower pressure caused saturation. Therefore, the higher pressure extraction resulted in significantly greater extraction.The effects of pressure are more pronounced for extraction from an amorphous rubber. Burgess and Jackson66 found that extraction of carbon tetrachloride from chlorinated poly- (isoprene) could be completed within 40 min at 60 °C and 21 MPa, and not at higher temperatures or pressures. The normal Tg is 120 °C, but the softening point is lowered by the CO2 at high pressures. Therefore, increasing the pressure at high temperatures lowered the softening point and the polymer particles coalesced, reducing the surface area. Extractions on Analyst, September 1997, Vol. 122 105Rcrystalline polymers at low temperatures usually show little improvement with increasing pressure. However, in this case at 40 °C the extraction improves with increasing pressure. At this low temperature the extraction is unlikely to be solubilitylimited, and therefore the greater extraction is probably due to increased swelling of the polymer at higher pressures.The polymer matrix affects the extraction efficiency, as demonstrated by Juo et al.,67 who extracted Chimassorb 944, distearyl 3,3A-thiodipropionate (DSTDP), Irganox 1010, Irganox 1076, Tinuvin 144, Irganox 1098, Ionox 330 and oxidised Nauguard 524 from LDPE and high-density PE (HDPE). Both polymers were extracted at 60.8 MPa and 40 °C, for 30 min for LDPE and 5 h for HDPE. All additives were extracted from the LDPE, whereas some additives would not extract from HDPE even with 3% methanol modifier.Generally, diffusion from the polymer is the rate-limiting process, particularly during the later stages. Therefore, for optimum extraction, a high temperature should be used to maximise diffusion. High pressure should also be used for rapid extractions, both to increase solubility of the analyte and to plasticise the polymer. However, high pressures may lower the softening point of the polymer, and therefore the optimum conditions need to be experimentally determined. Effect of Flow Rate The flow rate will affect the extraction only if the extraction is solubility-limited.In this way changes to flow rate are similar to changes in pressure. The effects of pressure and flow rate on SFE from polymers have been modelled.38 For a solubilitylimited extraction, increasing the pressure will increase the rate of extraction up to the point at which diffusion becomes ratelimiting. At faster flow rates, the pressure limit will occur at lower pressure.Therefore, increasing the flow rate has a similar effect to increasing the pressure. This does not take into account the increased plasticisation of the polymer at higher pressures of CO2. Hawthorne et al.68 found that the flow rate had little effect on extraction of alkylbenzenes from PS beads except at a very low flow rate of 0.25 ml min21. For extraction of antioxidant from PE powder, there was little effect of flow rate at 50 °C, but some effect was noted at 80 °C as the faster diffusion at the higher temperature raised the concentration in the fluid.63 In contrast, Baner et al.69 found a marked reduction in the time to determine total extractables from Biopol (a biodegradable polymer) with increased flow rate.At a flow rate of 1.5 ml min21, the extraction took 40–50 min, reducing to about 15 min with a flow rate of 4.5 ml min21 and 8 min for 8.5 ml min21. As the total volume of CO2 is similar in each case, it seems that this extraction is almost completely solubilitylimited.Effect of Modifiers CO2 is characterised as a non-polar solvent with a solubility parameter similar to hexane. It does, however, have some affinity with slightly polar molecules because of its molecular quadrupole. For more polar molecules the addition of polar modifiers is used to increase the polarity of the supercritical phase and the solubility of polar compounds. In extraction from environmental matrices, binding of the extractant to active sites is very important, and the modifier can displace the extractant from the matrix, increasing extractability. This is less likely to be a factor in extraction of polymers.However, if they are similar in molecular character, modifiers can act by swelling the polymer, increasing diffusion and extraction rates. Modifiers used for this purpose may be non-polar or aromatic as well as polar. Therefore, modifiers may enhance extraction even when solubility is not a limiting factor.Hunt and Dowie53 found that only 50% of Topanol CA would extract from PVC at 45 MPa and 90 °C within 30 min. The addition of methanol modifier enhanced the extraction. The recoveries increased with increasing methanol concentration in the SF from 2 to 15%. The total present in the polymer was not accurately known; therefore, it was not known whether the recovery was quantitative. Mixed messages on modifiers come from the work of Kuppers.50 Dichloromethane (DCM) enhanced extraction of oligomers from PET, the higher oligomers being most affected. Using methanol or propan-2-ol completely prevented extraction.The reason for the latter finding is not explained. Recovery of oligomers and caprolactam from ground nylon 6 was poor using CO2 at 60 °C.70 Addition of methanol to the extraction vessel during a 5 min static extraction period enhanced extraction, which was almost complete for caprolactam in 15 min. For oligomers the recovery was lower, but was further improved by using 7.5% methanol during the dynamic extraction stage.Methanol enhanced the extraction of caprolactam from nylon for Jordan et al.65 Spiking the extraction cell with methanol before extraction with CO2 at 75 °C and 25.3 MPa extracted approximately three times as much caprolactam as the same extraction conditions without methanol. The same amount of caprolactam spiked onto Celite was completely recovered using CO2 only, indicating that solubility limitations were not the reason for the low recovery from the polymer.These workers suggest that the methanol was swelling the polymer and therefore enhancing the extraction rate. Lou et al.64 examined the effect of modifiers on extraction of caprolactam from nylon 6, and dimer and trimer from PBT. The modifiers (hexane, chloroform, methanol and benzene) were spiked into the extraction cell for an initial static extraction period. Methanol was the most effective modifier for nylon and chloroform for extraction from PBT.When compared with extraction using unmodified CO2, the modifiers had the most effect at lower temperatures. In fact, at higher temperature the amount extracted at longer times was lower than when using CO2 alone. This was thought to be due to a shrinking of the polymer after the modifier had been extracted from the cell. Improvements in extraction efficiencies could be achieved at both diffusion and solubility limiting conditions, indicating that the modifiers worked both by swelling the polymer and increasing solubility.Addition of benzene during a static extraction with CO2 for 30 min improved recoveries of antioxidant from PE powder.63 Benzene was selected as it is known to swell PE, and this was thought to be the main reason for the greater recoveries. However, for Irganox 1010 the greater solubility was also thought to be significant. Garde et al.71 optimised the extraction of antioxidants from PP.They found that the best methods used two static extractions of 30 min, one with hexane and one with methanol. Use of methanol as modifier in a 60 min dynamic extraction only slightly enhanced the recoveries, indicating that swelling of the polymer is the most important factor. Propane added to CO2 had little effect on the extraction of chlorofluorocarbons (CFCs) from PU foams.72 Using unground foam, the time required for 99% extraction was 1.5 h, whereas with addition of propane (6% m/m) 2 h were needed.However, using material ground to 0.2 cm particle size, 1 h was needed with CO2 alone and only 0.6 h with addition of propane. Compounds may be added to the supercritical phase as a reactant rather than as a simple modifier. Roston et al.62 used formic acid to hydrolyse the bond between the polymer and a drug. Using methanol as a modifier there was almost no extraction. Cedergren et al.73 extracted nicotine from PE patches using modified CO2.They found that 1 m triethylamine in methanol was the best modifier. The tertiary amine was added to prevent reaction with the CO2, which could lead to insoluble carbamates. Pre-treatment of the PE with concentrated triethylamine together with a static extraction stage gave the best recoveries. 106R Analyst, September 1997, Vol. 122Modifiers generally accelerate extraction when the modifier interacts with the polymer more than CO2. The modifier can then cause greater swelling than with CO2 alone.Therefore, methanol is useful when extracting from polar nylons, and aromatic modifiers from non-polar poly(olefins). The greater solubility in modified CO2 is most effective when extracting large or polar molecules. Nature of the Extractant The nature of the extractant affects extraction in both solubility and diffusion limiting cases. The larger the molecule the slower the diffusion in the polymer, and hence the slower the extraction.High molecular weight compounds tend to be less soluble in supercritical CO2 and hence solubility will also limit the extraction more for larger molecules. The extraction of DIOP from PVC at 45 MPa and 90 °C was almost complete after 20 min, whereas the polar Topanol CA was only 50% extracted.53 Irganox 1010 has proved difficult to extract in a number of cases when smaller compounds were extracted.56,63 Higher nylon oligomers also proved difficult to extract.70 Extraction of flame retardants at 30.4 MPa for 10 min at 60 °C was complete for molecules of molecular weights 286, 388 and 472 Da, but the largest molecule with a molecular weight of 571 Da was only 77% extracted.57 There are some reports of smaller molecules extracting slower than larger molecules.The plasticiser DOP extracted faster from a 100 + 50 PVC–plasticiser blend than the smaller DBP.54 This was attributed to DBP being more strongly bonded to matrix sorption sites than DOP.However, it should be noted that DOP may be a more effective plasticiser than DBP, hence leading to faster extraction rates. Effect of Particle Size As one of the limiting steps in extraction is diffusion to the surface of the polymer, the particle size or film thickness is extremely important.52,53,59,60,70,74 The diffusion coefficient of additives in polymers at 40 °C is typically about 10210 cm2 s21. The rate of diffusion (s21) is proportional to D/L2, where L is the length of the shortest dimension. As a first approximation, therefore, for an extraction time of 1000 s (17 min) a particle diameter of 0.3 mm is required.Therefore, grinding of the polymer is often an essential step in the analysis. An exception to this is the extraction of thin films and foams, for which the shortest dimension is small. Garde et al.71 could extract no more than 50% of antioxidants from PP pellets, but could achieve 90% recoveries from the same polymer extruded into film. Loss of volatile additives is possible owing to the heat generated by grinding polymers.Therefore, the polymer must be frozen, usually with liquid nitrogen, before grinding. Hexabromocyclododecane was extracted from PS styrofoam and Irganox 1010 from PE ethafoam within 30 min at 150 °C at 45.6 MPa.55 Flame retardants were extracted from PU foam within 5 min at 30.4 MPa, 60 °C.57 A different approach to avoiding the need to grind the sample was taken by Mackay and Smith.75 Examination of polymer samples (unplasticised PVC) after SFE revealed that the CO2 had not penetrated into the core of the sample, and recoveries were low.The polymer was dissolved in THF and an internal standard added. The solvent was evaporated and the resulting polymer samples were extracted with CO2 at 50 °C, 35.5 MPa for 10 min. Extraction was far from complete, but the internal standard extracted to the same extent as the analyte and the correct analysis result was obtained.Extraction of additives from liquid polymers has been achieved by bonding them onto silica.76 Additives from poly(alkylene glycol) and sorbitan ester were extracted by SFE with CO2 at 45 °C. Oligomers were co-extracted, but most of these could be removed by a silica guard column. Non-quantitative extraction of additives from a range of polymers was demonstrated by Braybrook and Mackay77 for biocompatability testing. Solvent extraction is problematical for this purpose because solvent residues interfere with the biocompatability test.Mackay and Smith78 showed the value of on-line SFE–SFC–MS by identifying a range of additives from PU. Some of these additives could not be analysed by GC–MS. Summary of SFE SFE can provide a method for extracting additives from polymers. It is much faster than Soxhlet extraction. Grinding of the sample is usually necessary except for thin films and foams. Extractions are likely to be fastest from amorphous polymers, but the temperature must be carefully chosen to be lower than the softening point under experimental conditions.This softening point is likely to be lower than that of the polymer at atmospheric pressure, and the conditions of temperature and pressure must be carefully selected. For crystalline polymers, the temperature should be as high as possible without onset of melting, and the pressure also should be as high as possible to ensure no solubility limitations and maximum plasticisation of the polymer.Most difficulty is likely to be experienced with extraction of high molecular weight and/or polar compounds. The use of a modifier is likely to enhance the extraction in these cases. The modifier should be one that is known to swell the polymer. For extractions that still take a long time, an extrapolation procedure can be used to determine concentration in the plastic without complete extraction. Microwave Heating Microwave-assisted sample preparation techniques are widely used in analytical laboratories, largely in the field of digestion of samples.Zlotorzynski79 reviewed the use of microwave radiation in analysis and suggested that microwave extraction is in its infancy. Reviews on the applications of microwaveassisted sample preparation have recently been published.80,81 There are several publications on MAE in environmental analysis, but few on extraction from polymers. The advantages of MAE are that samples can be rapidly heated and several samples can be extracted simultaneously. The sample can be contained in a pressure-resistant vessel with safety valves.The solvent can therefore be heated above its normal boiling-point. At 1.2 MPa, the temperature reached in a pressure vessel with acetone is 164 °C, with dichloromethane 140 °C, and with acetonitrile 194 °C.81 The high temperatures will accelerate diffusion through the polymer and hence improve extraction rates.The high temperature may increase the swelling of the polymer owing to greater solvent–polymer interaction at higher temperatures. Another advantage compared with Soxhlet extraction is that any composition of solvent mixtures can be used. During Soxhlet extraction the solvent is the vapour condensate, which will only have the same composition as a mixture of solvents if an azeotropic mixture is used. The solvent selected must have a high relative permittivity to be heated by microwaves; therefore, pure hydrocarbons cannot be used. Freitag and John82 used acetone–heptane (1 + 1) to extract additives from LDPE, HDPE and PP.After grinding to 20 mesh, 91–97% of Irgafos 168, Chimassorb 81 and Irganox 1010 were extracted within 6 min from HDPE and within 3 min from PP and LDPE. Molecular weight had a significant effect, with Irganox 1010 being the slowest to extract. 1,1,1-Trichloroethane gave slightly faster extractions, but is environmentally less desirable.The order of extraction of Irganox 1010 from the polymers was PP > LDPE > HDPE, although the scatter of the LDPE results was fairly high. Dissolution of larger particles in toluene–1,2-dichlorobenzene was effected in 5 min. However, Analyst, September 1997, Vol. 122 107Rthe dissolution method gave only 85% recovery of the additives from the PE samples, although 95% recovery was achieved from the PP. Extractions from larger pellets were less efficient.The vessels were pressurised, but the temperatures and pressures reached were not reported. Nielson83 compared microwave extraction with sonication for HDPE, PP and LDPE. BHT, Irganox 1010 and Irganox 1076 could be extracted at > 90% recoveries from ground HDPE in 20 min at 50% power. Two solvent systems were used: propan- 2-ol–cyclohexane (1 + 1) and DCM–propan-2-ol (98 + 2). In each case the propan-2-ol was present to absorb the microwave energy and the other solvent to swell the polymer.The sample was stirred at 5 min intervals. The polymer was ground to 20 mesh, and 5 g of plastic were used to 50 ml of solvent. This is a much larger sample than the 50–100 mg typically used in SFE. Butylated hydroxy ethylbenzene (BHEB), Isonox 129 and erucamide slip agent were extracted from LDPE using the same conditions. From PP, Irganox 3114, Irganox 1010, Irganox 1076, Irgafos 168, Tinuvin 328, Ultranox 626, Cyasorb UV 531, BHT and AM 340 could be extracted from the ground polymer (20 mesh) [5 g of plastic to 50 ml of solvent, DCM–propan-2-ol (98 + 2)] in 20 min at 20% power, with stirring required every 5 min.Only the Irganox 3114 had a low recovery of 79%. When larger pellets were used the recoveries were also high, except for Irganox 1010, for which only 50% recovery was possible without grinding. The vessels were not pressurised and the temperature using the DCM–propan-2-ol mixture did not exceed 50 °C. Costley et al.84 report on extraction of cyclic trimer from PET using a variety of solvents heated to 120 °C in pressure vessels.The polymer fused at temperatures above 120 °C with DCM; therefore, higher temperatures were not investigated. Extraction for 2 h with DCM at 120 °C gave the same extraction as 24 h Soxhlet extraction with xylene as solvent. MAE using hexane– acetone (1 + 1), water, acetone and acetone–DCM (1 + 1) all gave much lower recoveries than DCM at 120 °C. From these examples, MAE appears to be a rapid and effective technique for polymer extractions.The solvent can be selected to swell the polymer, provided that some microwave absorbing solvent is also present. The polymer needs to be ground for efficient extraction. However, there are too few reports for firm conclusions to be drawn. Ultrasonic Extraction Ultrasonic extraction works principally by agitating the solution and producing cavitation in the liquid. This would be expected to enhance the rate of transfer across the polymer/liquid boundary layer, but not to increase the diffusion of compounds within the polymer. There are several reports of ultrasonic extraction from polymers.Brandt85 extracted tri(nonylphenyl) phosphite (TNPP) from a styrene–butadiene polymer using 2 3 20 min extractions with isooctane as solvent. This compares with 2 3 1 h extractions for boiling under reflux. Nielson83 compared ultrasonic extraction with MAE for extraction of a variety of analytes from PP, LDPE and HDPE (see under Microwave Heating).For all samples, the ultrasonic extraction could be achieved within 1 h, provided that the samples were stirred every 10 min. For LDPE and PP most compounds were extracted within 10 min. The exception was Irganox 1010, which required 1 h for > 90% extraction. Further experiments by Nielson86 on extraction from HDPE using the same regime confirmed these results. However, where phosphite antioxidants are present the use of DCM–cyclohexane was preferred as it prevented hydrolysis of the phosphite by the alcohol. Caceres et al.87 used the same solvent mixtures as Nielson83 to extract Tinuvin 770 and Chimassorb 944 from HDPE.The additives could only be extracted at less than 20% recoveries from pellets using ultrasonic extractions of up to 5 h. The size of the pellets is not given, but the fact that the sample was not ground may be the reason for the difference in results. Extraction of Chimassorb 81 from LDPE and ethylene-vinyl acetate polymer (EVA) was achieved with 6 h standing of the sample under DCM (maceration) followed by 3 3 20 min sonication.88 The initial maceration time allowed swelling of the polymer.The extraction time using maceration alone was 48 h. In this case it was important to avoid high temperatures as these could degrade the analyte. Ultrasonic extraction from polymers has given some reasonably fast extractions, but the advantages over shaking the sample have not been widely demonstrated.ASE Lou et al.89 extracted monomers and oligomers from nylon and PBT using hexane as extraction solvent in a home-made ASE. They investigated the effect of temperature, pressure and flow rate with 20 min static followed by 30 min dynamic extractions. Pressure was found to have no effect other than to keep the solvents liquid at high temperature and flow rate had little effect between 0.4 and 2 ml min21. Extraction efficiencies increased in all cases as the temperature was raised from 50 to 170 °C, which was attributed to faster diffusion rates.These workers observed that solvents which are good swelling agents, and hence give fastest extractions during Soxhlet extraction, tend to dissolve the polymer at the high temperatures used during ASE. Dissolved polymer re-precipitates on cooling and can block transfer lines in the instrument. Solvents therefore cannot be selected on the basis of those used for atmospheric pressure extractions.Hexane was used in the extractions even though it gives poor recoveries during Soxhlet extraction. Lou et al. point out that selection of a suitable extraction solvent is probably the most difficult step in optimising ASE, as there are few data on the solubility of polymers in solvents at high temperatures. These workers had previously analysed the same polymers using SFE with pure and modified CO2 64 and compared the result with that obtained using pure CO2 at 170 °C and 30.7 MPa.The recoveries for ASE for caprolactam from nylon and the dimer and trimer from PBT are 1.1, 6.5 and 37.6 times higher, respectively, than those obtained with SFE. However, at these conditions the SFE was not optimum, particularly for the dimer and trimer, where the peak extraction after 30 min occurred at 110 and 90 0C, respectively. This extraction peak at low temperatures clearly indicates solubility-limited extractions.Addition of modifier (methanol for nylon and chloroform for PBT) during the static extraction stage further increased recoveries from SFE, particularly for the dimer and trimer, but recoveries were still higher with ASE by approximately 1.5 times for the dimer and trimer. No experiments were performed with modified CO2 during the dynamic extraction. From these results, it appears that ASE offers significant advantages over SFE with CO2 alone for extraction of compounds with a low solubility in CO2.Other Extraction Methods Some new variations on the dissolution theme have recently been published. Staal et al.90 dissolved polycarbonate and polysulfone in THF and precipitated them onto a C18 guard column. A gradient elution from 50 + 50 water–THF as nonsolvent to 100% THF successively eluted the additives, then the oligomers and finally the polymer itself. No quantitative work was reported The method could be adapted by altering the solvent programme to separate the compounds of interest.Another way to separate the additives from the polymer after dissolution was explored by Nerin et al.91 They linked a highperformance size-exclusion chromatography (SEC) column with a normal-phase HPLC column, via a three-way switching 108R Analyst, September 1997, Vol. 122valve. The polymer elutes from the SEC column first and is drained through the valve. The valve is then switched to allow the additives onto the analytical HPLC column.One problem with the dissolution method is that high boiling solvents are usually required to dissolve the polymer. The solvent is therefore difficult to remove after precipitation of the polymer. This was addressed by Macko and co-workers92,93 who used an autoclave to dissolve HDPE in heptane at 160–170 °C, well above the normal boiling-point. The polymer was precipitated by cooling, and after filtration the additives could be determined by direct analysis of the resulting solution by normal-phase HPLC.Alternatively, the solvents were relatively easy to remove by evaporation. The dissolution took 1 h, and the complete analysis time was 3 h. Caceres et al.87 compared several methods for extraction of Tinuvin 770 and Chimassorb 944 from HDPE pellets. Room temperature diffusion into chloroform and ultrasonication gave less than 20% extraction. Soxtec extraction with DCM for 4 h resulted in only 50% extraction. Dissolution of the polymer in dichlorobenzene at 160 °C for 1 h followed by re-precipitation of the polymer with propan-2-ol gave 65–70% recovery. The most successful method was boiling under reflux with toluene at 160 °C for 2–4 h, which extracted 95% of both additives.The relatively poor performance of the Soxtec extraction compared with the reflux extraction is probably due to the large difference in temperature between the boiling solvents. The pellets were not ground and the size was not specified.Conclusions There are several methods that can be used for the extraction of low molecular weight material from polymers. The principal points of each are given in Table 2. Table 3 shows a summary of the extraction papers discussed here. There is a choice to be made between inexpensive, simple equipment giving long extraction times and more expensive but rapid techniques. Soxhlet extraction will generally extract all additives, but extraction times can be as long as 48 h. During Table 2 Summary of extraction techniques Soxhlet SFE Microwave Sonication ASE Solvent Any CO2 (possibly Must contain a Any Any modified) microwave absorbing component Typical sample size 1–5 g 10–100 mg 1–5 g 1–5 g 1–10 g Analysis time 6–24 h 20 min–2 h 30–60 min 40–60 min 15 min Solvent usage 50–100 ml 10 ml 30 ml 30–50 ml 30–50 ml Advantages Inexpensive, Low solvent Fast, low Economical Low solvent widely accepted use, fast, can be solvent use, can use automated, automated extract multiple any solvent solvents possible simultaneously Disadvantages Slow, large solvent Expense, may take Expense, not Not always effective Expense, not use time to optimise sufficient body sufficient body method of evidence for of evidence for extraction from extraction from plastics plastics Table 3 Polymer extraction summary Extraction Polymer Additive technique Comments Ref. Solid–liquid extraction— Poly (ethylene terephthalate) Cyclic trimer and other oligomers up to the heptamer Dissolution followed by precipitation 0.1 g of polymer was dissolved in dichloromethane and hexafluoropropan-2-ol (7 + 3 v/v 10 ml) and then acetone was added to precipitate the high molecular mass polymer. The sample was filtered, concentrated to dryness and the residue dissolved in dimethylacetamide.Positive ion atmospheric-pressure chemical ionisation (APCI) was used to analyse the extracts 15 Poly (ethylene terephthalate) Butyric acid (1), Malathion (2), Diazinon (3).All are recycling byproducts Dissolution/ precipitation (1) PET was dissolved in a large volume of hexafluoropropanol and dichloromethane. Large oligomers were then removed from the solution by polymer precipitation using acetone. Additives (2) and (3) as before but precipitated using methanol. Average recoveries from spiked samples were 80–98% 16 Poly (vinyl chloride) Dioctyl phthalate Dissolution/ precipitation Sample dissolved in THF and centrifuged at 20 000 rev min21.Polymer is then precipitated using ethanol and isolated by filtration. The filtrate is evaporated to dryness and analysed 94 Poly(vinyl chloride) and poly(vinyl chloride)–vinyl acetate copolymer Tinuvin 320; Cyasorb UV 9; Uvinul N- 539 Soxhlet extraction PVC film or finely ground PVC particles (1 g). Soxhlet extraction with diethyl ether for 16 h; evaporation to dryness; precipitate dissolved in THF (10 ml); filtration through Millipore Teflon filter of 0.5 mm pore size.Recoveries ranged from 65% (unground PVC, diameter 2–10 mm) to 94% (ground, i.e., < 0.5 mm) for Tinuvin 320; 59% (unground PVC, diameter 2–10 mm) to 97% (ground, i.e., < 0.5 mm) for Cyasorb UV-9; and 89% (film, thickness 150 mm) to 95% (ground, i.e., < 0.5 mm) for Uvinul N-539 95 Table continued on next page Analyst, September 1997, Vol. 122 109RTable 3 continued Extraction Polymer Additive technique Comments Ref. Poly(propylene) Chimassorb 944 Dissolution/precipitation Polymer was dissolved in decalin at 150 °C and re-precipitated by cooling 13 Polyamides Lubricants, e.g., stearic acid and ethylene bis-stearamide Dissolution followed by liquid– liquid extraction Dissolve 5 g of polymer in 30 ml of formic acid and extract with 150 ml of toluene for 18 h 14 Polycarbonate, polysulfone Additives, monomers and oligomers Dissolution THF dissolution of polymer samples.Preliminary note on the on-line extraction of polymers by multiple solvents on packed HPLC columns. 90 Poly(ethylene) Cyasorb 531 Soxhlet extraction Powdered sample (100 g) Soxhlet-extracted with 500 ml hexane or chloroform for 12 h. This was repeated three times. Then the extract was filtered and filtrate evaporated to dryness. Soluble fractions of low molecular weight polymers were removed from the individual extracts using methanol followed by evaporation to dryness and dissolution in chloroform. Chromatographic identification only 96 Poly(ethylene) Irganox 1010, Irgafos 168, a-tocopherol Disolution with hot heptane under pressure Polymer (1 g, cut into slices) was dissolved in heptane at 160–170 °C under elevated pressure in an autoclave.Polymer precipitated by cooling and supernatant analysed by HPLC. 2 h per sample needed for complete analysis. 92, 93 Poly(ethylene) (high-density) BHT, Irganox anti-oxidants, Isonox, Cyasorb, Am 340, MD 1024, Irgafos 168 Ultrasonic extraction Use of cyclohexane–dichloromethane as extraction solvent reduces risk of hydrolysis of Irgafos 168. 30–40 min of ultrasonic extraction needed for complete extraction. 86 Poly(ethylene) (high-density) Tinuvin 770, Chimassorb 944 Soxtec, ultrasonic extraction, room temperature diffusion, dissolution, reflux Boiling HDPE pellets under reflux with toluene was the most successful method 87 Poly(ethylene) (low-density) Ionol, Santanox, oleamide Flask-shaker Pelletised (7-mesh) LDPE (5 g) was shaken with solvents (10 ml).Carbon disulfide extracted Ionol and Santonox in about 2 h, much faster than isooctane. Extraction with powdered (50-mesh) LDPE was much faster 17 Poly(ethylene) (low-density) Chimassorb 81 Ultrasonic extraction and maceration Ground polymer (0.25 g) was placed in dichloromethane (3 ml), shaken and kept in the dark for 6 h, followed by 3 3 15 min ultrasonic extractions. This gave same extraction as 48 h of standing in the dark 88 Poly(ethylene) (low-density) DSTDP; Irganox 1035; Santonox R; peroxide initiator Vulcup Soxhlet extraction Poly(ethylene) film (5 g) Soxhlet-extracted with 100 ml THF, extract concentration to 5 ml.Chromatographic identification only 97 Polymer DLTDP; DSTDP; TNPP; Goodrite 3114; Weston 618; Topanol CA, Irganox 1076; Cyasorb UV 531 Solid–liquid extraction from polymers by soaking in boiling solvents Reflux pressed foil samples with boiling CH2Cl2 for 2 h; evaporate to dryness; solution in 5 ml THF; filtration through 0.5 mm Millipore Teflon filter 20 Poly(propylene) Irganox 1010, Irganox 1330, Irgafos 168, Irganox 3114, Atmer 163 and Tinuvin 326 Soxhlet extraction Soxhlet extraction using chloroform (50 ml).Optimum extraction obtained in 1 h using sieved ( < 1.18 mm) freeze-ground samples (3 g). For film and granular samples, 3 h extraction necessary 14 Poly(propylene) Tinuvin 770; Hostavin TMN 20; Tinuvin 144 Soxhlet extraction Sample (10 g) in powder or pellet form was extracted. Kumagawa extraction with CHCl3 for 16 h, extract concentration to 20 ml under a flow of nitrogen; then 80 ml of acetone added to precipitate oligomers. Sample filtered and washed with hot acetone; filtrate concentrated under a flow of nitrogen and finally made up to volume (10 ml) with CHCl3.Results for poly(propylene) pellets were 96.2% for Tinuvin 770 and 95.6% for Hostavin TMN 20 95 Poly(propylene) DLTDP; DSTDP; TNPP; BHT; Goodrite 3114; Weston 618; Topanol CA; Irganox 1076; Cyasorb UV 531; oleamide; erucamide; Ethyl 330; stearamide; Irganox 1010 Soxhlet extraction Sample (50 g) in pellet form.Soxhlet extraction with 250 ml CH2Cl2 for 48 h; evaporate to dryness; re-dissolving in 5 ml THF; filtration through Millipore Teflon filter of 0.5 mm pore size. Good results (88–120% recovery) reported for two out of three samples 20 Poly(propylene) Irganox 1010; Irgafos 168; Tinuvin 770; erucamide; Irganox 3114; Tinuvin 440; Tinuvin P Soxhlet extraction Sample (10 g) in pellet form.Soxhlet extraction with diethyl ether for 15 h; evaporation to dryness; wax precipitation by refluxing with 5 ml ethanol; cooling; filtration. Qualitative data only 98 Table continued on next page 110R Analyst, September 1997, Vol. 122Table 3 continued Extraction Polymer Additive technique Comments Ref. Poly(propylene) BHT, Topanol CA, Irganox 1076 Cold liquid solvent extraction Sample of beads or shavings (1 g). Overnight extraction with 5 ml of acetonitrile at ambient temperature in a sealed amber-coloured vial with constant stirring.Qualitative only 99 Poly(propylene) DSTDP, BHT, Topanol CA, Santowhite powder Cold liquid solvent extraction Sample (4 g of 8-mesh pellets) shaken in a 50 ml screw-capped, darkened glass vial with (a) 20 ml THF or (b) CH2Cl2 for 24 h at room temperature 22 Poly(propylene) copolymer Irganox 1330 Solid–liquid extraction from polymers by soaking in boiling solvents Sample frozen and pulverised (0.5–5.0 g).Refluxing for 40 min under nitrogen purging using 25–100 ml of decalin, hexane, CHCl3 and THF. After cooling, samples filtered through Whatman GF/A microfibre filter. For quantitative work a 100-fold molar excess of BHT (0.5–1.5 mg l21) was added to the extraction solvent to protect Irganox 1330 from oxidation during extraction. Results indicate that all solvents gave good recoveries provided that the sample was milled to a particle size of < 1.0 mm 100 Poly(styrene) (styrenic polymer) External lubricants Dissolution followed by evaporation Wash 20 g of poly(styrene) resin with ethanol.Collect washings and evaporate to dryness with nitrogen. Dissolve residue in chloroform. Analyse using FTIR 101 Poly(styrene) (styrenic polymer) Internal lubricants Dissolution followed by evaporation Dissolve 20 g of resin in 150 ml of dichloromethane. Add 100 ml of ethanol dropwise to precipitate polymer. Analyse using FTIR 101 PVC, PE BHT, tinuvin 326, Tinuvin 327, Irganox 1076, Cyasorb UV 9, Cyasorb UV 1084 Dissolution, separation of polymer by size-exclusion chromatography Polymer dissolved and separated by SEC.Polymeric fraction diverted to waste and fraction containing analytes directed to silica HPLC column. Qualitative data only 88 Styrene–butadiene rubber Tris(nonylated phenyl) phosphite Ultrasonic extraction Ultrasonic extraction with isooctane for 2 3 20 min gave same extraction as 2 3 1 h for boiling under reflux 85 Supercritical fluid extraction— Biodegradable lactide- co-glycoside Oligomers, stearic acid, anthraquinone- based dye SFE Screening method for in vitro cytotoxicity testing.SFE extracts not contaminated with solvent residues 77 Biopol Triacetin, total extractables SFE 40 °C, 35 MPa, 10 min; CO2 only. Pieces were cut from a bottle. Faster flow rates gave faster extraction with almost complete recovery after 8 min at a flow rate of 8.5 ml min21 69 Nylon Caprolactam SFE SFE–SFC with CO2 only on milled polymer at 35 MPa and 75 °C showed extractable caprolactam.Addition of methanol modifier to the extraction cell significantly increased the amount of extracted caprolactam. The methanol was thought to act by swelling the polymer rather than through enhanced solubility. 65 Nylon 6 Caprolactam and oligomers SFE SFE conditions: temperature, 60–80 °C; CO2 density, 0.85 g ml21. Extraction effeciencies higher than methanol Soxhlet extractions.Use of 7.5% methanol modifier and additional modifier in extraction cell necessary for high extraction efficiencies 70 Nylon 6 Caprolactam SFE and Soxhlet extraction Study of modifier addition and temperature variation. Conditions for SFE: modified supercritical CO2; temperature, 50–170 °C; pressure, 30 MPa; extraction time, 20 min static and 30 min dynamic. Comparable results obtained by both techniques. Chloroform, benzene and methanol as modifiers had a large effect at low temperatures, the effect decreased at higher temperatures 64 Nylon pellets Cyclic trimer and ethyl bis-stearamide SFE On-line SFE–SFC.Conditions for SFE: CO2 only; temperature, 70 °C; pressure, 5–40 MPa sample size: 5–10 mg. Qualitative only 32 Poly(butylene terephthalate) Dimer/trimer SFE and Soxhlet extraction Study of modifier addition and temperature variation. Conditions for SFE: chloroform-modified supercritical CO2; temperature, 50 °C; pressure, 30 MPa; extraction time, 20 min static and 30 min dynamic.Comparable results obtained for the dimer extraction; lower recovery by SFE for the trimer (73%). Chloroform was the most effective modifier, followed by benzene and methanol 64 Poly(butylene terephthalate) polymers Volatiles SFE On-line–GC. Conditions for SFE: CO2 only; temperature, 55 °C; pressure, 20 MPa; extraction time, 10 min. Average recovery from a 25 mm sample thickness is 98% 52 Poly(ethylene terephthalate) Cyclic trimer SFE and Soxhlet extraction On- and off-line SFE–SFC.SFE conditions: CO2 only; temperature, 70 °C; pressure, 40 MPa; extraction time, 13 consecutive 30 min extractions, i.e., over a period of 6.5 h. Poor recovery obtained, as compared with Soxhlet, by SFE 35 Poly(ethylene terephthalate) Cyclic trimer SFE and liquid solvent extraction CO2 only; temperature, 90–215 °C; pressure, 40 MPa; extraction time, 30 min. Sample size: 0.02–3 g of film. Soxhlet conditions: Sample size, 12–13 g of film; solvent, xylene; extraction time, 24 h.Good results obtained by SFE at elevated temperature 51 Poly(ethylene terephthalate) Cyclic trimer SFE On-line SFE–SFC: conditions for SFE: CO2 only; temperature, 70 °C; pressure, 5–40 MPa. Sample size: 5–10 mg. Qualitative only Table continued on next page Analyst, September 1997, Vol. 122 111RTable 3 continued Extraction Polymer Additive technique Comments Ref. Poly(ethylene terephthalate) Cyclic trimer and other oligomers SFE Conditions: supercritical CO2 with and without modifiers used (methanol, isopropyl alcohol, dichloromethane and acetone); temperature, 40–150 °C; density, 0.5–0.9 g ml21.Sample size: 0.75 g of ground PET chips or 1 g of PET fibres. The use of isopropyl alcohol and methanol as modifiers prevented extraction; a 7% dichloromethane modified supercritical CO2 was beneficial. A three-stage SFE procedure was recommended using supercritical CO2 only 50 Poly(vinyl chloride) Diisooctyl phthalate (DIOP) as plasticizer, chlorinated polyethylene wax, Topanol SFE and liquid solvent extraction Conditions for SFE: CO2 only; temperature, 45–115 °C; pressure, 7–45 MPa; extraction time, 0.5–435 min.Sample size: 0.2–0.4 g. Analysis using off-line packed SFC. Results for SFE compare favourably with those obtained using liquid extraction and the actual formulation value 53 Poly(vinyl chloride) Stabilizers SFE and dissolution/ precipitation Screening method for food contact plastics, non-quantitative.SFE extracted same stabilizers as precipitation method 5 Poly(vinyl chloride) Tinuvin P SFE Polymer was dissolved in THF and internal standard added. The polymer was re-cast and extracted with CO2 at 50 °C and a pressure of 30 MPa for 10 min. Extraction was far from complete, but the internal standard compensated and good analyses were obtained 75 Poly(vinyl chloride) Dibutyl phthalate, dioctyl phthalate SFE and Soxhlet extraction Best SFE conditions: temperature,95 °C; pressure, 48 MPa; extraction time, 25 min; unmodified CO2.Extraction efficiency was 98% compared with 5 h Soxhlet extraction with cyclohexane 54 Poly(vinyl chloride) Organotin stabiliser SFE SFE conditions: pressure, 18 MPa; formic acid modifier; temperature optimum at 90 °C. Extraction complete within 60 min 60 Poly(vinyl chloride) Diisooctyl phthalate, diethylhexyl phthalate, BHT, Tinuvin P, tributyltin chloride, vintyl chloride monomer SFE Screening method for in vitro cytotoxicity testing; therefore, not quantitative. SFE conditions: pressure, 42 MPa; temperature, 50–150 °C.Main advantage over solvent extraction is that the extracts are not contaminated with toxic solvent residues 77 Poly(alkylene glycol (PAG) Additives SFE SFE conditions: temperature, 45 °C; pressure, 30 MPa; CO2 only. PAG was adsorbed into silica and extracted with CO2. An in-line silica column was used to remove co-extracted oligomers 76 Poly(isoprene) Carbon tetrachloride SFE Optimum extraction conditions 60 °C at 21 MPa, CO2 only.Increasing the temperature caused softening and agglomeration of rubber particles. Increasing the pressure lowered the softening temperature of the rubber and therefore also caused the particles to agglomerate 66 Poly(styrene) Ethylbenzene SFE CO2 only, temperature from 50 to 100 °C, pressure: 7–12 MPa. Supercritical CO2 swelled the polymer, lowered Tg and resulted in a 106-fold increase in diffusivity of ethylbenzene in PS.A model used to estimate diffusivity in the swelled polymer is presented 61 Poly(styrene) Alkylbenzenes SFE CO2 only. SFE was limited by diffusion within the polymer rather than solubility in the CO2. Fastest extractions therefore employ small particle size and either static or dynamic extraction 68 Poly(styrene) Styrene dimers and trimers SFE CO2 only. 9–11 MPa, 80–120 °C. On-line SFE–SFC was effective at identifying additives and oligomers.The rate of extraction increased with temperature 65 Poly(urethane) foams CFCs SFE SFE conditions: 50 °C at 11 MPa. Supercritical CO2 extracted much more than liquid CO2 and nitrogen. Addition of propane modifier seems to increase extraction rate slightly. Ground foam extracted much faster than un-ground foam 72 Poly(ethylene) Erucamide acid amide SFE On-line SFE-capillary SFC. CO2 only; temperature, 45 °C; pressure, 15 MPa; extraction time, 15 min. Sample size: 2.7 mg.Example chromatogram presented 58 Poly(ethylene) Irganox SFE On-line SFE–capillary SFC. CO2 only; temperature, 45 °C; pressure, 20 MPa; extraction time, 30 min. Sample size: 9 mg. Example chromatogram presented 58 Poly(ethylene) Chimassorb, Tinuvin 144, Irganox 1098, oxidised Nauguard 524, Irganox 1010, DSTDP, Irganox 1076 and Ionox 330 SFE and Soxhlet extraction SFE conditions: CO2 only; temperature, 40 °C; pressure, 60 MPa; extraction time, 30 min and 5 h for low- and high-density poly- (ethylene) samples, respectively.Sample size: 0.2 g. Soxhlet: 1 g sample extracted with 100 ml of toluene for 3 h. After cooling, 20 ml of ethanol added to precipitate low molecular weight polymer. Extract solutions filtered, dried and reconstituted in CH2Cl2 prior to analysis. Similar mass spectra obtained by both extraction techniques 67 Poly(ethylene) Additives SFE Conditions for SFE: CO2 only; temperature, 65 °C; pressure, 15 MPa; extraction time, 10 min.Sample size: 3 mg of film 102 Poly(ethylene) Irganox 1010, Irganox 1076, Irgafos 168 SFE and Soxhlet extraction Conditions for SFE: temperature, 50–80 °C; pressure, 15–30 MPa, CO2 and benzene-modified CO2. Results at 80 °C and 30 MPa generally fastest, with amount extracted sometimes exceeding that from Soxhlet extraction. Benzene modifier increased extraction rates, particularly at lower temperatures 63 Poly(ethylene) Nicotine SFE SFE conditions: pressure, 21 MPa; temperature, 60 °C; CO2 modified with triethylamine; extraction time, 20 min 73 Table continued on next page 112R Analyst, September 1997, Vol. 122Table 3 continued Extraction Polymer Additive technique Comments Ref. Poly(ethylene) (low-density) SFE On-line SFE–SFC analysis. SFE conditions: temperature, 100 °C; pressure, 45 MPa. Typical extraction time 20 min for ground polymer samples (30–50 mesh) 74 Poly(ethylene) (low-density) Paraffins and olefins SFE Screening method for in vitro cytotoxicity testing 77 Poly(ethylene) and poly(propylene) Irganox 1010, BHT, erucamide, Tinuvin 770, Irgafos 168, Isonox 129 and DLTDP SFE On-line SFE–SFC.SFE conditions: CO2 only; temperature, 50 °C; pressure, 14 and 43 MPa; extraction time, 15 and 30 min. Accurate quantification and high extraction efficiency reported. 33 Poly(ethylene), poly(styrene) BHT, BHEB, Isonox 129, Irganox 1076, Irganox 1010, Irgafos 168, Cyasorb 3346, Cyanox 1790, stearyl stearamide, HBCD and erucamide SFE On-line SFE–SFE. CO2 only; temperature, 150 °C; pressure, 46 MPa; extraction time, 30 min.Typical recoveries ranged from 86 to 108% 55 Polymer Misoprostol SFE Formic acid modified CO2 used to extract misoprostol steroid drug from polymer to which it is covalently bonded. Formic acid hydrolyses covalent bond 62 Poly(propylene) BHT, Tinuvin 326, Seenox DM and Irganox 1010 SFE and liquid solvent extraction On-line SFE–capillary SFC. CO2 only; temperature, 30–90 °C; pressure, 30 MPa; extraction time, 30 min.Qualitative results only. Selectivity of extraction investigated 56 Poly(propylene) Apple aroma compounds SFE and liquid solvent extraction Conditions for SFE: CO2 only; temperature, 10–70 °C; pressure, 6–12 MPa; extraction time, 5–20 min. Sample size: 10 mg of film. Analysis using on-line GC. Results for SFE compare favourably (96–105% recovery) with those obtained using liquid solvent (dichloromethane) extraction 94 Poly(propylene) Irganox 1010, Irganox 1076, total extractables SFE 40 °C, 35 MPa, 10 min, CO2 only, 30 mm film.SFE extracted more material than 4 h Soxhlet extraction with dichloromethane 69 Poly(propylene) Irganox 1076, Irgafos 168, Hostanox SE- 02 SFE SFE conditions: temperature, 60 °C; CO2 density, 0.85 g ml21. Optimum extraction used two static extractions with hexane and methanol modifiers 71 Poly(propylene), poly(ethylene) Cyasorb UV 531, Topanol OC, Irganox 1330, Irganox 1010, Irganox 1076 SFE SFE conditions: 25.6 MPa, 35 °C, CO2 only. 2 h extraction was required for complete extraction 31 Poly(urethane) foams Amgard TCEP, Amgard TMCP, Amgard V6 and Thermolin 101 SFE On-line SFE–SFC. SFE conditions: CO2 only; temperature, 60 °C; pressure, 30 MPa; extraction time, 5 min. Good recoveries obtained compared with solvent extraction 75 Poly(urethane) Oligomers, TPP, BHT, Irganox 1010, Irganox 1076 SFE Screening method for in vitro cytotoxicity testing 77 Poly(urethanes) [Pellethane, estane, and 90-N poly(urethane) Ellastolan] BHT, residual ethers, plasticisers (either an adipate acid or a phthalate ester) SFE On-line SFE–SFC–MS.Conditions for SFE: CO2 only; temperature, 60 °C; pressure, 30 MPa; extraction time, 10 min. Feasibility study 78 Microwave-assisted extraction (MAE) and ASE— Poly(propylene) and poly(ethylene) (powdered) Irganox 1010, Irgafos 168, Chimassorb 81 MAE Hexane–acetone (1 + 1)—max. 6 min. 1,1,1-Trichloroethane—max. 3 min. Microwave power, 70%; solvent volume, 30 ml for both 82 HDPE, LDPE, PP Irganox 1010, Irgafox 168, Cyasorb UV 531, BHT Microwave heating, atmospheric pressure Cyclohexane–propan-2-ol (1 + 1) found to be the best solvent for extraction from poly(olefins). 20 min extraction time typical 83 Poly(ethylene terephthalate) Cyclic trimer MAE and Soxhlet extraction MAE conditions: sample size, 8 g of film; solvent, xylene + water or dichloromethane (40 ml); extraction time, 30–120 min; temperature, 70–120 °C.Soxhlet conditions: sample size, 15–20 g of film; solvent, xylene (190 ml); extraction time, 24 h. Good agreement between both extraction methods 84 PBT Dimers and trimers ASE Freeze-ground samples extracted with hexane at temperature from 50 to 70 °C. Extraction rate increased with temperature. Pressure had no effect other than to keep solvent liquid. Extraction faster than SFE with CO2 alone 89 Nylon-6 Caprolactam ASE Freeze-ground samples extracted with hexane at temperatures from 50 to 170 °C.Extraction rate increased with temperature. Pressure had no effect other than to keep solvent liquid. Extraction faster than with SFE with CO2 alone, but very similar to CO2 with methanol modifier 89 Analyst, September 1997, Vol. 122 113Rsuch long extractions and subsequent concentration steps there is a possibility of losses of volatile or thermally labile components. The selection of extraction solvent can make a large difference to the extraction time.However, the equipment is inexpensive, and once set up requires little ‘hands-on’ attention. SFE has been shown to provide a much faster extraction method than Soxhlet extraction. In most cases rapid and quantitative extraction has been achieved using ground samples or thin films. The method needs to be optimised for pressure, temperature and modifier. Equipment exists which can extract several samples simultaneously and which can be programmed to extract up to 28 samples consecutively.Microwave-assisted extraction offers a rapid method for the extraction of up to 12 samples simultaneously. The equipment is also much more expensive than that for Soxhlet extraction, and there are relatively few publications concerning this method. ASE has been found to be effective in extraction from environmental samples and has great potential for extraction from polymers. The main problem is likely to be selection of extraction solvent which does not dissolve the polymer at high temperature.Although ASE uses liquid solvents, the total solvent usage may not be higher than through the use of modifiers with SFE. In conclusion, Soxhlet extraction is inexpensive and well established, but time consuming. Of the newer methods, only SFE has been available long enough for large amounts of published evidence to become available. Amongst the applications of analytical SFE, extractions of polymers have been the most successful, although a few cases are problematical.Other newer methods show promise, but cannot yet be assessed from a wide range of published data. References 1 EEC 89/109, Off. J. Eur. Comm., 1989, L40, 38. 2 EEC 90/128, Off. J. Eur. Comm., 1990, L349, 26. 3 EEC 92/39, Off. J. Eur. Comm., 1992 , L168, 21. 4 Schilling, F. C., and Kuck, V. J., Polym. Degradation Stab., 1991, 31, 141. 5 Brauer, B., Funke, T., and Schulenbergschell, H., Dtsch. Lebensm. Rundsch., 1995, 91, 381. 6 Wright, S. J., Dale, M. J., Langridge-Smith, P. R. R., Zhan, Q., and Zenobi, R., Anal. Chem., 1966, 68, 3585. 7 Crompton, T. R., The Analysis of Plastics, Pergamon Press, Oxford, 1984. 8 Haslam, J., Willis, H. A., and Squirrel, D. C. M., Identification and Analysis of Plastics, Iliffe Books, London, 2nd edn., 1977. 9 Wheeler, D. A., Talanta, 1968, 15, 1315. 10 Cotton, N. J., PhD Thesis, University of Leeds, 1992 11 British Standard 2782, Part 4, Method 405D, 1965. 12 Schabron, J. F., and Fenska, L.E., Anal. Chem., 1980, 52, 1411. 13 Freitag, W., Fresenius’ Z. Anal. Chem., 1983, 316, 495. 14 Newton, I. D., in Polymer Characterisation, ed. Hunt, B. J., and James, M. I., Blackie, Glasgow, 1993, pp. 8–36. 15 Barns, K. A., Damant, A. P., Startin, J. R., and Castle, L., J. Chromatogr. A, 1995, 712, 191. 16 Komolprasert, V., Lawson, A. R., and Hargraves, W. A., J. Agric. Food Chem., 1995, 43, 1963. 17 Spell, H. L., and Eddy, R. D., Anal. Chem., 1960, 32, 1811. 18 Crompton, T. R., Eur. Polym. J., 1968, 4, 473. 19 Majors, R. E., J. Chromatog. Sci., 1970, 8, 339. 20 Haney, M. A., and Dark, W. A., J. Chromatogr. Sci., 1980, 18, 655. 21 Perlstein, P., Anal. Chim. Acta, 1983, 149, 21 22 Wims, A. M., and Swarin, S. J., J. Appl. Polym. Sci., 1975, 19, 1243. 23 Majors, R. E., LC–GC, 1995, 13, 82. 24 Majors, R. E., LC–GC, 1996, 14, 88. 25 Chester, T. L., Pinkston, J. D., and Raynie, D. E., Anal. Chem., 1992, 64, 153R. 26 Chester, T. L., Pinkston, J. D., and Raynie, D. E., Anal. Chem., 1994, 66, 106R. 27 Levy, J. M., HRC–J. High Resolut. Chromatogr., 1994, 17, 212. 28 Camel, V., Tambute, A., and Caude, M., Analusis, 1992, 20, 503. 29 Smith, C. G., Smith, P. B., Pasztor, A. J., Mckelvy, M. L., Meunier, D. M., Froelicher, S. W., and Ellaboudy, A. S., Anal. Chem., 1993, 65, R 217. 30 Anton, K., Menes, R., and Widmer, H. M., Chromatographia, 1988, 26, 221. 31 Hirata, Y., and Okamoto, Y., J. Microcol. Sep., 1989, 1, 46. 32 Cotton, N. J., Bartle, K. D., Clifford, A. A., Ashraf, S., Moulder, R., and Dowle, C. J., HRC–J. High Resolut. Chromatogr., 1991, 14, 165. 33 Ryan, T. W., Yocklovich, S. G., Watkins, J. C., and Levy, E. J., J. Chromatogr., 1990, 505, 273. 34 Bartle, K. D., Clifford, A. A., Hawthorne, S. B., Lagenfield, J. J., Miller, D. J., and Robinson, R., J. Supercrit. Fluids, 1990, 3, 143. 35 Bartle, K. D., Boddington, T., Clifford, A. A., Cotton, N. J., and Dowle, C. J., Anal. Chem., 1991, 63, 2371. 36 Ashby, R., Food Addit. Contam., 1988, 5, 485. 37 Bartle, K. D., Boddington, T., Clifford, A. A., and Hawthorne, S. B., J. Supercrit. Fluids, 1992, 5, 207. 38 Clifford, A. A., Bartle, K. D., and Zhu, S. A., Anal. Proc., 1995, 32, 227. 39 Howdle, S. M., Ramsay, J. M., and Cooper, A. I., J. Polym. Sci. B: Polym. Phys., 1994, 32, 541. 40 Fahmy, T. M., Paulitis, M. E., Johnson, D. M., and McNally, M. E. P., Anal. Chem., 1993, 65, 1462. 41 Shieh, Y. T., Su, J. H., Manivannan, G., Lee, P. H. C., Sawan, S. P., and Spall, W. D., J. Appl. Polym. Sci., 1996, 59, 695. 42 Shieh, Y. T., Su, J. H., Manivannan, G., Lee, P. H. C., Sawan, S. P., and Spall, W. D., J. Appl. Polym. Sci., 1996, 59, 707. 43 Chiou, J. S., Barlow, J. W., and Paul, D. R., J. Appl. Polym. Sci., 1985, 30, 3911. 44 Michaels, A. S., and Bixler, H. J., J. Polym. Sci., 1961, L, 393. 45 Kazarian, S. G., Vincent, M. F., Bright, F. V., Liotta, C. L., and Eckert, C. A., J. Am. Chem. Soc., 1996, 118, 1729. 46 Kalospiros, N. S., and Paulaitis, M. E., Chem. Eng. Sci., 1994, 49, 659. 47 Handa, Y. P., Roovers, J., and Wang, F., Macromolecules, 1994, 27, 5511. 48 Condo, P. D., and Johnston, K. P., J. Polym. Sci. B: Polym. Phys., 1994, 32, 523. 49 Condo, P. D., Paul, D. R., and Johnston, K. P., Macromolecules, 1994, 27, 365. 50 Kuppers, S., Chromatographia, 1992, 33, 434. 51 Cotton, N. J., Bartle, K. D., Clifford, A. A., and Dowle, C. J., J. Chromatogr. Sci., 1993, 31, 157. 52 Schmidt, S., Blomberg, L., and Wannnman, T., Chromatographia, 1989, 28, 400. 53 Hunt, T. P., and Dowle, C. J., Analyst, 1991, 116, 1299. 54 Marin, M. L., Jiminez, A., Lopez, J., and Vilaplana, J., J. Chromatogr. A, 1996, 750, 183. 55 Ashraf-Khorassani, M., Boyer, D. S., and Levy, J. M., J. Chromatogr. Sci., 1991, 29, 517. 56 Daimon, H., and Hirata, Y., Chromatographia, 1991, 32, 549. 57 Mackay, G. A., and Smith, R. M., Analyst, 1993, 118, 741. 58 Engelhardt, H., Zapp, J., and Kolla, P., Chromatographia, 1991, 32, 527. 59 Cotton, N. J., Bartle, K. D., Clifford, A. A., and Dowle, C. J., J. Appl. Polym. Sci., 1993, 48, 1607. 60 Oudsema, J. W., and Poole, C. F., HRC–J. High Resolut. Chromatogr., 1993, 16, 198. 61 Dooley, K. M., Launey, D., Becnel, J. M., and Caines, T., ACS Symp. Ser., 1995, 608, 269. 62 Roston, D. A., Sun, J. J., Collins, P. W., Perkins, W. E., and Tremont, S. J., J. Pharm. Biomed. Anal., 1995, 13, 1513. 63 Lou, X. W., Janssen, H. G., and Cramers, C. A., J. Microcol. Sep., 1995, 7, 303. 64 Lou, X. W., Janssen, H. G., and Cramers, C. A., J. Chromatogr. Sci., 1996, 34, 282. 65 Jordan, S. L., Taylor, L. T., Seemuth, P. D., and Miller, R. J., Text. Chem. Color., 1997, 29, 25. 66 Burgess, A. N., and Jackson, K., J. Appl. Polym. Sci., 1992, 46, 1395. 67 Juo, C. G., Chen, S. W., and Her, G. R., Anal. Chim. Acta, 1995, 311, 153. 68 Hawthorne, S. B., Galy, A. B., Schmitt, V. O., and Miller, D. J., Anal. Chem., 1995, 67, 2723. 114R Analyst, September 1997, Vol. 12269 Baner, L., Bucherl, T., Ewender, J., and Franz, R., J. Supercrit. Fluids, 1992, 5, 213. 70 Venema, A., Vandeven, H. J. F. M., David, F., and Sandra, P., HRC– J. High Resolut. Chromatogr., 1993, 16, 522. 71 Garde, J. A., Galotto, J., Catala, R., and Gavara, R., presented at the ILSI Symposium on Food Packaging: Ensuring the Quality and Safety of Foods, September 11–13, 1996, Budapest, Hungary. 72 Filardo, G., Galia, A., Gambino, S., Silvestri, G., and Poidomani, M., J. Supercrit. Fluids, 1996, 9, 234. 73 Cedergren, L., Fors, S., and Jonn, S., presented at the 3rd European Symposium on Analytical SFC and SFE organized in conjunction with the 6th International Symposium on SFC and SFE, September 6–8, 1995, Uppsala, Sweden. 74 Ashraf-Khorassani, M., and Levy, J. M., HRC–J. High Resolut. Chromatogr., 1990, 13, 742. 75 Mackay, G. A., and Smith, R. M., HRC–J. High Resolut. Chromatogr., 1995, 18, 607. 76 Hunt, T. P., Dowle, C. J., and Greenway, G., Analyst, 1993, 118, 17. 77 Braybrook, J. H., and Mackay, G. A., Polym. Int., 1992, 27, 157. 78 Mackay, G. A., and Smith, R. M., J. Chromatogr. Sci., 1994, 32, 455. 79 Zlotorzynski, A., Crit. Rev. Anal. Chem., 1995, 25, 43. 80 Smith, F. E., and Arsenault, E. A., Talanta, 1996, 43, 1207. 81 Renoe, B. W., Am. Lab., 1994, August, 34. 82 Freitag, W., and John, O., Angew. Makromol. Chem., 1990, 175, 181. 83 Nielson, R. C., J. Liq. Chromatogr., 1991, 14, 503. 84 Costley, C. T., Dean, J. R., Newton, I., and Carroll, J., Anal. Commun., 1997, 34, 89. 85 Brandt, H. J., Anal. Chem., 1961, 33, 1390. 86 Nielson, R. C., J. Liq. Chromatogr., 1993, 16, 1625. 87 Caceres, A., Ysambert, F., Lopez, J., and Marquez, N., Sep. Sci. Technol., 1996, 31, 2287. 88 Nerin, C., Salafranca, J., and Cacho, J., Food Addit. Contam., 1996, 13, 243. 89 Lou, X., Janssen, H., and Cramers, C. A., Anal. Chem., 1997, 69, 1598. 90 Staal, W. J., Cools, P., Vanherk, A. M., and German, A. L., Chromatographia, 1993, 37, 218. 91 Nerin, C., Salafranca, J., Cacho, J., and Rubio, C., J. Chromatogr. A, 1995, 690, 230. 92 Macko, T., Siegl, R., and Lederer, K., Angew. Makromol. Chem., 1995, 227, 179. 93 Macko, T., Furtner, B., and Lederer, K., J. Appl. Polym. Sci., 1996, 62, 2201. 94 Nielson, T. J., Jagerstad, I. M., Oste, R. E., and Sivik, B. T. G., J. Agric. Food Chem., 1991, 39, 1234. 95 Sevini, F., and Marcato, B., J. Chromatogr. A, 1983, 260, 507. 96 Lehotay, J., Danecek, J., Liska, O., Lesko, O. J., and Brandsteterova, E., J. Appl. Polym. Sci., 1980, 25, 1943. 97 Majors, R. E., and Johnson, E. L., J. Chromatogr. Sci., 1978, 167, 17. 98 Raynor, M. W., Bartle, K. D., Davies, I. L., Williams, A., Clifford, A. A., Chalmers, J. M., and Cook, B. W., Anal. Chem., 1988, 60, 427. 99 Vargo, J. D., and Olson, K. L., Anal. Chem., 1985, 57, 672. 100 Gasslander, U., and Jaegfeldt, H., Anal. Chim. Acta, 1984, 166, 243. 101 Kumur, T., Analyst, 1990, 115, 1319. 102 Hirata, Y., Nakata, F., and Horihata, M., J. High Resolut. Chromatogr., 1988, 11, 81. Paper 7/04052K Received June 10, 1997 Accepted July 7, 1997 Analyst, September 1997, Vol. 122 115R
ISSN:0003-2654
DOI:10.1039/a704052k
出版商:RSC
年代:1997
数据来源: RSC
|
2. |
Evaluation of Parallel Factor Analysis for the Resolution ofKinetic Data by Diode-array High-performance Liquid Chromatography |
|
Analyst,
Volume 122,
Issue 9,
1997,
Page 871-877
Peter Hindmarch,
Preview
|
|
摘要:
Evaluation of Parallel Factor Analysis for the Resolution of Kinetic Data by Diode-array High-performance Liquid Chromatography Peter Hindmarch, Keyhandokht Kavianpour and Richard G. Brereton* School of Chemistry, University of Bristol, Cantock’s Close, Bristol, BS8 1TS, UK The PARAFAC algorithm for factor analysis of three or higher way datasets is summarised. A series of simulations of kinetic profiles of two-way diode-array HPLC data is described. A three-phase reaction system of reactant, intermediate and product is used to illustrate the method, each closely eluting and with similar spectra based on experimental HPLC with diode-array detection of chlorophyll degradation products.A kinetic parameter is varied to change the relative concentration of the intermediate in each series of simulations. Several indices of quality of reconstruction are introduced. It is concluded that the number of factors used to model the data is crucial to the quality of reconstruction.A good approach is first to use fewer factors than are expected, then increasing the number until each elution profile shows a single maximum. Keywords: Deconvolution; PARAFAC; high-performance liquid chromatography; kinetics; chlorophyll Three-way data are common in analytical chemistry.1,2 An example is a series of chromatograms recorded in time. If these chromatograms consist, in turn, of two-way data such as in HPLC with diode-array detection (DAD) or GC–MS, the full series of chromatograms may be regarded as a three-way dataset.One mode is time or sample number, whereas the other modes are elution time and a spectroscopic parameter, such as wavelength or mass number. Conventionally, each chromatogram is analysed independently by factor analysis or multivariate calibration, but this ignores the fact that there are components common to the entire series of chromatograms with similar spectra and elution profiles. Treating the entire dataset as one three-dimensional block provides more information than treating each chromatogram separately.There are several methods for three-dimensional factor analysis,3–7 and it is the purpose of this paper to evaluate one of the most common, called PARAFAC. In this approach, the three-dimensional data are decomposed into a series of factors, each relating to one of the three physical variables. Theory PARAFAC (parallel factor analysis) is a method of decomposing a three-way data array, or tensor, into a series of two-way arrays.The original algorithms were developed by psychometricians for the decomposition of multiblock data.8–11 Mathematically, PARAFAC can be seen as a simplification of the Tucker3 Model proposed by Tucker,12 in which a three-way I3 J 3 K array is decomposed into three loadings matrices A(I 3 L), B(J 3 M) and C(K 3 N), where L, M and N are the number of factors in the first, second and third modes, respectively. I, J, and K may be regarded as the number of samples, elution times and wavelengths, respectively.In most areas of chemistry, L, M and N will be equal and are the number of detectable components in a mixture, making the chemometric problem simpler than the psychometric problem. In this case, a three-way array (or tensor) X, whose dimensions are sample number, elution time and spectral wavelength in the case of HPLC–DAD, is decomposed into three matrices A, B and C such that, xi, j,k = ai, f bj, f 1 F å ck, f + ei, j,k (1) where F is the number of factors used in the model and e is the error term.A is a matrix of I rows consisting of sample numbers and L columns consisting of the number of detectable components in the mixture; B and C correspond to the elution profiles and spectra of these L components. This model can also be written as X = af 1 F å Ä bfÄ cf + E (2) where # represents the ternary tensor product of the three vectors. The field of tensor algebra as applied to chemical data is discussed extensively elsewhere.13 The definition and representation of tensor products varies depending on the context, but it is sufficient here to state that the tensor product of vectors I,1a, J,1b and K,1c is a vector with co-ordinates xi,yj,zk.Combining all of these vectors over all factors gives the threeway data matrix. Graphically, the PARAFAC model for a threeway, two-component system is shown in Fig. 1. The simplest way of implementing a PARAFAC model is by alternating least squares.Starting with a known three-way matrix, X, and two randomly initiated loadings matrices, A and B, the third loadings matrix, C, can be estimated. Then, from this new estimate of C, A and B then C can be successively estimated and so on until there is convergence in the model. The advantage of this approach is that, apart from the number of factors, F, no prior knowledge of the system is required. Furthermore, apart from scaling considerations, PARAFAC produces a unique solution.However, as it is a numeric rather than an analytical method, care must be taken to ensure that the algorithm converges properly and operates at an acceptable speed. The theory of the PARAFAC algorithm is discussed further elsewhere,14–16 as also are further applications to calibration.17–20 Practical applications, however, have been limited.21–23 Method Experimental Spectra of three chlorophyll degradation products were obtained experimentally from a Waters (Milford, MA, USA) Fig. 1 Graphical representation of the PARAFAC method. Analyst, September 1997, Vol. 122 (871–877) 871Model 990 HPLC–DAD system. For chlorophyll degradation mixtures, closely eluting compounds often possess very similar spectral characteristics, posing real problems in resolution by chemometric means. Spectra were recorded between 350 and 800 nm at 2 nm intervals. A common problem involves detecting crucial intermediates that are present in low concentrations, e.g., in the reaction A ? B ? C, where the first reaction is slow and the second is fast.A full kinetic model for the pathway requires the detection of B, which may be present in small amounts, dependent on the relative rates of the two reactions. If HPLC is employed to study such reactions, the intermediate, which may be a stereoisomer of the parent compound, could have very similar elution profiles and spectra to one of the other components.This situation is well established in the study of the degradation of chlorophyll by HPLC . Simulation Design The data were designed to simulate the degradation of chlorophyll-a products investigated elsewhere.24 Each threeway dataset is constructed from modes representing elution profiles, spectra and degradation profiles of each component. Three spectra obtained experimentally were used to represent three compounds with very similar spectral characteristics.The elution profiles were simulated and the degradation profiles represent a series of HPLC–DAD traces used to monitor a threereactant system. The three species formed are assumed to form part of a reaction series, where the reactant (R) is converted into an intermediate (I), which is then converted into a product (P). Elution Profiles The elution data matrix is represented by 20,3A, representing 20 chromatographic points in time. The chromatographic profile of compound l (where l = 1, 2 and 3 for the reactant, intermediate and product, respectively) at time I is given by al,i.Each column of A is an individual elution profile represented in this simulation by Gaussians centred at points 10, 14 and 6 in time and given by ai,1 = e - (i-10)2 6 (5) ai,2 = e - (i-14)2 6 (6) a e i i , ( ) 3 6 6 2 = - - (7) The data are designed so that the product elutes first, followed by the reactant and then the intermediate. The elution profiles are shown graphically in Fig. 2. Spectra Spectra were chosen from previous work to represent each component. Brereton and co-workers24,25 have shown that in chlorophyll degradation studies the ratio of absorbances between the two absorbance maxima, at approximately 430 and 665 nm, respectively, is very diagnostic. These spectra were chosen so that the reactant and the intermediate had spectra with similar features. Table 1 gives the spectral characteristics of each component. The spectra, shown in Fig. 3, were scaled to constant maximum absorbance and stored as 226,3B, where each column represents a spectrum taken between 350 and 800 nm, digitised at a resolution of 2 nm, i.e., 226 readings per spectrum. Degradation profiles The degradation profiles were designed to represent a reactant decreasing in concentration as the experiment proceeds, a minor intermediate increasing to a maximum, then decreasing and a product increasing in concentration. Concentrations are such that at any time the total concentration of all species is constant.The degradation data matrix 20,3C represents the concentration of the three components at 20 sampling points throughout the experiment where the three columns represent the reactant, intermediate and product respectively. The profiles for each component are given by ck,1 = 5e2k/8 (8) ck,2 = (5 2 5e2k/8)e2k/d (9) ck,3 = 5 2 5e2k/8 + e2k/d 2 5e2(k/8 + k/d) (10) where k is the sample number arranged sequentially in time and d is a parameter varying according to the relative significance and kinetic stability of the intermediate. The greater the value of d the slower is the decomposition of the intermediate, and so the easier it is to detect.Different simulations were performed at different values of d and simulated degradation profiles are shown in Fig. 4. Fig. 2 Simulated elution profiles. 8, Reactant; 3, intermediate; and 5, product. Table 1 Position of maxima of designed and predicted elution profiles. Elution maximum Reactant Intermediate Product 10 14 6 Rate parameter, d Factor 1 Factor 2 Factor 3 1 10 10 6 3 10 10 6 5 10 6 14 10 6 10 14 20 14 6 10 Fig. 3 Experimentally obtained spectra used in the simulations.Solid line, reactant; dotted line, intermediate; and dashed line, product. 872 Analyst, September 1997, Vol. 122Formation of three-way data set The three-way, three-component model 20,226,20 X is formed by xi, j,k = ai, f bj, f ck, f f =1 F å (11) where f is the component number and F is the number of components, which in this study is three.Five datasets were created in which the rate constant, d, was 1, 3, 5, 10 and 20. Application of PARAFAC Algorithm No pre-processing is performed on the data in this paper. The issue of pre-processing of three-way arrays is more complex than for the two-way case. Centring can be performed across either one, two or three of the modes and can distort the trilinear model. The order of any pre-processing is also critical.These issues have been discussed elsewhere.9,15,26,27 The datasets were decomposed by the PARAFAC written in Matlab 4.2 (Mathworks, Natick, MA, USA). The algorithm was used to extract three factor matrices from each simulated dataset initialised using random vectors and a convergence limit of 1 3 1026 between successive estimates of the sum of squares of the misfit. Indicators of Quality of Reconstruction Various functions can be used to compare the results from the simulations with the design data.Component sum of squares This gives a measure of the size of each component and factor, which aids in the identification of factors and gives an indication of their purity. The size of each predicted component, �S f, is given by �S f = ( �a i, f � bj, f � ck, f )2 k=1 K å j=1 J å i=1 I å (12) The values obtained for the predicted model using eqn. (12) can be compared to the size of the true components, Sf, calculated in the same manner as above, but with the estimated vectors replaced by their true equivalents.The square root of the ratio of the estimated to the true sum of squares, Qf, is given by Qf = �S f / Sf (13) The closer this value is to unity, the better is the modelling of the factor f. The concentration of the reactant will decrease identically from sample to sample in each simulation. The relative concentration of the intermediate will increase with increasing d and the product will decrease. The total concentration of the reactant, intermediate and product at any one point will always be constant, but the sum of squares will not.Regression For each of the elution, spectral and degradation modes, the predicted data are regressed on to the real data. In each case a matrix R can be obtained, often called a rotation or transformation matrix. For example, for the elution data, if A is the true elution datum, �A is the predicted data and RA is the rotation matrix, then �A = ARA (14) RA is found by the pseudo-inverse: RA = (AAA)21 AA �A (15) For a good model, each column and row of the rotation matrix contains only one value significantly greater than zero.Using the rotation matrices obtained above and the rotation matrix, a ‘predicted true’ dataset can be obtained, denoted by a circle overscript, e.g. for the elution data °A = � AR21 A (16) Calculating a residual root mean sum of squares RMSEP(A) between the actual true and predicted true data, across each matrix, gives a further indication of the quality of regression: RMSEP( ) ( ° � ) , , A a a I F i f i f f F i I = - � = = å å 2 1 1 (17) Spectral characteristics and elution maxima A simple measure of the success of the decomposition can obtained by comparing the predicted design parameters with those listed in Table 1.From the predicted elution data the position of the maximum of each component can be obtained as the maximum of each column of �A .Similarly, the positions of the two absorbance maxima and their ratio can be computed from the estimated spectral data. Results Most methods for factor analysis depend first on determining the number of significant factors. This is particularly true when the aim is to model the entire dataset. The importance of detecting and modelling all significant components in two-way factor analysis has been discussed in the context of mid-infrared (MIR) spectrometry.28,29 If a third significant factor is ignored, then the information from this compound is mixed with the other two compounds.In contrast, if a third factor is small it may become confused with the other two factors if a threefactor model is employed. PARAFAC depends crucially on a prior estimate of the number of significant factors as shown below. The following section reports the results assuming a three component mixture and the subsequent section a two component mixture.Another important aspect is that the order in which the factors are extracted may differ according to how the algorithm is implemented, e.g., the starting point of the iterations. This Fig. 4 Simulated degradation profiles. 8, Reactant; 3, intermediate; and 5, product. Analyst, September 1997, Vol. 122 873means that, over a series of datasets, the first factor may correspond to physically different compounds in each run, so it is first necessary to reorder the factors according to presumed physical significance.In some cases, where the interpretation of each factor is in doubt, this can be difficult. In the tables, the factors are ordered according to the order in which they were extracted. Three-factor Systems The predicted maxima positions of the elution profiles, in terms of elution index for the three component, three-factor system, are given in Table 2. At the levels where d, the kinetic rate parameter, is low, i.e., 1 and 3, the PARAFAC algorithm fails to position the three components correctly, whereas at higher levels of d all three components are correctly determined.Table 2 lists the spectral parameters determined from the predicted data. In all cases the product (factor 3 at d = 1 and 3, factor 2 at d = 5 and 20 and factor 1 at d = 10) is predicted well, with a peak ratio of 1.65, and a low-wavelength absorbance maximum at 424 or 426 nm. There is a slight problem with predicting the high-wavelength absorbance maximum at higher values of d, presumably because the prediction ability decreases as the amount of intermediate increases.However, the highwavelength absorption maximum is always @660 nm. A 4 nm shift in position represents only two sampling points in the wavelength direction. In this study, the data were designed with very similar spectral algorithm has successively determined these, so it is trivial to establish the correspondence between components and factors.However, this will not always be true and in cases where there are several very similar components, a confident identification of the factors based on spectral parameters may not be possible. The component sum of squares for the true and predicted data are given in Table 3. At all levels the size of the product is predicted remarkably well. The reactant and intermediate, however, are only closely estimated at the two higher levels of d.This can be understood by considering that the reactant and product had similar spectral characteristics. At the lower levels of d, the intermediate is relatively minor compared with the reactant, but as the intermediate increases in significance at higher levels of d, it is easier for the algorithm to distinguish between them. The elution profiles for d = 1 and 20 are presented graphically in Fig. 5. It is obvious that the product and intermediate are not distinguished when d is low; these two Table 2 Design and predicted spectral parameters for the three factor model.Design Reactant Intermediate Product Spectral ratio 1.28 1.24 1.65 Absorbance max. 1/nm 434 434 424 Absorbance max. 2/nm 670 670 660 d = 1— Factor 1 Factor 2 Factor 3 Spectrum ratio 1.29 1.28 1.65 Absorbance max. 1/nm 436 436 426 Absorbance max. 2/nm 670 670 660 d = 3— Spectrum ratio 1.30 1.27 1.65 Absorbance max. 1/nm 434 434 424 Absorbance max. 2/nm 670 670 660 d = 5— Spectrum ratio 1.28 1.65 1.24 Absorbance max. 1/nm 434 426 436 Absorbance max. 2/nm 670 660 670 d = 10— Spectrum ratio 1.65 1.28 1.24 Absorbance max. 1/nm 424 424 434 Absorbance max. 2/nm 658 668 668 d = 20— Spectrum ratio 1.24 1.65 1.28 Absorbance max. 1/nm 434 424 434 Absorbance max. 2/nm 668 656 668 Table 3 Size of each design component and factors for the two- and threecomponent systems Rate parameter, d 1 3 5 10 20 Reactant 9100 9100 9100 9100 9100 Intermediate 8 167 591 2570 7165 Product 14827 13416 11750 7637 3547 Three-component system— Factor 1 842 1825 8201 7640 6334 Factor 2 4741 3803 11751 8463 3547 Factor 3 14285 13412 572 2483 9461 Two-component system— Factor 1 9209 9872 11684 13111 5317 Factor 2 14312 13473 10795 7549 15294 Table 4 Prediction ratios, Qf, for the two- and three-component models Ratio Qf Three-component model Two-component model d Reactant Intermediate Product Reactant Product 1 0.9999 10.2884 0.7218 1.0060 1.0009 3 0.9999 3.3054 0.6465 1.0416 1.0021 5 1.0000 0.9836 0.9493 0.9972 1.0892 10 1.0002 0.9829 0.9646 1.2003 0.9942 20 1.0197 0.9402 1.0001 n/a n/a Table 5 Root mean square error of prediction (RMSEP) for the three-factor models Data mode d A B C 1 1.1 3 1027 4.19 3 1027 3.5 3 1028 3 4.2 3 1028 2.89 3 1028 2.3 3 1028 5 1.4 3 1028 2.89 3 1027 3.5 3 1028 10 3.7 3 1028 2.3 3 1028 3.6 3 1028 20 3.37 3 1028 3.55 3 1027 2.63 3 1028 Fig. 5 Elution profiles obtained for the three factor models at (a) d = 1 and (b) d = 20. 874 Analyst, September 1997, Vol. 122species have similar spectral characteristics. Figs. 6(a) and (b) are representations of the corresponding spectra and it can be seen that they are recovered well. It can be concluded that when the number of components is correctly known, the PARAFAC algorithm produces excellent decomposition results. These results are the best when all of the components are relatively significant, as shown by the square root of the ratios of predicted to true sum of squares, Qf, given in Table 4, and the RMSEP in Table 5.For the elution data, A, improves considerably from the d = 1 to the d = 20 level. There is also an improvement, but to a lesser extent, for the spectral data. The error in the kinetic profiles is reasonably constant at each level. Two factor Systems The PARAFAC algorithm was repeated on the datasets but with two rather than three factors used to model the data. The elution and spectral parameters found are given in Tables 6 and 7, respectively. As can be seen from Table 6, at each level of d the algorithm appears to detect successfully the reactant and product without any interference from the intermediate.Note that the product should elute at datapoint 6 and the reactant at datapoint 10. Again, in Table 7, it appears that the two-component model produces good predictions of the spectrum ratios and absorbance maxima at each level of d, although the peak ratio for the product (1.56) is lower at d = 20.However, when the sum of squares of the factors and components are computed (Table 3), the situation is not so straightforward. At the lower two levels of the intermediate the Table 6 Position of maxima of predicted elution profiles for the two factor model Elution maximum d Factor 1 Factor 2 1 10 6 3 10 6 5 6 10 10 10 6 20 6 10 Fig. 6 Predicted spectra for (a) three factors at d = 1, (b) three factors at d = 2, (c) two factors at d = 1 and (d) two factors at d = 20.Fig. 7 Elution profiles obtained for the two factor models at (a) d = 1, (b) d = 3, (c) d = 5, (d) d = 10 and (e) d = 20. Analyst, September 1997, Vol. 122 875factors predict the size of the component reasonably well, but at the higher levels it becomes more difficult to distinguish the factors. This result reinforces the observation from above that a univariate measure such as elution maximum is not a sophisticated measure of data quality and a multivariate method, utilising the data from all available modes, should always be used in preference.This is important as the PARAFAC algorithm distributes all of the observed systematic variance between the factors in the model, so that these are not necessarily pure factors. Unlike the three-factor model above, predictions for the two factor model are better when the unmodelled component is relatively insignificant. As can be seen in Table 4, as the level of the intermediate increases the quality of reconstruction of the spectrally similar reactant decreases, but this is only observed when a multivariate measure such as the sum of squares or the rotation matrices is used.The quality of the product also falls but less significantly. Because at the d = 20 level a confident determination of the identification of the reactant and product cannot be made, the prediction ratio therefore cannot be calculated. The five recovered elution profiles are shown in Fig. 7. These supplement the data in Table 6; it is obvious that for d = 20 the first factor has, in fact, two clear maxima. Interestingly, the intermediate is confused with the product and predicted as one factor, despite the difference in both spectral characteristics and elution profiles. This unexpected result can be explained in terms of kinetic profiles; the level of intermediate builds up rapidly and then decreases with time, and so the kinetics of the two compounds are fairly similar.Two components with identical kinetics but different spectra and elution profiles could be modelled as a single factor. Visual inspection of the predicted chromatograms in Fig. 7(c)–(e) should provide clues that the number of predicted components is too few, and so lead to rerunning the model including further components. The spectra, Fig. 6(c) and (d), are recovered well again. Conclusions PARAFAC is a powerful approach for resolving out series for two-way chromatograms recorded over a number of samples.The methods can be extended to three-way or higher data, e.g., chromatograms could be recorded at different pH values and times; the change in chromatography with pH complements the change in intensity with time. The dataset in this paper is demanding, with the following properties. The middle chromatographic peak has no composition 1 or selective region, and most factor analysis methods find it difficult to resolve out unselective peaks. The spectra of the reactant and intermediate are almost identical, with similar spectra ratios and absorbance maxima, and also partially coelute.Approaches such as windows factor analysis and evolutionary factor analysis will not resolve out neighbouring peaks with very similar spectra; these will simply be modelled by one principal component. Even two-dimensional peak purity methods such as derivatives depend on change in spectral composition over elution time and simply would not detect a difference between the reactant and intermediate. By using PARAFAC on a series of chromatograms, these peaks can be distinguished provided that the concentration of the intermediate is not too low.Hence PARAFAC has potential as a major technique for the resolution and quantification of a series of two-way of chromatograms, often in cases where normal factor analysis methods will fail. The major drawback is that a good estimate of the number of components is required in advance for sensible models.If this is unknown, it is better to perform the models with fewer components first to see whether there are any elution profiles with more than one maximum. If so, the algorithm can be repeated, increasing the number of components until unimodal elution profiles are achieved. The authors thank R. Bro for providing the Matlab PARAFAC algorithm and EPSRC for providing financial support for this project. Appendix List of Notation Used i Elution time index I Total number of elution points (20) j Spectral wavelength index J Number of points in each spectrum (226) k Sample number K Number of reaction times (20) A Elution data matrix, with individual point al,i B Spectra data matrix, with individual point bl,j C Concentration data matrix, with individual point cl,k l Component number L Number of components d Kinetic rate parameter X Three-way data, with individual point xi,j,k f Factor number F Total number of factors Three-way matrices are represented by underlined uppercase bold italic characters, e.g., X, two-way matrices by uppercase bold italic characters, e.g., A, vectors by lower case bold italic characters, e.g., af, and scalars by non-bold characters.Estimated variables are denoted by a ‘hat,’ e.g., �A , except in eqns. (16) and (17), where the ‘estimated true’ data are represented by a circle overscript. Dimensions of matrices are given as left-hand side subscripts, e.g., 20,10A is a matrix of 20 rows by 10 columns.References 1 Geladi, P., Chemom. Intell. Lab. Syst., 1989, 7, 11. 2 Ståhle, L., Chemom. Intell. Lab. Syst., 1989, 7, 95. 3 Mitchell, B. C., and Burdick, D. S., Chemom. Intell. Lab. Syst., 1993, 20, 149. Table 7 Predicted spectral parameters for the two factor model. The design parameters are given in Table 1 d = 1— Factor 1 Factor2 Spectrum ratio 1.28 1.65 Absorbance max. 1/nm 436 426 Absorbance max. 2/nm 670 660 d = 3— Spectrum ratio 1.28 1.65 Absorbance max. 1/nm 436 426 Absorbance max. 2/nm 670 660 d = 5— Spectrum ratio 1.65 1.28 Absorbance max. 1/nm 426 436 Absorbance max. 2/nm 660 670 d = 10— Spectrum ratio 1.27 1.65 Absorbance max. 1/nm 436 426 Absorbance max. 2/nm 670 660 d = 20— Spectrum ratio 1.56 1.27 Absorbance max. 1/nm 426 436 Absorbance max. 2/nm 660 670 876 Analyst, September 1997, Vol. 1224 Wold, S., Geladi, P., Esbensen, K., and � Ohman, J., J. Chemom., 1987, 1, 41. 5 Kvalheim, O.M., and Grung, B., Chemom. Intell. Lab. Syst., 1995, 29, 213. 6 Smilde, A. K., and Doornbos, D. A., J. Chemom., 1991, 5, 345. 7 Tauler, R., Smilde, A. K., Kowalski, B., J. Chemom., 1995, 9, 31. 8 Cattell, R., Psychol. Bull., 1952, 49, 499. 9 Kruskal, J. B., Psychometrika, 1976, 41, 281. 10 Sands, R., and Young, F., Psychometrika, 1980, 45, 39. 11 Kroonberg, P. M., and de Leeuw, J., Psychometrika, 1980, 45, 69. 12 Tucker, L., in Problems of Measuring Change, ed. Harris, C., University of Wisconsin Press, Madison, WI, 1963, p. 122. 13 Burdick, D. S., Chemom. Intell. Lab. Syst., 1995, 28, 229. 14 Harshman, R. A., and Lundy, M. E., Comput. Stat. Data Anal., 1994, 18, 39. 15 Smilde, A. G., Chemom. Intell. Lab. Syst., 1992, 15, 143. 16 Henrion, R., Chemom. Intell. Lab. Syst., 1994, 25, 1. 17 Bro, R., and Heimdal, H., Chemom. Intell. Lab. Syst., 1996, 34, 85. 18 Smilde, A. K., J. Chemom., 1992, 6, 11. 19 Smilde, A. K., Van der Graaf, P. H., Doornbos, D. A., Steerneman, T., and Sleurink, A., Anal. Chim.Acta, 1990, 235, 41. 20 Bro, R., J. Chemom., 1996, 10, 47. 21 Booksh, K. S., Muroski, A. R., and Myrick, M. L., Anal. Chem., 1996, 68, 3539. 22 Bro, R., Chemom. Intell. Lab. Syst., 1996, 34, 85. 23 Smilde, A. K., Tauler, R., Henshaw, J. M., Burgess, L. W., and Kowalski, B. R., Anal. Chem., 1994, 66, 3345. 24 Brereton, R. G., Rahmani, A., Liang, Y., Z., and Kvalheim, O. M., Photochem. Photobiol., 1993, 57, 1048. 25 Elbergali, A. K., Brereton, R.G., and Rahmani A., Analyst, 1995, 120, 2207. 26 Harshman, R. A., and Lundy, M. E., in Research Methods for Multimode Data Analysis, ed. Law, H. G., Snyder, C. W., Hattie, J. A., and McDonald, R. P., Praeger, New York, 1984, p. 216. 27 Ten Berge, J. M. F., in Multiway Data Analyses, ed. Coppi, R., and Bolasco, S., Elsevier, Amsterdam, 1989, p. 53. 28 Gurden, S. P., Brereton. R. G., and Groves, J. A., Analyst, 1996, 121, 441. 29 Gurden, S. P., Brereton, R. G., and Groves, J.A., Chemom. Intell. Lab. Syst., 1994, 23, 123. Paper 7/02232H Received April 2, 1997 Accepted May 12, 1997 Analyst, September 1997, Vol. 122 877 Evaluation of Parallel Factor Analysis for the Resolution of Kinetic Data by Diode-array High-performance Liquid Chromatography Peter Hindmarch, Keyhandokht Kavianpour and Richard G. Brereton* School of Chemistry, University of Bristol, Cantock’s Close, Bristol, BS8 1TS, UK The PARAFAC algorithm for factor analysis of three or higher way datasets is summarised.A series of simulations of kinetic profiles of two-way diode-array HPLC data is described. A three-phase reaction system of reactant, intermediate and product is used to illustrate the method, each closely eluting and with similar spectra based on experimental HPLC with diode-array detection of chlorophyll degradation products. A kinetic parameter is varied to change the relative concentration of the intermediate in each series of simulations.Several indices of quality of reconstruction are introduced. It is concluded that the number of factors used to model the data is crucial to the quality of reconstruction. A good approach is first to use fewer factors than are expected, then increasing the number until each elution profile shows a single maximum. Keywords: Deconvolution; PARAFAC; high-performance liquid chromatography; kinetics; chlorophyll Three-way data are common in analytical chemistry.1,2 An example is a series of chromatograms recorded in time.If these chromatograms consist, in turn, of two-way data such as in HPLC with diode-array detection (DAD) or GC–MS, the full series of chromatograms may be regarded as a three-way dataset. One mode is time or sample number, whereas the other modes are elution time and a spectroscopic parameter, such as wavelength or mass number. Conventionally, each chromatogram is analysed independently by factor analysis or multivariate calibration, but this ignores the fact that there are components common to the entire series of chromatograms with similar spectra and elution profiles.Treating the entire dataset as one three-dimensional block provides more information than treating each chromatogram separately. There are several methods for three-dimensional factor analysis,3–7 and it is the purpose of this paper to evaluate one of the most common, called PARAFAC. In this approach, the three-dimensional data are decomposed into a series of factors, each relating to one of the three physical variables. Theory PARAFAC (parallel factor analysis) is a method of decomposing a three-way data array, or tensor, into a series of two-way arrays.The original algorithms were developed by psychometricians for the decomposition of multiblock data.8–11 Mathematically, PARAFAC can be seen as a simplification of the Tucker3 Model proposed by Tucker,12 in which a three-way I3 J 3 K array is decomposed into three loadings matrices A(I 3 L), B(J 3 M) and C(K 3 N), where L, M and N are the number of factors in the first, second and third modes, respectively. I, J, and K may be regarded as the number of samples, elution times and wavelengths, respectively.In most areas of chemistry, L, M and N wie equal and are the number of detectable components in a mixture, making the chemometric problem simpler than the psychometric problem. In this case, a three-way array (or tensor) X, whose dimensions are sample number, elution time and spectral wavelength in the case of HPLC–DAD, is decomposed into three matrices A, B and C such that, xi, j,k = ai, f bj, f 1 F å ck, f + ei, j,k (1) where F is the number of factors used in the model and e is the error term.A is a matrix of I rows consisting of sample numbers and L columns consisting of the number of detectable components in the mixture; B and C correspond to the elution profiles and spectra of these L components.This model can also be written as X = af 1 F å Ä bfÄ cf + E (2) where # represents the ternary tensor product of the three vectors. The field of tensor algebra as applied to chemical data is discussed extensively elsewhere.13 The definition and representation of tensor products varies depending on the context, but it is sufficient here to state that the tensor product of vectors I,1a, J,1b and K,1c is a vector with co-ordinates xi,yj,zk. Combining all of these vectors over all factors gives the threeway data matrix.Graphically, the PARAFAC model for a threeway, two-component system is shown in Fig. 1. The simplest way of implementing a PARAFAC model is by alternating least squares. Starting with a known three-way matrix, X, and two randomly initiated loadings matrices, A and B, the third loadings matrix, C, can be estimated. Then, from this new estimate of C, A and B then C can be successively estimated and so on until there is convergence in the model. The advantage of this approach is that, apart from the number of factors, F, no prior knowledge of the system is required.Furthermore, apart from scaling considerations, PARAFAC produces a unique solution. However, as it is a numeric rather than an analytical method, care must be taken to ensure that the algorithm converges properly and operates at an acceptable speed. The theory of the PARAFAC algorithm is discussed further elsewhere,14–16 as also are further applications to calibration.17–20 Practical applications, however, have been limited.21–23 Method Experimental Spectra of three chlorophyll degradation products were obtained experimentally from a Waters (Milford, MA, USA) Fig. 1 Graphical representation of the PARAFAC method. Analyst, September 1997, Vol. 122 (871–877) 871Model 990 HPLC–DAD system. For chlorophyll degradation mixtures, closely eluting compounds often possess very similar spectral characteristics, posing real problems in resolution by chemometric means.Spectra were recorded between 350 and 800 nm at 2 nm intervals. A common problem involves detecting crucial intermediates that are present in low concentrations, e.g., in the reaction A ? B ? C, where the first reaction is slow and the second is fast. A full kinetic model for the pathway requires the detection of B, which may be present in small amounts, dependent on the relative rates of the two reactions. If HPLC is employed to study such reactions, the intermediate, which may be a stereoisomer of the parent compound, could have very similar elution profiles and spectra to one of the other components.This situation is well established in the study of the degradation of chlorophyll by HPLC . Simulation Design The data were designed to simulate the degradation of chlorophyll-a products investigated elsewhere.24 Each threeway dataset is constructed from modes representing elution profiles, spectra and degradation profiles of each component. Three spectra obtained experimentally were used to represent three compounds with very similar spectral characteristics.The elution profiles were simulated and the degradation profiles represent a series of HPLC–DAD traces used to monitor a threereactant system. The three species formed are assumed to form part of a reaction series, where the reactant (R) is converted into an intermediate (I), which is then converted into a product (P).Elution Profiles The elution data matrix is represented by 20,3A, representing 20 chromatographic points in time. The chromatographic profile of compound l (where l = 1, 2 and 3 for the reactant, intermediate and product, respectively) at time I is given by al,i. Each column of A is an individual elution profile represented in this simulation by Gaussians centred at points 10, 14 and 6 in time and given by ai,1 = e - (i-10)2 6 (5) ai,2 = e - (i-14)2 6 (6) a e i i , ( ) 3 6 6 2 = - - (7) The data are designed so that the product elutes first, followed by the reactant and then the intermediate.The elution profiles are shown graphically in Fig. 2. Spectra Spectra were chosen from previous work to represent each component. Brereton and co-workers24,25 have shown that in chlorophyll degradation studies the ratio of absorbances between the two absorbance maxima, at approximately 430 and 665 nm, respectively, is very diagnostic.These spectra were chosen so that the reactant and the intermediate had spectra with similar features. Table 1 gives the spectral characteristics of each component. The spectra, shown in Fig. 3, were scaled to constant maximum absorbance and stored as 226,3B, where each column represents a spectrum taken between 350 and 800 nm, digitised at a resolution of 2 nm, i.e., 226 readings per spectrum. Degradation profiles The degradation profiles were designed to represent a reactant decreasing in concentration as the experiment proceeds, a minor intermediate increasing to a maximum, then decreasing and a product increasing in concentration. Concentrations are such that at any time the total concentration of all species is constant.The degradation data matrix 20,3C represents the concentration of the three components at 20 sampling points throughout the experiment where the three columns represent the reactant, intermediate and product respectively. The profiles for each component are given by ck,1 = 5e2k/8 (8) ck,2 = (5 2 5e2k/8)e2k/d (9) ck,3 = 5 2 5e2k/8 + e2k/d 2 5e2(k/8 + k/d) (10) where k is the sample number arranged sequentially in time and d is a parameter varying according to the relative significance and kinetic stability of the intermediate. The greater the value of d the slower is the decomposition of the intermediate, and so the easier it is to detect.Different simulations were performed at different values of d and simulated degradation profiles are shown in Fig. 4.Fig. 2 Simulated elution profiles. 8, Reactant; 3, intermediate; and 5, product. Table 1 Position of maxima of designed and predicted elution profiles. Elution maximum Reactant Intermediate Product 10 14 6 Rate parameter, d Factor 1 Factor 2 Factor 3 1 10 10 6 3 10 10 6 5 10 6 14 10 6 10 14 20 14 6 10 Fig. 3 Experimentally obtained spectra used in the simulations. Solid line, reactant; dotted line, intermediate; and dashed line, product. 872 Analyst, September 1997, Vol. 122Formation of three-way data set The three-way, three-component model 20,226,20 X is formed by xi, j,k = ai, f bj, f ck, f f =1 F å (11) where f is the component number and F is the number of components, which in this study is three. Five datasets were created in which the rate constant, d, was 1, 3, 5, 10 and 20. Application of PARAFAC Algorithm No pre-processing is performed on the data in this paper. The issue of pre-processing of three-way arrays is more complex than for the two-way case.Centring can be performed across either one, two or three of the modes and can distort the trilinear model. The order of any pre-processing is also critical. These issues have been discussed elsewhere.9,15,26,27 The datasets were decomposed by the PARAFAC written in Matlab 4.2 (Mathworks, Natick, MA, USA). The algorithm was used to extract three factor matrices from each simulated dataset initialised using random vectors and a convergence limit of 1 3 1026 between successive estimates of the sum of squares of the misfit.Indicators of Quality of Reconstruction Various functions can be used to compare the results from the simulations with the design data. Component sum of squares This gives a measure of the size of each component and factor, which aids in the identification of factors and gives an indication of their purity. The size of each predicted component, �S f, is given by �S f = ( �a i, f � bj, f � ck, f )2 k=1 K å j=1 J å i=1 I å (12) The values obtained for the predicted model using eqn.(12) can be compared to the size of the true components, Sf, calculated in the same manner as above, but with the estimated vectors replaced by their true equivalents. The square root of the ratio of the estimated to the true sum of squares, Qf, is given by Qf = �S f / Sf (13) The closer this value is to unity, the better is the modelling of the factor f.The concentration of the reactant will decrease identically from sample to sample in each simulation. The relative concentration of the intermediate will increase with increasing d and the product will decrease. The total concentration of the reactant, intermediate and product at any one point will always be constant, but the sum of squares will not. Regression For each of the elution, spectral and degradation modes, the predicted data are regressed on to the real data.In each case a matrix R can be obtained, often called a rotation or transformation matrix. For example, for the elution data, if A is the true elution datum, �A is the predicted data and RA is the rotation matrix, then �A = ARA (14) RA is found by the pseudo-inverse: RA = (AAA)21 AA �A (15) For a good model, each column and row of the rotation matrix contains only one value significantly greater than zero. Using the rotation matrices obtained above and the rotation matrix, a ‘predicted true’ dataset can be obtained, denoted by a circle overscript, e.g.for the elution data °A = � AR21 A (16) Calculating a residual root mean sum of squares RMSEP(A) between the actual true and predicted true data, across each matrix, gives a further indication of the quality of regression: RMSEP( ) ( ° � ) , , A a a I F i f i f f F i I = - � = = å å 2 1 1 (17) Spectral characteristics and elution maxima A simple measure of the success of the decomposition can obtained by comparing the predicted design parameters with those listed in Table 1.From the predicted elution data the position of the maximum of each component can be obtained as the maximum of each column of �A . Similarly, the positions of the two absorbance maxima and their ratio can be computed from the estimated spectral data. Results Most methods for factor analysis depend first on determining the number of significant factors. This is particularly true when the aim is to model the entire dataset.The importance of detecting and modelling all significant components in two-way factor analysis has been discussed in the context of mid-infrared (MIR) spectrometry.28,29 If a third significant factor is ignored, then the information from this compound is mixed with the other two compounds. In contrast, if a third factor is small it may become confused with the other two factors if a threefactor model is employed. PARAFAC depends crucially on a prior estimate of the number of significant factors as shown below.The following section reports the results assuming a three component mixture and the subsequent section a two component mixture. Another important aspect is that the order in which the factors are extracted may differ according to how the algorithm is implemented, e.g., the starting point of the iterations. This Fig. 4 Simulated degradation profiles. 8, Reactant; 3, intermediate; and 5, product. Analyst, September 1997, Vol. 122 873means that, over a series of datasets, the first factor may correspond to physically different compounds in each run, so it is first necessary to reorder the factors according to presumed physical significance. In some cases, where the interpretation of each factor is in doubt, this can be difficult. In the tables, the factors are ordered according to the order in which they were extracted. Three-factor Systems The predicted maxima positions of the elution profiles, in terms of elution index for the three component, three-factor system, are given in Table 2.At the levels where d, the kinetic rate parameter, is low, i.e., 1 and 3, the PARAFAC algorithm fails to position the three components correctly, whereas at higher levels of d all three components are correctly determined. Table 2 lists the spectral parameters determined from the predicted data. In all cases the product (factor 3 at d = 1 and 3, factor 2 at d = 5 and 20 and factor 1 at d = 10) is predicted well, with a peak ratio of 1.65, and a low-wavelength absorbance maximum at 424 or 426 nm.There is a slight problem with predicting the high-wavelength absorbance maximum at higher values of d, presumably because the prediction ability decreases as the amount of intermediate increases. However, the highwavelength absorption maximum is always @660 nm. A 4 nm shift in position represents only two sampling points in the wavelength direction.In this study, the data were designed with very similar spectral parameters and the PARAFAC algorithm has successively determined these, so it is trivial to establish the correspondence between components and factors. However, this will not always be true and in cases where there are several very similar components, a confident identification of the factors based on spectral parameters may not be possible. The component sum of squares for the true and predicted data are given in Table 3.At all levels the size of the product is predicted remarkably well. The reactant and intermediate, however, are only closely estimated at the two higher levels of d. This can be understood by considering that the reactant and product had similar spectral characteristics. At the lower levels of d, the intermediate is relatively minor compared with the reactant, but as the intermediate increases in significance at higher levels of d, it is easier for the algorithm to distinguish between them.The elution profiles for d = 1 and 20 are presented graphically in Fig. 5. It is obvious that the product and intermediate are not distinguished when d is low; these two Table 2 Design and predicted spectral parameters for the three factor model. Design Reactant Intermediate Product Spectral ratio 1.28 1.24 1.65 Absorbance max. 1/nm 434 434 424 Absorbance max. 2/nm 670 670 660 d = 1— Factor 1 Factor 2 Factor 3 Spectrum ratio 1.29 1.28 1.65 Absorbance max. 1/nm 436 436 426 Absorbance max. 2/nm 670 670 660 d = 3— Spectrum ratio 1.30 1.27 1.65 Absorbance max. 1/nm 434 434 424 Absorbance max. 2/nm 670 670 660 d = 5— Spectrum ratio 1.28 1.65 1.24 Absorbance max. 1/nm 434 426 436 Absorbance max. 2/nm 670 660 670 d = 10— Spectrum ratio 1.65 1.28 1.24 Absorbance max. 1/nm 424 424 434 Absorbance max. 2/nm 658 668 668 d = 20— Spectrum ratio 1.24 1.65 1.28 Absorbance max. 1/nm 434 424 434 Absorbance max. 2/nm 668 656 668 Table 3 Size of each design component and factors for the two- and threecomponent systems Rate parameter, d 1 3 5 10 20 Reactant 9100 9100 9100 9100 9100 Intermediate 8 167 591 2570 7165 Product 14827 13416 11750 7637 3547 Three-component system— Factor 1 842 1825 8201 7640 6334 Factor 2 4741 3803 11751 8463 3547 Factor 3 14285 13412 572 2483 9461 Two-component system— Factor 1 9209 9872 11684 13111 5317 Factor 2 14312 13473 10795 7549 15294 Table 4 Prediction ratios, Qf, for the two- and three-component models Ratio Qf Three-component model Two-component model d Reactant Intermediate Product Reactant Product 1 0.9999 10.2884 0.7218 1.0060 1.0009 3 0.9999 3.3054 0.6465 1.0416 1.0021 5 1.0000 0.9836 0.9493 0.9972 1.0892 10 1.0002 0.9829 0.9646 1.2003 0.9942 20 1.0197 0.9402 1.0001 n/a n/a Table 5 Root mean square error of prediction (RMSEP) for the three-factor models Data mode d A B C 1 1.1 3 1027 4.19 3 1027 3.5 3 1028 3 4.2 3 1028 2.89 3 1028 2.3 3 1028 5 1.4 3 1028 2.89 3 1027 3.5 3 1028 10 3.7 3 1028 2.3 3 1028 3.6 3 1028 20 3.37 3 1028 3.55 3 1027 2.63 3 1028 Fig. 5 Elution profiles obtained for the three factor models at (a) d = 1 and (b)ember 1997, Vol. 122species have similar spectral characteristics. Figs. 6(a) and (b) are representations of the corresponding spectra and it can be seen that they are recovered well. It can be concluded that when the number of components is correctly known, the PARAFAC algorithm produces excellent decomposition results.These results are the best when all of the components are relatively significant, as shown by the square root of the ratios of predicted to true sum of squares, Qf, given in Table 4, and the RMSEP in Table 5. For the elution data, A, improves considerably from the d = 1 to the d = 20 level. There is also an improvement, but to a lesser extent, for the spectral data. The error in the kinetic profiles is reasonably constant at each level.Two factor Systems The PARAFAC algorithm was repeated on the datasets but with two rather than three factors used to model the data. The elution and spectral parameters found are given in Tables 6 and 7, respectively. As can be seen from Table 6, at each level of d the algorithm appears to detect successfully the reactant and product without any interference from the intermediate. Note that the product should elute at datapoint 6 and the reactant at datapoint 10. Again, in Table 7, it appears that the two-component model produces good predictions of the spectrum ratios and absorbance maxima at each level of d, although the peak ratio for the product (1.56) is lower at d = 20.However, when the sum of squares of the factors and components are computed (Table 3), the situation is not so straightforward. At the lower two levels of the intermediate the Table 6 Position of maxima of predicted elution profiles for the two factor model Elution maximum d Factor 1 Factor 2 1 10 6 3 10 6 5 6 10 10 10 6 20 6 10 Fig. 6 Predicted spectra for (a) three factors at d = 1, (b) three factors at d = 2, (c) two factors at d = 1 and (d) two factors at d = 20. Fig. 7 Elution profiles obtained for the two factor models at (a) d = 1, (b) d = 3, (c) d = 5, (d) d = 10 and (e) d = 20. Analyst, September 1997, Vol. 122 875factors predict the size of the component reasonably well, but at the higher levels it becomes more difficult to distinguish the factors. This result reinforces the observation from above that a univariate measure such as elution maximum is not a sophisticated measure of data quality and a multivariate method, utilising the data from all available modes, should always be used in preference.This is important as the PARAFAC algorithm distributes all of the observed systematic variance between the factors in the model, so that these are not necessarily pure factors. Unlike the three-factor model above, predictions for the two factor model are better when the unmodelled component is relatively insignificant.As can be seen in Table 4, as the level of the intermediate increases the quality of reconstruction of the spectrally similar reactant decreases, but this is only observed when a multivariate measure such as the sum of squares or the rotation matrices is used. The quality of the product also falls but less significantly. Because at the d = 20 level a confident determination of the identification of the reactant and product cannot be made, the prediction ratio therefore cannot be calculated.The five recovered elution profiles are shown in Fig. 7. These supplement the data in Table 6; it is obvious that for d = 20 the first factor has, in fact, two clear maxima. Interestingly, the intermediate is confused with the product and predicted as one factor, despite the difference in both spectral characteristics and elution profiles. This unexpected result can be explained in terms of kinetic profiles; the level of intermediate builds up rapidly and then decreases with time, and so the kinetics of the two compounds are fairly similar.Two components with identical kinetics but different spectra and elution profiles could be modelled as a single factor. Visual inspection of the predicted chromatograms in Fig. 7(c)–(e) should provide clues that the number of predicted components is too few, and so lead to rerunning the model including further components.The spectra, Fig. 6(c) and (d), are recovered well again. Conclusions PARAFAC is a powerful approach for resolving out series for two-way chromatograms recorded over a number of samples. The methods can be extended to three-way or higher data, e.g., chromatograms could be recorded at different pH values and times; the change in chromatography with pH complements the change in intensity with time. The dataset in this paper is demanding, with the following properties. The middle chromatographic peak has no composition 1 or selective region, and most factor analysis methods find it difficult to resolve out unselective peaks. The spectra of the reactant and intermediate are almost identical, with similar spectra ratios and absorbance maxima, and also partially coelute.Approaches such as windows factor analysis and evolutionary factor analysis will not resolve out neighbouring peaks with very similar spectra; these will simply be modelled by one principal component.Even two-dimensional peak purity methods such as derivatives depend on change in spectral composition over elution time and simply would not detect a difference between the reactant and intermediate. By using PARAFAC on a series of chromatograms, these peaks can be distinguished provided that the concentration of the intermediate is not too low. Hence PARAFAC has potential as a major technique for the resolution and quantification of a series of two-way of chromatograms, often in cases where normal factor analysis methods will fail.The major drawback is that a good estimate of the number of components is required in advance for sensible models. If this is unknown, it is better to perform the models with fewer components first to see whether there are any elution profiles with more than one maximum. If so, the algorithm can be repeated, increasing the number of components until unimodal elution profiles are achieved.The authors thank R. Bro for providing the Matlab PARAFAC algorithm and EPSRC for providing financial support for this project. Appendix List of Notation Used i Elution time index I Total number of elution points (20) j Spectral wavelength index J Number of points in each spectrum (226) k Sample number K Number of reaction times (20) A Elution data matrix, with individual point al,i B Spectra data matrix, with individual point bl,j C Concentration data matrix, with individual point cl,k l Component number L Number of components d Kinetic rate parameter X Three-way data, with individual point xi,j,k f Factor number F Total number of factors Three-way matrices are represented by underlined uppercase bold italic characters, e.g., X, two-way matrices by uppercase bold italic characters, e.g., A, vectors by lower case bold italic characters, e.g., af, and scalars by non-bold characters.Estimated variables are denoted by a ‘hat,’ e.g., �A , except in eqns.(16) and (17), where the ‘estimated true’ data are represented by a circle overscript. Dimensions of matrices are given as left-hand side subscripts, e.g., 20,10A is a matrix of 20 rows by 10 columns. References 1 Geladi, P., Chemom. Intell. Lab. Syst., 1989, 7, 11. 2 Ståhle, L., Chemom. Intell. Lab. Syst., 1989, 7, 95. 3 Mitchell, B. C., and Burdick, D. S., Chemom. Intell. Lab. Syst., 1993, 20, 149. Table 7 Predicted spectral parameters for the two factor model. The design parameters are given in Table 1 d = 1— Factor 1 Factor2 Spectrum ratio 1.28 1.65 Absorbance max. 1/nm 436 426 Absorbance max. 2/nm 670 660 d = 3— Spectrum ratio 1.28 1.65 Absorbance max. 1/nm 436 426 Absorbance max. 2/nm 670 660 d = 5— Spectrum ratio 1.65 1.28 Absorbance max. 1/nm 426 436 Absorbance max. 2/nm 660 670 d = 10— Spectrum ratio 1.27 1.65 Absorbance max. 1/nm 436 426 Absorbance max. 2/nm 670 660 d = 20— Spectrum ratio 1.56 1.27 Absorbance max. 1/nm 426 436 Absorbance max. 2/nm 660 670 876 Analyst, September 1997, Vol. 1224 Wold, S., Geladi, P., Esbensen, K., and � Ohman, J., J. Chemom., 1987, 1, 41. 5 Keim, O. M., and Grung, B., Chemom. Intell. Lab. Syst., 1995, 29, 213. 6 Smilde, A. K., and Doornbos, D. A., J. Chemom., 1991, 5, 345. 7 Tauler, R., Smilde, A. K., Kowalski, B., J. Chemom., 1995, 9, 31. 8 Cattell, R., Psychol. Bull., 1952, 49, 499. 9 Kruskal, J. B., Psychometrika, 1976, 41, 281. 10 Sands, R., and Young, F., Psychometrika, 1980, 45, 39. 11 Kroonberg, P. M., and de Leeuw, J., Psychometrika, 1980, 45, 69. 12 Tucker, L., in Problems of Measuring Change, ed. Harris, C., University of Wisconsin Press, Madison, WI, 1963, p. 122. 13 Burdick, D. S., Chemom. Intell. Lab. Syst., 1995, 28, 229. 14 Harshman, R. A., and Lundy, M. E., Comput. Stat. Data Anal., 1994, 18, 39. 15 Smilde, A. G., Chemom. Intell. Lab. Syst., 1992, 15, 143. 16 Henrion, R., Chemom. Intell. Lab. Syst., 1994, 25, 1. 17 Bro, R., and Heimdal, H., Chemom. Intell. Lab. Syst., 1996, 34, 85. 18 Smilde, A. K., J. Chemom., 1992, 6, 11. 19 Smilde, A. K., Van der Graaf, P. H., Doornbos, D. A., Steerneman, T., and Sleurink, A., Anal. Chim. Acta, 1990, 235, 41. 20 Bro, R., J. Chemom., 1996, 10, 47. 21 Booksh, K. S., Muroski, A. R., and Myrick, M. L., Anal. Chem., 1996, 68, 3539. 22 Bro, R., Chemom. Intell. Lab. Syst., 1996, 34, 85. 23 Smilde, A. K., Tauler, R., Henshaw, J. M., Burgess, L. W., and Kowalski, B. R., Anal. Chem., 1994, 66, 3345. 24 Brereton, R. G., Rahmani, A., Liang, Y., Z., and Kvalheim, O. M., Photochem. Photobiol., 1993, 57, 1048. 25 Elbergali, A. K., Brereton, R. G., and Rahmani A., Analyst, 1995, 120, 2207. 26 Harshman, R. A., and Lundy, M. E., in Research Methods for Multimode Data Analysis, ed. Law, H. G., Snyder, C. W., Hattie, J. A., and McDonald, R. P., Praeger, New York, 1984, p. 216. 27 Ten Berge, J. M. F., in Multiway Data Analyses, ed. Coppi, R., and Bolasco, S., Elsevier, Amsterdam, 1989, p. 53. 28 Gurden, S. P., Brereton. R. G., and Groves, J. A., Analyst, 1996, 121, 441. 29 Gurden, S. P., Brereton, R. G., and Groves, J. A., Chemom. Intell. Lab. Syst., 1994, 23, 123. Paper 7/02232H Received April 2, 1997 Accepted May 12, 1997 Analyst, September 1997, Vol. 122 877
ISSN:0003-2654
DOI:10.1039/a702232h
出版商:RSC
年代:1997
数据来源: RSC
|
3. |
Raman Spectral Estimation via Fast OrthogonalSearch |
|
Analyst,
Volume 122,
Issue 9,
1997,
Page 879-882
Michael J. Korenberg,
Preview
|
|
摘要:
Raman Spectral Estimation viaFast Orthogonal Search Michael J. Korenberg*a, Colin J. H. Brenanb and Ian W. Hunterb a Department of Electrical and Computer Engineering, Queen’s University, Kingston, Ontario, Canada K7L 3N6 b Department of Mechanical Engineering, Room 3-147, Massachusetts Institute of Technology, Cambridge, MA 02139, USA A Fourier transform (FT) spectrometer measures the autocorrelation (interferogram) of radiation emitted from a source and estimates the optical power spectral density through application of the discrete Fourier transform (DFT) to the recorded interferogram.Although a widely used method, FT spectrometry suffers because its frequency resolution is limited to the sampling rate divided by the number of time-series data points. A large number of points are therefore required to resolve an optical spectrum properly. In this paper, it is shown that a noise-resistant technique known as fast orthogonal search (FOS) can be used to achieve accurate optical spectrum estimation. Further, it is shown that frequency accuracy comparable to the DFT applied to the full interferogram can be obtained with FOS even if the original interferogram is contaminated with noise and then reduced by a factor of up to 10 by irregularly spaced sampling.The FOS application presented here is for the estimation of Raman spectra from interferograms acquired with an FT Raman spectrometer. Keywords: Spectrum estimation; Raman spectroscopy; irregular sampling; discrete Fourier transform Beginning with the early work of Michelson,1 the Fourier transform (FT) spectrometer has become an increasingly important tool for optical spectral measurement.The FT spectrometer in its most general form is constructed from a twobeam, wavefront- or amplitude-division interferometer. After division of an incident optical field of wavenumber s0 ( = 1/l0) into two equi-amplitude beams, an optical path length difference, Ds, is imposed on the field of each beam on propagation through the interferometer.Interference of the two fields at the interferometer output generates an irradiance, J, that varies as a function of Ds: J(Ds) = B(s0)cos(2ps0 Ds) (1) where B is the optical power at s0 and Ds = 2(n1x12n2x2) (2) where, for i = 1 or 2, ni and xi are the refractive index and length of the ith interferometer arm, respectively. For a polychromatic field illuminating the interferometer, the interferogram, J(Ds), and optical spectrum are a cosine Fourier transform pair:2,3 J s B s ( ) ( )cos( ) D D = -¥ +¥ ò s ps s 2 d (3) and the interferogram is formally equivalent to the autocorrelation of the input optical field through the Wiener–Khitnitche theorem.4 Hence the essence of FT spectrometry is to measure the optical field autocorrelation with the interferometer and recover the optical power spectrum by application of the discrete Fourier transform (DFT).DFT-based spectral estimation requires the interferogram to be sampled at equi-spaced intervals and at a rate at least twice the maximum wavenumber, smax, of the optical signal (Nyquist sampling criterion).For an N-point data record sampled at frequency ss = 2smax, the DFT estimate will have a spectral resolution Ds s = 2 max N (4) and a spectral range, ds, equal to smax. The FT spectrometer resolving power R ( = ds/Ds) equals N/2 or R s = D max l (5) for a maximum optical path length difference, Dsmax, and light of mean wavelength l–. The time required to record an N-point interferogram, T, depends on the means by which Ds is systematically varied.Most commonly, one interferometer mirror is moved relative to the other and measurement of J at specified optical path differences yields the interferogram as a time-series signal.5–7 If the measurement time per interferogram point equals dt, the total time to measure the interferogram is simply T = Ndt.Assuming that optical shot noise dominates, the FT spectral signal-to-noise ratio (S/N) has been shown to scale as 1/ANB.8 Clearly, for an FT spectrometer based on a scanning mirror configuration, operation of the spectrometer at a high resolving power greatly increases the time to acquire an interferogram that yields an acceptable spectral S/N. Attempts to minimize T can adversely impact on the spectral resolution, spectral range and S/N. These restrictions are alleviated with an FT spectrometer fabricated from a static twobeam interferometer and an array photosensor with N equispaced photosensitive elements for recording the interferogram distributed across the detector array.9–12 For the same S/N, the acquisition time is decreased by N (known as the multi-channel advantage9) compared with a scanning mirror-type FT spectrometer but with a restricted spectral resolution since the number of array photoelements is usually a small fixed number.Given the above considerations, it would be particularly advantageous in many spectroscopic applications (e.g., imaging or time-resolved measurement) to estimate spectra from a smaller number of sample points, N, in order to minimize total measurement time, T, whilst simultaneously retaining a high spectral S/N and resolving power, R. Alternatively, for the static, two-beam interferometer, minimizing N will result in a smaller instrument or, equivalently, achieve higher resolution of the optical spectrum for a given size of instrument.A number of parametric methods have been proposed, such as the Prony and Pisarenko methods, the maximum entropy method, and various autoregressive moving average-based estimators,13 to obtain frequency resolutions comparable to the DFT but with a reduced number of sample points (i.e., smaller N) or, conversely, to obtain finer frequency resolution than is afforded by the DFT for the same N. However, in all of these methods the model order must be selected, and the first two additionally Analyst, September 1997, Vol. 122 (879–882) 879require solution of a polynomial equation, which may be of high degree. Accordingly, we demonstrate the application of a recent parametric spectral estimator, fast orthogonal search (FOS),14 to the estimation of Raman spectra acquired with an FT-Raman spectrometer.7 As opposed to the techniques listed above, model order is automatically determined in FOS; moreover, there is no need to solve a polynomial equation.Most important, we will show that reduction in the number of Raman interferogram points by a factor of 10 through irregular sampling and application of FOS to the reduced data set generates a spectral estimate comparable in frequency resolution to a DFT applied to the full interferogram. We also demonstrate that the FOS spectral estimator is relatively insensitive (compared with the DFT estimator) to additive noise over a large range of interferogram S/N.Optical spectrometers utilizing the FOS algorithm and irregularly spaced sampling may have important applications in process control or monitoring. Method There is a special benefit provided by FOS which is not readily available with the other spectral methods discussed above, namely the ability to handle conveniently time series which are unequally spaced or missing some data.14 This capability stems from the implicit orthogonalization procedure that FOS employs, and does not involve interpolation to ‘fill in’ missing data values, which would introduce error.The ability of FOS to cope with unequally spaced data permits irregular sampling of the interferogram and, further, the irregular sampling permits accurate resolution of high frequencies in the optical spectrum using fewer interferogram points than required by the DFT and other parametric spectral estimators. This can be understood through consideration of a time series obtained by sampling at the Nyquist frequency.Suppose next that a large number of data points are randomly deleted. Since some points would still remain closely proximate, high-frequency information would not be lost, yet far fewer points would be required for processing the time series to extract its spectral content. As noted, this exploits the capability of FOS to accept unequally spaced data without introducing error, unlike the other above-mentioned spectral methods. Why FOS has this ability is explained next.Denote the irregularly-spaced data points by y(n), sampled at times t = t(n), n = 0, ..., N 2 1. Then FOS enables one to build up a concise sinusoidal series model: y n a p n e n m M m m ( ) ( ) ( ) = + = å0 (6) where p0(n) = 1, and for i = 1,2, ..., p2i21(n) = cos wit(n) (7) p2i(n) = sin wit(n) (8) and e(n) is the equation error. The frequencies wi in eqns. (7) and (8) are found by systematically searching through a candidate set of frequencies wA, wB, ...These candidate frequencies are not required to be commensurate with, or integral multiples of, the fundamental frequency corresponding to the record length. The candidate frequencies can be selected with a priori knowledge of the specific frequencies sought or they could simply be frequencies distributed in frequency bands of interest. In particular, for i = 1, 2, ..., and M = 2i we set wi equal to that candidate frequency resulting in the greatest reduction in mean-square error (MSE) when the term pair Ti(n) = a2i21p2i21(n) + a2ip2i(n) (9) is added to the model of eqn.(6). An implicit orthogonalization of the term pairs, achieved via a slightly modified Cholesky decomposition, is used to obtain a computationally efficient procedure14 for building up the sinusoidal series model. Because the t(n) in eqns. (7) and (8) are the actual instants when the samples y(n) were taken, and these instants are used in defining the Ti(n) (which are implicitly orthogonalized), the unequal spacing of the data contributes no error.Results A comparative analysis of DFT and FOS spectral estimates is based on a double-sided Raman interferogram for neat methanol measured at equi-spaced scan mirror positions with an FT Raman spectrometer to yield an interferogram containing 6294 data points. The interferogram was zero-padded to 8192 points in order to treat the data set as a time series obtained with a sampling rate of 8192 Hz.After application of the Hamming window, the DFT was applied to obtain the Raman spectrum shown in Fig. 1(a), having a spectral resolution of 26 cm21 (2.6 Hz). As expected, the spectrum derived here compares well with the previously published Raman spectrum of methanol.15 Application of FOS to the original 6294 point interferogram generates a Raman spectrum similar to that obtained with the 8192 point DFT but with a slightly higher spectral resolution of 20 cm21 (2 Hz) [Fig. 1(b)].To obtain this Raman spectral estimate, FOS searched through 200 candidate frequencies equi-spaced between 1700 and 2098 Hz (23480 to 500 cm21) inclusive and selected the 40 most significant frequencies. The amplitudes of those frequencies are plotted in Fig. 1(b) and the amplitudes of unselected frequencies were set to zero. The small difference in spectral peak positions between the FOS and DFT spectra is attributed to their slightly different spectral resolutions.We now proceed to demonstrate a major difference between the DFT and FOS spectral estimators by recovering via FOS the methanol Raman spectrum with only 10% of the original interferogram data. At the same time, we will illustrate the pronounced capability of FOS to cope with noise contamination Fig. 1 Comparison of the Raman spectrum of methanol estimated from a 6294 point interferogram with two different algorithms. (a) Raman spectrum resulting from application of the DFT to the 6294 point interferogram, Hamming windowed and zero padded to 8192 points.(b) Raman spectral estimate that results from the direct application of FOS to the 6294 point raw interferogram. Divisions on the vertical axes are proportional to the photoelectron count and the horizontal axes indicate spectral position in wavenumbers. 880 Analyst, September 1997, Vol. 122of the interferogram. We began by adding to the full interferogram zero-mean white Gaussian noise whose variance was 10% of that of the interferogram. Then, 621 of the 6294 points in the noisy interferogram were randomly selected and FOS was directly applied without any additional processing to this reduced data set.Again, FOS searched through 200 candidate frequencies equi-spaced between 1700 and 2098 Hz, and the amplitudes of selected frequencies are shown in Fig. 2(a). The following stopping criterion was used.16 FOS required, in order to continue, that the greatest reduction achievable by adding a further frequency, divided by the mean square of the current residue, exceed a threshold divided by the number of data points.This criterion (which follows immediately from using a standard correlation test) helps to avoid choosing frequencies which are merely fitting noise. A threshold of 10.9 was used here, roughly corresponding to 99.9% confidence limits. Comparison of Fig. 2(a) with 1(a) shows that the Raman spectrum estimated from 10% of the noisy data using FOS is remarkably similar to the DFT result obtained from the full 6294 point interferogram with no added noise.Fig. 2(b) and (c) show, for 50% and 100% noise contamination, respectively, the FOS estimated Raman spectra using the same 621 point irregular sampling sequence. It is impressive that with only 10% of the original interferogram points and interferogram S/Ns as low as 1 (100% additive noise) that the FOS algorithm recovers the major spectral components of the Raman spectrum.Note that the irregular spacing of the 621 point reduced data set was crucial to successful recovery of the spectrum. Had, instead, every tenth point been selected from the original data, the sampling rate would have been only about 820 Hz, suitable for recovering frequencies up to about 410 Hz but certainly not those at higher frequencies where most, if not all, of the spectral information in the interferogram is situated. Fig. 3(a) shows the unproductive result of applying the DFT in this situation, where no noise was added to the reduced data set.On the other hand, if the original sampling frequency of 8192 Hz is maintained but 621 data points were selected equally spaced to either side of the zero lag position (310 data points to either side) then the resulting DFT spectral estimate will have a resolution limit of 260 cm21 (26 Hz) [Fig. 3(b)], over ten times larger than the resolution observed with FOS applied to a data set with the same number of points.Again, no noise was added to the reduced data set used to obtain the DFT spectral estimate. Thus a FOS spectral analysis of a randomly reduced noisy data set results in substantially higher spectral resolution than is possible with a DFT spectral estimator applied to a clean interferogram with the same number of points. Comparison of Fig. 3(a) and (b) with Fig. 2(a) is particularly relevant in those situations where the number of sampled interferogram points must be minimized. Similarly accurate FOS Raman spectral estimates from reduced interferograms have also been observed for other condensed phase organic and inorganic samples. Moreover, it has been shown that the recovery of the major spectral components in a Raman spectrum from a randomly reduced interferogram via FOS is not critically dependent on the particular random sampling sequence employed.Conclusions We have introduced for optical spectrum estimation the use of a recent spectral estimator called fast orthogonal search14,16 and compared its performance with the discrete Fourier transform on an interferogram recorded with an FT spectrometer.The specific example given here is the estimation of the Raman spectrum of methanol from its interferogram measured with an FT Raman spectrometer. The FOS algorithm does not impose the same restriction on sampling of the interferogram as the DFT; therefore, FOS requires far fewer interferogram data points than the DFT to obtain a spectral estimate of comparable resolution to a DFT estimate that utilizes the entire interferogram data record.Consequently, the reduced number of interferogram points needed to obtain a given spectral resolu- Fig. 2 Raman spectral estimate of methanol by FOS applied to a 621 point reduced interferogram generated by resampling a noisy 6294 point interferogram with a random sampling sequence. In (a), the noise added to the original interferogram had a variance equal to 10% of the variance of the original interferogram.In (b) and (c), the noise variance was 50% and 100%, respectively, and the peaks occurred at the same positions as marked in (a). Fig. 3 Comparison of the Raman spectrum of methanol estimated from the DFT applied to (a) a reduced 621 point interferogram produced by resampling the original interferogram at a new sampling rate that is one tenth the original sampling frequency and (b) a reduced 621 point interferogram composed of 310 points equally distributed to either side of the original interferogram zero lag position.The sampling rate of this reduced interferogram is identical with that of the original interferogram. Note that no noise was added to the original interferogram before the reduced interferograms were obtained. In (a) and (b), the reduced interferograms were Hamming windowed and zero padded to 8192 points. These spectra should be compared with that in Fig. 2(a), estimated by FOS from an equal number of interferogram points (which had additionally been corrupted with noise). Analyst, September 1997, Vol. 122 881tion via FOS implies substantial performance improvements for existing FT spectrometers when measurement time must be minimized and suggests new interferometric spectrometer designs based on the irregular sampling strategy allowed by FOS. This is especially pertinent for a spectrometer having an array photosensor where the number of array photoelements is typically small.To illustrate the differences between the DFT and FOS spectral estimators, a 6294 point Raman interferogram of neat methanol acquired with an FT Raman spectrometer was analyzed with each algorithm. First, the FOS and DFT Raman spectra computed from the full Raman interferogram were shown to be similar. Next, the interferogram was contaminated with additive Gaussian white noise and then resampled with an irregular (random) sampling sequence to generate a new interferogram having one tenth the points (621 points) of the original data set.Application of FOS to the reduced noisy interferogram generated a Raman spectrum remarkably similar to the DFT spectral estimate using the full clean interferogram data record (6294 points). As expected, application of the DFT to reduced data sets of 621 points failed to generate estimates comparable to the Raman spectrum obtained from the original interferogram even though the reduced sets did not have added noise, unlike those analyzed by FOS.The FOS method is general and robust and could be an alternative to the DFT in other FT-based spectrochemical analytical techniques. This would include atomic emission, visible or infrared absorption, nuclear magnetic resonance, electron paramagnetic resonance and ion cyclotron spectrometry. M.J.K. acknowledges the support of the Natural Sciences and Engineering Research Council of Canada.C.J.H.B. and I.W.H. acknowledge the support of this work in part by the Institute for Robotics and Intelligent Systems (IRIS), a Canadian Center of Excellence, and the Office of Naval Research. References 1 Michelson, A. A., Philos. Mag., 1891, 5, 338. 2 Fourier Transform Infrared Spectroscopy: Applications to Chemical Systems, ed. Ferraro, J. R., and Basile, L. J., Academic Press, New York, 1978. 3 Nordstrom, R. J., in Fourier, Hadamard and Hilbert Transforms in Chemistry, ed.Marshall, A. G., Plenum Press, New York, 1982, p. 21. 4 Saleh, B. E. A., and Teich, M. C., Fundamental of Photonics., Wiley, New York, 1991. 5 Hirschfeld, T., and Chase, D. B., Appl. Spectrosc., 1986, 40, 133. 6 Bell, R. J., Introductory Fourier Transform Spectroscopy, Academic Press, New York, 1972. 7 Brenan, C. J. H., and Hunter, I. W., Appl. Spectrosc., 1995, 49, 1086. 8 Kahn, F. D., Astrophys. J., 1959, 129, 518. 9 Zhao, J., and McCreery, R.L., Appl. Spectrosc., 1996, 50, 1209. 10 Moller, K. D., Appl. Opt., 1995, 34, 1493. 11 Junttila, M.-L., Appl. Opt., 1992, 31, 4106. 12 Okamoto, T., Kawata, S., and Minami, S., Appl. Opt., 1984, 23, 269. 13 Kay, S. M., Modern Spectral Estimation: Theory and Application, Prentice-Hall, Englewood Cliffs, NJ, 1988. 14 Korenberg, M. J., Biol. Cybern., 1989, 60, 267. 15 Raman/IR Atlas of Organic Compounds, ed. Schrader, B., and Meier, W., Verlag Chemie, Weinheim, 1977, p. A3-04. 16 Korenberg, M., in Non-linear Vision: Determination of Neural Receptive Fields, Function, and Networks, ed. Pinter, R. B., and Nabet, B., CRC Press, Boca Raton, FL, 1992, ch.7. Paper 7/00902J Received February 10, 1997 Accepted May 15, 1997 882 Analyst, September 1997, Vol. 122 Raman Spectral Estimation viaFast Orthogonal Search Michael J. Korenberg*a, Colin J. H. Brenanb and Ian W. Hunterb a Department of Electrical and Computer Engineering, Queen’s University, Kingston, Ontario, Canada K7L 3N6 b Department of Mechanical Engineering, Room 3-147, Massachusetts Institute of Technology, Cambridge, MA 02139, USA A Fourier transform (FT) spectrometer measures the autocorrelation (interferogram) of radiation emitted from a source and estimates the optical power spectral density through application of the discrete Fourier transform (DFT) to the recorded interferogram.Although a widely used method, FT spectrometry suffers because its frequency resolution is limited to the sampling rate divided by the number of time-series data points.A large number of points are therefore required to resolve an optical spectrum properly. In this paper, it is shown that a noise-resistant technique known as fast orthogonal search (FOS) can be used to achieve accurate optical spectrum estimation. Further, it is shown that frequency accuracy comparable to the DFT applied to the full interferogram can be obtained with FOS even if the original interferogram is contaminated with noise and then reduced by a factor of up to 10 by irregularly spaced sampling.The FOS application presented here is for the estimation of Raman spectra from interferograms acquired with an FT Raman spectrometer. Keywords: Spectrum estimation; Raman spectroscopy; irregular sampling; discrete Fourier transform Beginning with the early work of Michelson,1 the Fourier transform (FT) spectrometer has become an increasingly important tool for optical spectral measurement.The FT spectrometer in its most general form is constructed from a twobeam, wavefront- or amplitude-division interferometer. After division of an incident optical field of wavenumber s0 ( = 1/l0) into two equi-amplitude beams, an optical path length difference, Ds, is imposed on the field of each beam on propagation through the interferometer. Interference of the two fields at the interferometer output generates an irradiance, J, that varies as a function of Ds: J(Ds) = B(s0)cos(2ps0 Ds) (1) where B is the optical power at s0 and Ds = 2(n1x12n2x2) (2) where, for i = 1 or 2, ni and xi are the refractive index and length of the ith interferometer arm, respectively. For a polychromatic field illuminating the interferometer, the interferogram, J(Ds), and optical spectrum are a cosine Fourier transform pair:2,3 J s B s ( ) ( )cos( ) D D = -¥ +¥ ò s ps s 2 d (3) and the interferogram is formally equivalent to the autocorrelation of the input optical field through the Wiener–Khitnitche theorem.4 Hence the essence of FT spectrometry is to measure the optical field autocorrelation with the interferometer and recover the optical power spectrum by application of the discrete Fourier transform (DFT).DFT-based spectral estimation requires the interferogram to be sampled at equi-spaced intervals and at a rate at least twice the maximum wavenumber, smax, of the optical signal (Nyquist sampling criterion).For an N-point data record sampled at frequency ss = 2smax, the DFT estimate will have a spectral resolution Ds s = 2 max N (4) and a spectral range, ds, equal to smax. The FT spectrometer resolving power R ( = ds/Ds) equals N/2 or R s = D max l (5) for a maximum optical path length difference, Dsmax, and light of mean wavelength l–. The time required to record an N-point interferogram, T, depends on the means by which Ds is systematically varied. Most commonly, one interferometer mirror is moved relative to the other and measurement of J at specified optical path differences yields the interferogram as a time-series signal.5–7 If the measurement time per interferogram point equals dt, the total time to measure the interferogram is simply T = Ndt.Assuming that optical shot noise dominates, the FT spectral signal-to-noise ratio (S/N) has been shown to scale as 1/ANB.8 Clearly, for an FT spectrometer based on a scanning mirror configuration, operation of the spectrometer at a high resolving power greatly increases the time to acquire an interferogram that yields an acceptable spectral S/N.Attempts to minimize T can adversely impact on the spectral resolution, spectral range and S/N. These restrictions are alleviated with an FT spectrometer fabricated from a static twobeam interferometer and an array photosensor with N equispaced photosensitive elements for recording the interferogram distributed across the detector array.9–12 For the same S/N, the acquisition time is decreased by N (known as the multi-channel advantage9) compared with a scanning mirror-type FT spectrometer but with a restricted spectral resolution since the number of array photoelements is usually a small fixed number.Given the above considerations, it would be particularly advantageous in many spectroscopic applications (e.g., imaging or time-resolved measurement) to estimate spectra from a smaller number of sample points, N, in order to minimize total measurement time, T, whilst simultaneously retaining a high spectral S/N and resolving power, R.Alternatively, for the static, two-beam interferometer, minimizing N will result in a smaller instrument or, equivalently, achieve higher resolution of the optical spectrum for a given size of instrument. A number of parametric methods have been proposed, such as the Prony and Pisarenko methods, the maximum entropy method, and various autoregressive moving average-based estimators,13 to obtain frequency resolutions comparable to the DFT but with a reduced number of sample points (i.e., smaller N) or, conversely, to obtain finer frequency resolution than is afforded by the DFT for the same N.However, in all of these methods the model order must be selected, and the first two additionally Analyst, September 1997, Vol. 122 (879–882) 879require solution of a polynomial equation, which may be of high degree.Accordingly, we demonstrate the application of a recent parametric spectral estimator, fast orthogonal search (FOS),14 to the estimation of Raman spectra acquired with an FT-Raman spectrometer.7 As opposed to the techniques listed above, model order is automatically determined in FOS; moreover, there is no need to solve a polynomial equation. Most important, we will show that reduction in the number of Raman interferogram points by a factor of 10 through irregular sampling and application of FOS to the reduced data set generates a spectral estimate comparable in frequency resolution to a DFT applied to the full interferogram.We also demonstrate that the FOS spectral estimator is relatively insensitive (compared with the DFT estimator) to additive noise over a large range of interferogram S/N. Optical spectrometers utilizing the FOS algorithm and irregularly spaced sampling may have important applications in process control or monitoring. Method There is a special benefit provided by FOS which is not readily available with the other spectral methods discussed above, namely the ability to handle conveniently time series which are unequally spaced or missing some data.14 This capability stems from the implicit orthogonalization procedure that FOS employs, and does not involve interpolation to ‘fill in’ missing data values, which would introduce error.The ability of FOS to cope with unequally spaced data permits irregular sampling of the interferogram and, further, the irregular sampling permits accurate resolution of high frequencies in the optical spectrum using fewer interferogram points than required by the DFT and other parametric spectral estimators. This can be understood through consideration of a time series obtained by sampling at the Nyquist frequency.Suppose next that a large number of data points are randomly deleted. Since some points would still remain closely proximate, high-frequency information would not be lost, yet far fewer points would be required for processing the time series to extract its spectral content. As noted, this exploits the capability of FOS to accept unequally spaced data without introducing error, unlike the other above-mentioned spectral methods. Why FOS has this ability is explained next.Denote the irregularly-spaced data points by y(n), sampled at times t = t(n), n = 0, ..., N 2 1. Then FOS enables one to build up a concise sinusoidal series model: y n a p n e n m M m m ( ) ( ) ( ) = + = å0 (6) where p0(n) = 1, and for i = 1,2, ..., p2i21(n) = cos wit(n) (7) p2i(n) = sin wit(n) (8) and e(n) is the equation error.The frequencies wi in eqns. (7) and (8) are found by systematically searching through a candidate set of frequencies wA, wB, ... These candidate frequencies are not required to be commensurate with, or integral multiples of, the fundamental frequency corresponding to the record length.The candidate frequencies can be selected with a priori knowledge of the specific frequencies sought or they could simply be frequencies distributed in frequency bands of interest. In particular, for i = 1, 2, ..., and M = 2i we set wi equal to that candidate frequency resulting in the greatest reduction in mean-square error (MSE) when the term pair Ti(n) = a2i21p2i21(n) + a2ip2i(n) (9) is added to the model of eqn. (6). An implicit orthogonalization of the term pairs, achieved via a slightly modified Cholesky decomposition, is used to obtain a computationally efficient procedure14 for building up the sinusoidal series model.Because the t(n) in eqns. (7) and (8) are the actual instants when the samples y(n) were taken, and these instants are used in defining the Ti(n) (which are implicitly orthogonalized), the unequal spacing of the data contributes no error. Results A comparative analysis of DFT and FOS spectral estimates is based on a double-sided Raman interferogram for neat methanol measured at equi-spaced scan mirror positions with an FT Raman spectrometer to yield an interferogram containing 6294 data points.The interferogram was zero-padded to 8192 points in order to treat the data set as a time series obtained with a sampling rate of 8192 Hz. After application of the Hamming window, the DFT was applied to obtain the Raman spectrum shown in Fig. 1(a), having a spectral resolution of 26 cm21 (2.6 Hz).As expected, the spectrum derived here compares well with the previously published Raman spectrum of methanol.15 Application of FOS to the original 6294 point interferogram generates a Raman spectrum similar to that obtained with the 8192 point DFT but with a slightly higher spectral resolution of 20 cm21 (2 Hz) [Fig. 1(b)]. To obtain this Raman spectral estimate, FOS searched through 200 candidate frequencies equi-spaced between 1700 and 2098 Hz (23480 to 500 cm21) inclusive and selected the 40 most significant frequencies.The amplitudes of those frequencies are plotted in Fig. 1(b) and the amplitudes of unselected frequencies were set to zero. The small difference in spectral peak positions between the FOS and DFT spectra is attributed to their slightly different spectral resolutions. We now proceed to demonstrate a major difference between the DFT and FOS spectral estimators by recovering via FOS the methanol Raman spectrum with only 10% of the original interferogram data.At the same time, we will illustrate the pronounced capability of FOS to cope with noise contamination Fig. 1 Comparison of the Raman spectrum of methanol estimated from a 6294 point interferogram with two different algorithms. (a) Raman spectrum resulting from application of the DFT to the 6294 point interferogram, Hamming windowed and zero padded to 8192 points. (b) Raman spectral estimate that results from the direct application of FOS to the 6294 point raw interferogram.Divisions on the vertical axes are proportional to the photoelectron count and the horizontal axes indicate spectral position in wavenumbers. 880 Analyst, September 1997, Vol. 122of the interferogram. We began by adding to the full interferogram zero-mean white Gaussian noise whose variance was 10% of that of the interferogram. Then, 621 of the 6294 points in the noisy interferogram were randomly selected and FOS was directly applied without any additional processing to this reduced data set.Again, FOS searched through 200 candidate frequencies equi-spaced between 1700 and 2098 Hz, and the amplitudes of selected frequencies are shown in Fig. 2(a). The following stopping criterion was used.16 FOS required, in order to continue, that the greatest reduction achievable by adding a further frequency, divided by the mean square of the current residue, exceed a threshold divided by the number of data points. This criterion (which follows immediately from using a standard correlation test) helps to avoid choosing frequencies which are merely fitting noise.A threshold of 10.9 was used here, roughly corresponding to 99.9% confidence limits. Comparison of Fig. 2(a) with 1(a) shows that the Raman spectrum estimated from 10% of the noisy data using FOS is remarkably similar to the DFT result obtained from the full 6294 point interferogram with no added noise. Fig. 2(b) and (c) show, for 50% and 100% noise contamination, respectively, the FOS estimated Raman spectra using the same 621 point irregular sampling sequence.It is impressive that with only 10% of the original interferogram points and interferogram S/Ns as low as 1 (100% additive noise) that the FOS algorithm recovers the major spectral components of the Raman spectrum. Note that the irregular spacing of the 621 point reduced data set was crucial to successful recovery of the spectrum. Had, instead, every tenth point been selected from the original data, the sampling rate would have been only about 820 Hz, suitable for recovering frequencies up to about 410 Hz but certainly not those at higher frequencies where most, if not all, of the spectral information in the interferogram is situated.Fig. 3(a) shows the unproductive result of applying the DFT in this situation, where no noise was added to the reduced data set. On the other hand, if the original sampling frequency of 8192 Hz is maintained but 621 data points were selected equally spaced to either side of the zero lag position (310 data points to either side) then the resulting DFT spectral estimate will have a resolution limit of 260 cm21 (26 Hz) [Fig. 3(b)], over ten times larger than the resolution observed with FOS applied to a data set with the same number of points. Again, no noise was added to the reduced data set used to obtain the DFT spectral estimate. Thus a FOS spectral analysis of a randomly reduced noisy data set results in substantially higher spectral resolution than is possible with a DFT spectral estimator applied to a clean interferogram with the same number of points.Comparison of Fig. 3(a) and (b) with Fig. 2(a) is particularly relevant in those situations where the number of sampled interferogram points must be minimized. Similarly accurate FOS Raman spectral estimates from reduced interferograms have also been observed for other condensed phase organic and inorganic samples.Moreover, it has been shown that the recovery of the major spectral components in a Raman spectrum from a randomly reduced interferogram via FOS is not critically dependent on the particular random sampling sequence employed. Conclusions We have introduced for optical spectrum estimation the use of a recent spectral estimator called fast orthogonal search14,16 and compared its performance with the discrete Fourier transform on an interferogram recorded with an FT spectrometer. The specific example given here is the estimation of the Raman spectrum of methanol from its interferogram measured with an FT Raman spectrometer. The FOS algorithm does not impose the same restriction on sampling of the interferogram as the DFT; therefore, FOS requires far fewer interferogram data points than the DFT to obtain a spectral estimate of comparable resolution to a DFT estimate that utilizes the entire interferogram data record.Consequently, the reduced number of interferogram points needed to obtain a given spectral resolu- Fig. 2 Raman spectral estimate of methanol by FOS applied to a 621 point reduced interferogram generated by resampling a noisy 6294 point interferogram with a random sampling sequence. In (a), the noise added to the original interferogram had a variance equal to 10% of the variance of the original interferogram. In (b) and (c), the noise variance was 50% and 100%, respectively, and the peaks occurred at the same positions as marked in (a).Fig. 3 Comparison of the Raman spectrum of methanol estimated from the DFT applied to (a) a reduced 621 point interferogram produced by resampling the original interferogram at a new sampling rate that is one tenth the original sampling frequency and (b) a reduced 621 point interferogram composed of 310 points equally distributed to either side of the original interferogram zero lag position. The sampling rate of this reduced interferogram is identical with that of the original interferogram.Note that no noise was added to the original interferogram before the reduced interferograms were obtained. In (a) and (b), the reduced interferograms were Hamming windowed and zero padded to 8192 points. These spectra should be compared with that in Fig. 2(a), estimated by FOS from an equal number of interferogram points (which had additionally been corrupted with noise). Analyst, September 1997, Vol. 122 881tion via FOS implies substantial performance improvements for existing FT spectrometers when measurement time must be minimized and suggests new interferometric spectrometer designs based on the irregular sampling strategy allowed by FOS.This is especially pertinent for a spectrometer having an array photosensor where the number of array photoelements is typically small. To illustrate the differences between the DFT and FOS spectral estimators, a 6294 point Raman interferogram of neat methanol acquired with an FT Raman spectrometer was analyzed with each algorithm.First, the FOS and DFT Raman spectra computed from the full Raman interferogram were shown to be similar. Next, the interferogram was contaminated with additive Gaussian white noise and then resampled with an irregular (random) sampling sequence to generate a new interferogram having one tenth the points (621 points) of the original data set. Application of FOS to the reduced noisy interferogram generated a Raman spectrum remarkably similar to the DFT spectral estimate using the full clean interferogram data record (6294 points).As expected, application of the DFT to reduced data sets of 621 points failed to generate estimates comparable to the Raman spectrum obtained from the original interferogram even though the reduced sets did not have added noise, unlike those analyzed by FOS. The FOS method is general and robust and could be an alternative to the DFT in other FT-based spectrochemical analytical techniques. This would include atomic emission, visible or infrared absorption, nuclear magnetic resonance, electron paramagnetic resonance and ion cyclotron spectrometry. M.J.K. acknowledges the support of the Natural Sciences and Engineering Research Council of Canada. C.J.H.B. and I.W.H. acknowledge the support of this work in part by the Institute for Robotics and Intelligent Systems (IRIS), a Canadian Center of Excellence, and the Office of Naval Research. References 1 Michelson, A. A., Philos. Mag., 1891, 5, 338. 2 Fourier Transform Infrared Spectroscopy: Applications to Chemical Systems, ed. Ferraro, J. R., and Basile, L. J., Academic Press, New York, 1978. 3 Nordstrom, R. J., in Fourier, Hadamard and Hilbert Transforms in Chemistry, ed. Marshall, A. G., Plenum Press, New York, 1982, p. 21. 4 Saleh, B. E. A., and Teich, M. C., Fundamental of Photonics., Wiley, New York, 1991. 5 Hirschfeld, T., and Chase, D. B., Appl. Spectrosc., 1986, 40, 133. 6 Bell, R. J., Introductory Fourier Transform Spectroscopy, Academic Press, New York, 1972. 7 Brenan, C. J. H., and Hunter, I. W., Appl. Spectrosc., 1995, 49, 1086. 8 Kahn, F. D., Astrophys. J., 1959, 129, 518. 9 Zhao, J., and McCreery, R. L., Appl. Spectrosc., 1996, 50, 1209. 10 Moller, K. D., Appl. Opt., 1995, 34, 1493. 11 Junttila, M.-L., Appl. Opt., 1992, 31, 4106. 12 Okamoto, T., Kawata, S., and Minami, S., Appl. Opt., 1984, 23, 269. 13 Kay, S. M., Modern Spectral Estimation: Theory and Application, Prentice-Hall, Englewood Cliffs, NJ, 1988. 14 Korenberg, M. J., Biol. Cybern., 1989, 60, 267. 15 Raman/IR Atlas of Organic Compounds, ed. Schrader, B., and Meier, W., Verlag Chemie, Weinheim, 1977, p. A3-04. 16 Korenberg, M., in Non-linear Vision: Determination of Neural Receptive Fields, Function, and Networks, ed. Pinter, R. B., and Nabet, B., CRC Press, Boca Raton, FL, 1992, ch.7. Paper 7/00902J Received February 10, 1997 Accepted May 15, 1997 882 Analyst, September 1997, Vol. 122
ISSN:0003-2654
DOI:10.1039/a700902j
出版商:RSC
年代:1997
数据来源: RSC
|
4. |
Microfabricated Flow Chamber for Fluorescence-based Chemistries andStopped-flow Injection Cytometry |
|
Analyst,
Volume 122,
Issue 9,
1997,
Page 883-887
Peter S. Hodder,
Preview
|
|
摘要:
Microfabricated Flow Chamber for Fluorescence-based Chemistries and Stopped-flow Injection Cytometry Peter S. Hodder*a, Gert Blankensteinb and Jaromir Ruzickaa a Department of Chemistry, University of Washington, P.O. Box 351700, Seattle, WA 98195, USA b Mikroelektronik Centret (MIC), Technical University of Denmark, Building 345E, DK-2800 Lyngby, Denmark A microfabricated flow chamber (MFC) suitable for performing liquid-based fluorimetric assays is introduced. Precision delivery of microliter volumes of sample and reagent to the MFC is accomplished by a double-syringe-pump flow injection analysis (FIA) apparatus. The FIA–MFC system also combines the ‘sheath flow’ technique (traditionally used in flow cytometry) and stopped-flow FIA as a way to allow sample and reagent streams to be mixed reproducibly.The applicability of this FIA–MFC system to bioanalytical assays is demonstrated by performing an enzymatic assay with an artificial fluorigenic substrate to determine the activity of Savinase, a proteolytic enzyme.When coupled to a fluorescence microscope platform, quantitative analysis of the reaction product is possible. Experiments showed that the FIA–MFC system was capable of performing the assay with good reproducibility of injection (1.5%), and linearity of response (r2 = 0.9997) in activity ranges of analytical interest. Owing to the incorporation of flow cytometry sheath flow principles into an FIA format, the FIA–MFC system is a suitable tool for cytometric studies.Keywords: Flow injection; microfabrication; enzyme; fluorescence; microscopy; flow cytometry Recently, microfabricated devices have demonstrated promise as alternatives to conventional analytical instruments. The most obvious benefits of using such technology are the lower sample and reagent volumes necessary for analysis, with a concomitant decrease in the amount of waste generated. These benefits are of particular interest to the biotechnology community, where both small amounts of sample (DNA, cellular suspensions) and expensive reagents (enzymes, fluorescent probes, antibodies) are available for bioanalysis.In addition, since the dimensions of a typical microfabricated structure are by definition small, microscopic arrays consisting of hundreds to thousands of copies of a particular structure can be fabricated in order to perform repetitive chemistries in a highly parallel fashion.1 Researchers have conceived the concept of placing many different microfabricated structures on a single chip where sampling, sample preparation, sample delivery, reaction with reagent and detection could be integrated.Termed micro-total analytical systems’, or m-TAS,2 this concept has recently been an area of very active research. Currently most m-TAS devices are ‘hybrid’ systems, where macroscale components (e.g., lasers, photomultiplier tubes, pumps, power supplies, microscopes) are clustered around a microfabricated structure.With a variety of detection methods having been successfully demonstrated in microfabricated structures,3–5 it is now important to consider the development of suitable automation of sample and reagent handling. The method ultimately chosen should allow manipulation of a variety of fluids (aqueous or non-aqueous, filtered or containing suspended matter) in an automated and highly reproducible fashion. Flow injection analysis (FIA) is an analytical technique that has demonstrated its utility in performing a variety of chemistries.6 It has been adapted as a mode of operation for a number of m-TAS systems, and a recent review has been published on microfabricated structures that employ FIA using in situ electroosmotic pumping schemes to manipulate small volumes of aqueous solutions.7 It would be desirable with chemistries of biotechnological interest to pump organic solvents in microfabricated structures (e.g., combinatorial synthesis chemistries for drug discovery), which ranges from difficult to impossible if one uses electroosmotic flow as a pumping scheme.8 However, at present FIA-based hydrodynamic fluid manipulation in microfabricated structures is problematic, presumably because of the imprecision that results from using a traditional FIA peristaltic pump to manipulate microliter-scale volumes in microfabricated structures. In addition, advances in the design of intricate miniature hydrodynamic-based pumps that can be incorporated directly into microfabricated structures have fragility concerns associated with their deployment in a m-TAS format.9 With few exceptions, such devices are subject to fouling (through entrapment of particulate or air bubbles), back-pressure and pulsation effects.Since the foremost research goal in our laboratory is to develop a robust yet microfabricated stopped-flow cytometer, where analysis of cell suspensions via FIA techniques is possible, we have chosen FIA-compatible, high-precision stepper-motor syringe pumps that represent a useful alternative for hydrodynamic manipulation of fluid in a microfabricated structure.Computer control of the pumps allows reproducible delivery of small volumes (nl–ml) at low flow rates (nl–ml min21) without sacrificing the application of FIA techniques based on reverse or stopped-flow modes. In addition, the incorporation of robust, commercially available syringe-pumps into a m-TAS format is convenient since the pumps and valves are located remotely from the microfabricated structure, allowing easy loading and delivery of sample and reagent and fast optimization of flow parameters. The apparatus presented in this paper takes advantage of both traditional FIA methodology and microfabrication techniques to study and exploit mixing by (i) construction a microfabricated flow chamber (MFC) with a novel architecture that serves as a confluence point for three streams, (ii) design of a two-stream (sample and reagent) ‘sheath flow’ pumping scheme and (iii) employment of stopped-flow FIA methodology.The aim of this work was the construction of a hybrid mFIA system that offers the biotechnology researcher flexibility in experimental design, allowing both biochemical and cytochemical studies to be performed in a robust research apparatus that consumes microliter volumes of sample and reagent per assay.Experimental Reagents The enzyme–substrate reaction used in the FIA–MFC system is similar to a method described in detail elsewhere.10 Savinase Analyst, September 1997, Vol. 122 (883–887) 883(Novo Nordisk Industries, Copenhagen, Denmark), is a subtilisin (EC 3.4.21.14) alkaline endoprotease of the serine type. Its proteolytic activity has a broad substrate specificity, and it is used as a laundry detergent additive for removing protein-based stains. A concentrated powder containing 1% pure enzyme was used for activity determination.The activity units of the powder were stated on the vial as 3.17 KNPU g21, where KNPU (Kilo Novo Protease Unit) is the Novo-Nordisk internal unit used for expression of protease activity.11 From a stock standard solution of 0.010 KNPU ml21 (i.e., 10 mKNPU ml21), nine working standard solutions of 0.078, 0.156, 0.313, 0.625, 1.25, 2.5, 5, 8 and 10 mKNPU ml21 were prepared. TRIS-HCl Buffer (0.100 m, pH 8.3) containing 0.010 m CaCl2 was used to dissolve the enzyme and substrate solutions. 7-[N-(a-Succinyl-l-alanyl-lalanyl- l-(phenylalanyl)amino]-4-methylcoumarin is an artificial fluorigenic substrate that consists of an 7-amino-4-methylcoumarin fluorophore (AMC) covalently attached to a tripeptide substrate (AAF). The AAF–AMC substrate allows quantitative determination of fluorescence as the fluorophore is released from the tripeptide by proteolytic action.12 A working solution of substrate (4 mm) was made by first dissolving the substrate in N,N-dimethylformamide and then buffer so that the final concentration of N,N-dimethylformamide was 10% v/v.For enzyme activity determination, 1 mm substrate solutions were made from this working solution as required. This concentration was chosen so that pseudo-zero-order reaction kinetics would be maintained during the course of the assay.13 All solutions were stored refrigerated (4 °C) and the substrate stock solution was protected from light to prevent photochemical degradation.All enzymatic reactions were carried out at room temperature. For tracer studies, sodium fluorescein was dissolved in sodium borate buffer (pH 10) so that its final concentration was 5 mm. Except for the enzyme, all chemicals were acquired from Sigma (St. Louis, MO, USA). Microfabricated Flow Chamber MFC construction started with the microfabrication of the desired design in a silicon {100} chip, prepared by standard photolithography and wet-etch techniques.14 The depth of all etched channels was 50 mm and widths of the channels varied from 150 to 580 mm.These dimensions were chosen to allow pumping of solutions containing dissolved particulate and gas with minimum clogging effects. After etching, the chip was electrostatically bonded (400 °C, 1000 V) on one side to a thin glass window (about 1 mm thick). On the other side of the chip, holes were made at the channel entrances and exits, and goldplated tubes (spring contact sockets, id 0.86 mm, od 0.97 mm; Newark Electronics, Chicago, IL, USA) were epoxied into place. The spring contact sockets served as tubing connectors that allowed coupling to FIA tubing.When completed, the typical size of the entire structure (including the tubing connectors) was 13 3 8 3 15 mm (Fig. 1). Flow Injection Analysis FIA system with two Cavro XL3000 high-precision computercontrolled stepper-motor syringe pumps (Cavro Scientific Instruments, Sunnyvale, CA, USA) was used (Fig. 2). Each syringe (250 or 500 ml ) was connected to a two-way Cavro XL Smart Valve. The valves switched positions between the syringes’ respective buffer and reagent reservoirs (to load the syringes) and the MFC (to inject of their contents). The FIA system was controlled through the serial port of a Toshiba laptop computer running a QuickBasic 4.5 computer program (Microsoft, Seattle, WA, USA). The FIA system was connected to the MFC via a sample injector valve (Upchurch, Oak Harbor, WA, USA) with a 20 ml sample loop.The valve was placed physically close to the MFC to minimize dispersion of the injected sample. An Upchurch tee valve was used to split the sheath stream so that it entered the two outer inlets of the MFC. Microline tubing (Cole Parmer, Chicago, IL, USA) of 0.51 mm id was used for all fluidic connections. Except for mating the tubing to MFC tubing connectors, Upchurch Teflon flangeless fittings with ferrules were used.Fluorescence Detection The MFC was mounted on the stage of a Axiovert 100 inverted epiluminescent microscope (Carl Zeiss, Oberkochen, Germany) with a fluorescence filter set to allow detection of the AMC fluorophore. The glass window bonded to the MFC permitted focusing of the etched channels by a Zeiss 103, 0.50 NA Fluar objective. The objective was used to bring a small region of the center channel of the MFC into focus. This region was then framed in the photomultiplier tube viewfinder.This region had the dimensions of 1800 3 580 3 50 mm, which corresponds to an analytical volume of 52 nl. A fluorescence spectrometer with a 75 W xenon short-arc light source (Photon Technology International, South Brunswick, NJ, USA) was coupled to the microscope via a fiber optic cable. The monochromators of the spectrometer were set to an excitation wavelength of 370 nm Fig. 1 Top view of a microfabricated flow chamber (MFC).The MFC has three fluid inlets that impinge upon a center channel, and has two outlets. Fig. 2 Schematic diagram of FIA–MFC system. C = ‘core’ syringe (filled with carrier buffer); S = sheath syringe (filled with substrate); E = enzyme sample introduced into injector valve, W = waste vial. Box in broken lines contains MFC in the fluorescence microscope platform. M = monochromator; PMT = photomultiplier tube detector; O = microscope objective. 884 Analyst, September 1997, Vol. 122with a 20 nm bandpass to allow selective excitation of the free AMC fluorophore over excitation of the substrate–fluorophore complex. A 400 nm long pass dichroic (Chroma Technology, Brattleboro, VT, USA) and 450 nm emission filter with a 65 nm bandpass (Omega Optical, Brattleboro, VT, USA) were used to allow only the fluorescence of the free fluorophore to be detected by the photomultiplier tube. Results and Discussion Theoretical Considerations for Assay Design The MFC was designed to act as a mixing tee integrated with a flow cell where sample and reagent streams merge in such a way that the sample stream is bounded on both sides by reagent streams [Fig. 3(a)]. This containment of a central stream by an outer stream is termed ‘sheath flow’, a hydrodynamic technique traditionally employed in flow cytometry.15 Sheath flow typically consists of a faster flowing annular outer stream (‘sheath’) that surrounds a slower flowing cylindrical inner (‘core’) stream.In flow cytometry, sheath flow serves the purpose of collimating cellular suspensions so that each cell follows a narrow, well defined trajectory and thereby cell-bycell analysis is possible. Other prototypes of the MFC design take advantage of the sheath flow principle for sorting of cellular suspensions.16 In the FIA–MFC system, it is important to note that as long as the sheath and core streams flow, laminar flow prevails and only a negligible amount of diffusional mixing occurs at the periphery of the sheath and core stream interface (see equations below).Mixing of sample and reagent in flow injection systems is facilitated by injecting a plug of sample into a flowing stream of reagent and allowing sufficient dispersion to occur before the sample–reagent mixture reaches the detector.6 In stopped-flow FIA, mutually dispersed reagent and sample zones flow into a cell which is interfaced with a detector suitable for measurement of some species of the ensuing chemical reaction.17 During the stopped-flow period, the species of interest (e.g., a fluorescent product of the reaction ) is formed and measured.The advantage of the stopped-flow method in biotechnological applications is that it allows continuous monitoring of a reaction rate, a parameter of great interest, without an excessive consumption of reagent (cf., continuous flow methodologies). When addressing mixing in microfabricated structures (and in many FIA systems), consideration must be given to the behavior of fluids at micrometer dimensions if one wishes to design a functional FIA-based assay.In this context, much emphasis has been placed on the Reynolds number (Re), which can be described as the ratio of inertial to viscous forces, and is useful for describing whether a flowing liquid is turbulent (Re > 2000) or laminar (Re << 2000).18 This can be readily seen by inspection of a hydraulic equation used to calculate Re: Re = ( ) Deq av n r m (1) where Deq is the effective diameter of the structure through which the liquid is flowed, vav is the average linear velocity of the liquid, r its density and m its viscosity.Assuming the liquid is Newtonian and incompressible and low volumetric flow rates are employed ( < 1 ml min21), the Deq term dominates at micrometer dimensions and low Res result. Therefore, microfabricated structures are being designed to exploit diffusionbased mass transport since convective mixing is thought to be difficult to achieve.The Einstein–Smoluchowski equation is useful for estimating the time required for a species to diffuse a given distance: t x D = ( ) D 2 2 (2) where .(Dx)2� is the average square of the displacement in the xdirection through which diffusion will occur and D is the diffusion coefficient of a molecule in a given phase (D is typically 1025 cm2 s21 for molecules in liquids at STP). With this in mind, researchers have constructed elaborate devices for promoting diffusional mixing in liquids by creating conditions where .(Dx)2� is minimized.For example, a microfabricated multi-stage, multi-layer lamination assembly was designed to optimize diffusional mixing of species by splitting and rejoining continuously flowing sample and rea streams through a complex three-dimensional network.19 However, owing to their small channel widths, such designs may have inherent drawbacks with handling biological samples where the presence of particulate is typical.Therefore, this work exploited a different path, viz., the use of stopped-flow FIA methodology in combination with a sheath flow technique to facilitate mixing in the MFC and allow the determination of a species of interest in a biochemical reaction. In terms of traditional wet chemistry, the laminar sheath flow in the MFC can be thought of as a means to confine sample and reagent in their separate beakers until the experimenter decides that the conditions are optimum to measure a chemical reaction.By simultaneously stopping the core and sheath flows, the sheath–core boundary collapses and mixing of the now stationary sample and reagent streams occurs [Fig. 3(b)–(d)]. It is important to note that the sheath flow serves as a type of multi-lamination technique, where the width of the streams (i.e., ‘channels’ of sample and reagent) can be set by changing the ratio of sheath to core flow rates.This is advantageous for experiments where cell suspensions are used in order to avoid clogging effects while at the same time allowing mixing to occur on a convienent time scale when the sheath and core flows have stopped: the dimensions of the channel, subdivided by the three streams, promotes mixing of sample and reagent species, since they need only displace a short distance (about 200 mm) before penetrating each other. Substitution of typical experimental run parameters into eqn.(2) yields 20 s as the required time for a substrate molecule on the periphery of the core– sheath boundary to diffuse to the other side of the enzyme stream. Furthermore, experiments also showed that mixing after stopping had an additional component to mixing: the abrupt Fig. 3 Snapshots of the FIA–sheath flow principle. Images taken from center channel of MFC. Core syringe loaded with fluorescein tracer and sheath syringe loaded with buffer. The time elapsed (in seconds) from the start of the FIA assay is given at the bottom.As the sheath and core syringes flow, laminar flow is observed at t = 25 s (a), just before stopping the flow at t = 26 s. Upon stopping of syringes, mixing of the core stream into the sheath stream is recorded at successive time intervals from t = 30 s (b), 35 s (c) and 55 s (d) after the beginning of the assay. Analyst, September 1997, Vol. 122 885stopping of flow caused an expansion of the core stream into the sheath stream.This phenomenon, which lasted of the order of < 1 s, promoted reproducible mixing of the sheath and core streams. This could result from artifactual processes, such as the syringes not stopping at precisely the same time once a command is given to stop. Also, it is not possible to describe this behavior in terms of Re. Once the flow is stopped, the Reynolds conditions do not apply. The utility of the stopped-flow approach for mixing slowly diffusing species (such as microbeads with cells) will be investigated further. To run an enzymatic assay (Fig. 4), the core and sheath syringes were loaded with 33.5 ml of TRIS buffer and 67.0 ml of substrate solution, respectively. While the syringes were loading, the downstream injector valve was turned to the load position and 20 ml of enzyme standard solution were loaded into the sample loop. After the syringes had been loaded, the injector valve was turned to the inject position and the contents of the core and sheath syringes were dispensed to the MFC.To allow rapid delivery of sample and reagent to the MFC, the volumetric flow rates were chosen to be 250 ml min21 for the sheath stream and 125 ml min21 for the core stream. Note that this choice of flow rates resulted in laminar flow (Re = 2) of three equally sized streams (about 200 mm width) when viewed under the microscope. Once the syringes had finished delivering their loaded volumes, they stopped simultaneously for 50 s.This cessation was coordinated so that the FIA peak of the injected enzyme solution was trapped in the MFC. Upon stopping, the increase in fluorescence with time was measured, corresponding to the increase in concentration of released AMC fluorophore from the AAF–AMC complex. After the measuring time, the syringes were loaded again (sheath 250 ml, core 125 ml) to rinse the MFC. This was followed by an additional rinse step with 250 ml of buffer by the core syringe.The assay was then repeated for each standard. The completeness of mixing was of some concern in preliminary experiments prior to discovering the effectiveness of the stopped-flow approach. Since preliminary experiments with tracer solutions suggested that agitation-based mixing schemes would be beneficial in decreasing the mixing time, this was investigated. However, experiments incorporating these mixing schemes with the enzyme–substrate system gave similar results to the stopped flow approach, and therefore the agitationbased mixing scheme was not explored further.To increase the sensitivity of the assay at low enzyme activity, it was found that increasing the measuring time was possible until enough fluorescent product was generated by enzymatic action to be detected. Therefore, a measuring time of 50 s was chosen as a convenient time scale in order to exploit the dynamic range of the system while at the same time quantify enzyme activities of analytical interest (0.1–0.5 mKNPU ml21).20 The total volumes of substrate and buffer solution volumes were, 317 and 408 ml per assay, respectively, exceeding manyfold the volume of the flow system.The majority of this volume comes from the washing steps of the FIA routine. This was found necessary to insure that the hydrophilic protease did not adhere to the walls of the flow chamber where it would give residual activity, resulting in a carryover which was especially conspicuous if higher activity samples were assayed in succession. Although other schemes were considered, such as acidic washes to deactivate residual enzyme activity or double injection of sample and reagent, they were not explored with the FIA–MFC system for reasons of fluidic simplicity.Otherwise, a further reduction in volumes would be facilitated by replacing the traditional FIA tubing used in this research with micrometer- diameter fused silica capillary tubing typically used in capillary electrophoresis.The recent introduction of commercially available connectors21 suitable for mating fused silica tubing to injector valves, splitting tees, etc., will facilitate the further minimization of sample and reagent volumes consumed per assay. Data Analysis Fluorescence spectra were collected and analyzed using the software utility that came with the spectrometer. The window of time (Dt) used for calculation of the slope of each spectra was 50–90 s.This window was found sufficient to allow mixing to occur and enough product to be detected. Since the slope of the linear portion of each spectrum is directly proportional to the enzyme activity in pseudo-zero-order kinetics, a linear equation should result if the slope of the linear portion of each spectra is plotted versus the enzyme activity of each of the enzyme standards. As can be seen in the spectra, the 8 and 10 mKNPU ml21 standards show non-linearity of the fluorescence signal (measured in cps, photon counts per second) over the selected time window, as the fluorescent signal measured at these activities, especially with the 10 mKNPU ml21 standard, exceeds the linear range of the photomultiplier tube detector.However, a linear calibration curve ([Dcps Dt21] = 2266 [mKNPU ml21] + 93.328, r2 = 0.9997, n = 8) was calculated for the range of enzyme activities from 0.08 to 8 mKNPU ml21. With the points used, the FIA–MFC technique was linear over a two decade range.Reproducibility of injection and mixing of the sheath flow–stopped-flow technique was tested by repeating the assay with a 0.8 mKNPU ml21 standard. The reproducibility was found to be 1.5% (n = 3). Conclusion Stopped flow of streams brought together in a microfabricated flow chamber that permits monitoring of reaction products in real time is a novel approach to mixing that combines aspects of flow cytometry and FIA to allow reaction rate measurements on biochemical and biological systems.Both the design of the MFC and the choice of the model enzyme chemistry were based on the employment of a fluorescence microscope platform as a detector. The use of the fluorescence microscope as a detector reflects its importance and utility in routine biological studies and its ability to image cellular processes. The hybrid mFIA system presented in this paper takes advantage of a highprecision syringe pump FIA apparatus, microfabrication technology and fluorescence microscopy to allow both mixing and subsequent chemical analysis of small volumes of fluid in an automated, highly reproducible fashion.The MFC itself can be thought of as both a nanoliter volume mixing chamber and Fig. 4 Fluorescence spectra of 10 successive Savinase enzymatic assays in the MFC–FIA system, 0–10 mKNPU ml21. Each assay is started by injecting contents of core and sheath syringes into the MFC. Upon stopped flow, an increase in fluorescence corresponding to the generation of fluorescent product is observed.Enzyme is then rinsed out at the end of the assay. 886 Analyst, September 1997, Vol. 122microscale flow cell suitable for performing a variety of fluorescence-based assays. Since hydrodynamically based pumping schemes are employed, the FIA–MFC system is insensitive to the use of different solvents, allowing greater flexibility in the choice of reagent chemistries (cf., electroosmotic- based pumping schemes).The microfabricated structure incorporates no moving parts and is therefore robust. However, in order to access even smaller volumes of sample and reagent per assay, further optimization of the system hardware is necessary. The study of mixing and reaction kinetics carried out in this mFIA format is a stepping stone towards the development of a stopped-flow cytometer designed to study reaction kinetics in cellular suspensions. Future research will be directed towards minimizing volumes of sample and reagent used for analysis and also applying the FIA–MFC system to cell studies where the sheath flow will contain certain fluorescent probes, agonists or antagonists, while the core stream will contain cell suspensions of interest.The authors thank L. Hallgren of Novo Nordisk Industries, Denmark for providing the Savinase enzyme. In addition, acknowledgment is made to Professor G. Christian for his thoughtful discussions, Dr.D. Holman for his help in constructing the computer program to power the Cavro syringes and Dr. L. Scampavia for his role in the inital designs of MFC prototypes. This research was supported by NIHGMS grant RO1 GM45260. References 1 Borman, S., Chem. Eng. News, 1995, 73, 37. 2 Manz, A., Graber, N., and Widmer, H. M., Sens. Actuators B, 1990, 1, 244. 3 Liang, Z., Chiem, N., Ocvirk, G., Tang, T., Fluri, K., and Harrison, J. D., Anal. Chem., 1996, 68, 1040. 4 Effenhauser, C., Manz, A., Widmer, H.M., Anal. Chem., 1995, 67, 2284. 5 Reay, R. J., Flannery, A. F., Storment, C. W., Kounaves, S. P., and Kovacs, G. T. A., Sens. Actuators B, 1996, 34, 450. 6 Ruzicka, J., and Hansen, E. H., Flow Injection Analysis, Wiley, New York, 2nd edn., 1988. 7 Haswell, S. J., Analyst, 1997, 122, 1R. 8 Dasgupta, P. K., and Liu, S., Anal. Chem., 1994, 66, 1792. 9 Zengerle, R., Stehr, M., Freygang, M., Haffner, H., Messner, S., Rossberg, R., and Sandmaier, H., in AMI Special Issue mTAS ’96, ed.Widmer, H. M., Verpoorte, E., and Barnard, S., AMI Basle, 1996, pp. 91–93. 10 Ruzicka, J., and G�ubeli, T., Anal. Chem., 1991, 63, 1680. 11 Savinase Product Sheet, Novo Nordisk Industries, Copenhagen, Detergent Enzyme Division. 12 Kanaoka, Y., Takahashi, T., Nakayama, H., and Tanizawa, K., Chem. Pharm. Bull., 1985, 33, 1721. 13 Hansen, E. H., and Jensen, A., Talanta, 1993, 40, 1891. 14 Kovacs, G. T. A., Petersen, K., and Albin, M., Anal. Chem., 1996, 68, 407A. 15 Kachel, V., Fellner-Feldegg, H., and Menke, E., in Flow Cytometry and Sorting, ed. Melamed, M. R., Lindmo, T., and Mendelsohn, M. L., Wiley-Liss, New York, 2nd edn., 1990, ch. 3. 16 Blankenstein, G., Scampavia, L., Branebjerg, J., Larsen, U. D., and Ruzicka, J., in AMI Special Issue mTAS ’96, ed. Widmer, H. M., Verpoorte, E., and Barnard, S., AMI, Basle, 1996, pp. 82–84. 17 Christian, G. D., and Ruzicka, J., Anal. Chim. Acta., 1992, 261, 11. 18 Bird, R. B., Stewart, W.E., and Lightfoot, E. N., Transport Phenomena, Wiley, New York, 1960, ch. 1–2. 19 Branebjerg, J., Gravesen, P., Krog, J. P., and Nielsen, C. R., in Proceedings of the 9th International Workshop on Micro Electro Mechanical Systems, San Diego, CA, 1996, pp. 441–446. 20 Analytical Method AF 220/1-GB, Novo Nordisk Industries, Copenhagen, Enzymes Division, 1986. 21 Catalog of Chromatography and Fluid Transfer Fittings, Upchurch Scientific, Oak Harbor, WA, 1997, pp. 25–26.Paper 7/01750B Received March 12, 1997 Accepted May 22, 1997 Analyst, September 1997, Vol. 122 887 Microfabricated Flow Chamber for Fluorescence-based Chemistries and Stopped-flow Injection Cytometry Peter S. Hodder*a, Gert Blankensteinb and Jaromir Ruzickaa a Department of Chemistry, University of Washington, P.O. Box 351700, Seattle, WA 98195, USA b Mikroelektronik Centret (MIC), Technical University of Denmark, Building 345E, DK-2800 Lyngby, Denmark A microfabricated flow chamber (MFC) suitable for performing liquid-based fluorimetric assays is introduced.Precision delivery of microliter volumes of sample and reagent to the MFC is accomplished by a double-syringe-pump flow injection analysis (FIA) apparatus. The FIA–MFC system also combines the ‘sheath flow’ technique (traditionally used in flow cytometry) and stopped-flow FIA as a way to allow sample and reagent streams to be mixed reproducibly. The applicability of this FIA–MFC system to bioanalytical assays is demonstrated by performing an enzymatic assay with an artificial fluorigenic substrate to determine the activity of Savinase, a proteolytic enzyme.When coupled to a fluorescence microscope platform, quantitative analysis of the reaction product is possible. Experiments showed that the FIA–MFC system was capable of performing the assay with good reproducibility of injection (1.5%), and linearity of response (r2 = 0.9997) in activity ranges of analytical interest.Owing to the incorporation of flow cytometry sheath flow principles into an FIA format, the FIA–MFC system is a suitable tool for cytometric studies. Keywords: Flow injection; microfabrication; enzyme; fluorescence; microscopy; flow cytometry Recently, microfabricated devices have demonstrated promise as alternatives to conventional analytical instruments. The most obvious benefits of using such technology are the lower sample and reagent volumes necessary for analysis, with a concomitant decrease in the amount of waste generated.These benefits are of particular interest to the biotechnology community, where both small amounts of sample (DNA, cellular suspensions) and expensive reagents (enzymes, fluorescent probes, antibodies) are available for bioanalysis. In addition, since the dimensions of a typical microfabricated structure are by definition small, microscopic arrays consisting of hundreds to thousands of copies of a particular structure can be fabricated in order to perform repetitive chemistries in a highly parallel fashion.1 Researchers have conceived the concept of placing many different microfabricated structures on a single chip where sampling, sample preparation, sample delivery, reaction with reagent and detection could be integrated.Termed micro-total analytical systems’, or m-TAS,2 this concept has recently been an area of very active research. Currently most m-TAS devices are ‘hybrid’ systems, where macroscale components (e.g., lasers, photomultiplier tubes, pumps, power supplies, microscopes) are clustered around a microfabricated structure.With a variety of detection methods having bn successfully demonstrated in microfabricated structures,3–5 it is now important to consider the development of suitable automation of sample and reagent handling. The method ultimately chosen should allow manipulation of a variety of fluids (aqueous or non-aqueous, filtered or containing suspended matter) in an automated and highly reproducible fashion.Flow injection analysis (FIA) is an analytical technique that has demonstrated its utility in performing a variety of chemistries.6 It has been adapted as a mode of operation for a number of m-TAS systems, and a recent review has been published on microfabricated structures that employ FIA using in situ electroosmotic pumping schemes to manipulate small volumes of aqueous solutions.7 It would be desirable with chemistries of biotechnological interest to pump organic solvents in microfabricated structures (e.g., combinatorial synthesis chemistries for drug discovery), which ranges from difficult to impossible if one uses electroosmotic flow as a pumping scheme.8 However, at present FIA-based hydrodynamic fluid manipulation in microfabricated structures is problematic, presumably because of the imprecision that results from using a traditional FIA peristaltic pump to manipulate microliter-scale volumes in microfabricated structures.In addition, advances in the design of intricate miniature hydrodynamic-based pumps that can be incorporated directly into microfabricated structures have fragility concerns associated with their deployment in a m-TAS format.9 With few exceptions, such devices are subject to fouling (through entrapment of particulate or air bubbles), back-pressure and pulsation effects. Since the foremost research goal in our laboratory is to develop a robust yet microfabricated stopped-flow cytometer, where analysis of cell suspensions via FIA techniques is possible, we have chosen FIA-compatible, high-precision stepper-motor syringe pumps that represent a useful alternative for hydrodynamic manipulation of fluid in a microfabricated structure.Computer control of the pumps allows reproducible delivery of small volumes (nl–ml) at low flow rates (nl–ml min21) without sacrificing the application of FIA techniques based on reverse or stopped-flow modes.In addition, the incorporation of robust, commercially available syringe-pumps into a m-TAS format is convenient since the pumps and valves are located remotely from the microfabricated structure, allowing easy loading and delivery of sample and reagent and fast optimization of flow parameters. The apparatus presented in this paper takes advantage of both traditional FIA methodology and microfabrication techniques to study and exploit mixing by (i) construction a microfabricated flow chamber (MFC) with a novel architecture that serves as a confluence point for three streams, (ii) design of a two-stream (sample and reagent) ‘sheath flow’ pumping scheme and (iii) employment of stopped-flow FIA methodology.The aim of this work was the construction of a hybrid mFIA system that offers the biotechnology researcher flexibility in experimental design, allowing both biochemical and cytochemical studies to be performed in a robust research apparatus that consumes microliter volumes of sample and reagent per assay.Experimental Reagents The enzyme–substrate reaction used in the FIA–MFC system is similar to a method described in detail elsewhere.10 Savinase Analyst, September 1997, Vol. 122 (883–887) 883(Novo Nordisk Industries, Copenhagen, Denmark), is a subtilisin (EC 3.4.21.14) alkaline endoprotease of the serine type. Its proteolytic activity has a broad substrate specificity, and it is used as a laundry detergent additive for removing protein-based stains. A concentrated powder containing 1% pure enzyme was used for activity determination. The activity units of the powder were stated on the vial as 3.17 KNPU g21, where KNPU (Kilo Novo Protease Unit) is the Novo-Nordisk internal unit used for expression of protease activity.11 From a stock standard solution of 0.010 KNPU ml21 (i.e., 10 mKNPU ml21), nine working standard solutions of 0.078, 0.156, 0.313, 0.625, 1.25, 2.5, 5, 8 and 10 mKNPU ml21 were prepared.TRIS-HCl Buffer (0.100 m, pH 8.3) containing 0.010 m CaCl2 was used to dissolve the enzyme and substrate solutions. 7-[N-(a-Succinyl-l-alanyl-lalanyl- l-(phenylalanyl)amino]-4-methylcoumarin is an artificial fluorigenic substrate that consists of an 7-amino-4-methylcoumarin fluorophore (AMC) covalently attached to a tripeptide substrate (AAF). The AAF–AMC substrate allows quantitative determination of fluorescence as the fluorophore is released from the tripeptide by proteolytic action.12 A working solution of substrate (4 mm) was made by first dissolving the substrate in N,N-dimethylformamide and then buffer so that the final concentration of N,N-dimethylformamide was 10% v/v.For enzyme activity determination, 1 mm substrate solutions were made from this working solution as required. This concentration was chosen so that pseudo-zero-order reaction kinetics would be maintained during the course of the assay.13 All solutions were stored refrigerated (4 °C) and the substrate stock solution was protected from light to prevent photochemical degradation.All enzymatic reactions were carried out at room temperature. For tracer studies, sodium fluorescein was dissolved in sodium borate buffer (pH 10) so that its final concentration was 5 mm. Except for the enzyme, all chemicals were acquired from Sigma (St. Louis, MO, USA). Microfabricated Flow Chamber MFC construction started with the microfabrication of the desired design in a silicon {100} chip, prepared by standard photolithography and wet-etch techniques.14 The depth of all etched channels was 50 mm and widths of the channels varied from 150 to 580 mm.These dimensions were chosen to allow pumping of solutions containing dissolved particulate and gas with minimum clogging effects. After etching, the chip was electrostatically bonded (400 °C, 1000 V) on one side to a thin glass window (about 1 mm thick).On the other side of the chip, holes were made at the channel entrances and exits, and goldplated tubes (spring contact sockets, id 0.86 mm, od 0.97 mm; Newark Electronics, Chicago, IL, USA) were epoxied into place. The spring contact sockets served as tubing connectors that allowed coupling to FIA tubing. When completed, the typical size of the entire structure (including the tubing connectors) was 13 3 8 3 15 mm (Fig. 1). Flow Injection Analysis FIA system with two Cavro XL3000 high-precision computercontrolled stepper-motor syringe pumps (Cavro Scientific Instruments, Sunnyvale, CA, USA) was used (Fig. 2).Each syringe (250 or 500 ml ) was connected to a two-way Cavro XL Smart Valve. The valves switched positions between the syringes’ respective buffer and reagent reservoirs (to load the syringes) and the MFC (to inject of their contents). The FIA system was controlled through the serial port of a Toshiba laptop computer running a QuickBasic 4.5 computer program (Microsoft, Seattle, WA, USA).The FIA system was connected to the MFC via a sample injector valve (Upchurch, Oak Harbor, WA, USA) with a 20 ml sample loop. The valve was placed physically close to the MFC to minimize dispersion of the injected sample. An Upchurch tee valve was used to split the sheath stream so that it entered the two outer inlets of the MFC. Microline tubing (Cole Parmer, Chicago, IL, USA) of 0.51 mm id was used for all fluidic connections.Except for mating the tubing to MFC tubing connectors, Upchurch Teflon flangeless fittings with ferrules were used. Fluorescence Detection The MFC was mounted on the stage of a Axiovert 100 inverted epiluminescent microscope (Carl Zeiss, Oberkochen, Germany) with a fluorescence filter set to allow detection of the AMC fluorophore. The glass window bonded to the MFC permitted focusing of the etched channels by a Zeiss 103, 0.50 NA Fluar objective.The objective was used to bring a small region of the center channel of the MFC into focus. This region was then framed in the photomultiplier tube viewfinder. This region had the dimensions of 1800 3 580 3 50 mm, which corresponds to an analytical volume of 52 nl. A fluorescence spectrometer with a 75 W xenon short-arc light source (Photon Technology International, South Brunswick, NJ, USA) was coupled to the microscope via a fiber optic cable. The monochromators of the spectrometer were set to an excitation wavelength of 370 nm Fig. 1 Top view of a microfabricated flow chamber (MFC). The MFC has three fluid inlets that impinge upon a center channel, and has two outlets. Fig. 2 Schematic diagram of FIA–MFC system. C = ‘core’ syringe (filled with carrier buffer); S = sheath syringe (filled with substrate); E = enzyme sample introduced into injector valve, W = waste vial. Box in broken lines contains MFC in the fluorescence microscope platform.M = monochromator; PMT = photomultiplier tube detector; O = microscope objective. 884 Analyst, September 1997, Vol. 122with a 20 nm bandpass to allow selective excitation of the free AMC fluorophore over excitation of the substrate–fluorophore complex. A 400 nm long pass dichroic (Chroma Technology, Brattleboro, VT, USA) and 450 nm emission filter with a 65 nm bandpass (Omega Optical, Brattleboro, VT, USA) were used to allow only the fluorescence of the free fluorophore to be detected by the photomultiplier tube.Results and Discussion Theoretical Considerations for Assay Design The MFC was designed to act as a mixing tee integrated with a flow cell where sample and reagent streams merge in such a way that the sample stream is bounded on both sides by reagent streams [Fig. 3(a)]. This containment of a central stream by an outer stream is termed ‘sheath flow’, a hydrodynamic technique traditionally employed in flow cytometry.15 Sheath flow typically consists of a faster flowing annular outer stream (‘sheath’) that surrounds a slower flowing cylindrical inner (‘core’) stream.In flow cytometry, sheath flow serves the purpose of collimating cellular suspensions so that each cell follows a narrow, well defined trajectory and thereby cell-bycell analysis is possible. Other prototypes of the MFC design take advantage of the sheath flow principle for sorting of cellular suspensions.16 In the FIA–MFC system, it is important to note that as long as the sheath and core streams flow, laminar flow prevails and only a negligible amount of diffusional mixing occurs at the periphery of the sheath and core stream interface (see equations below). Mixing of sample and reagent in flow injection systems is facilitated by injecting a plug of sample into a flowing stream of reagent and allowing sufficient dispersion to occur before the sample–reagent mixture reaches the detector.6 In stopped-flow FIA, mutually dispersed reagent and sample zones flow into a cell which is interfaced with a detector suitable for measurement of some species of the ensuing chemical reaction.17 During the stopped-flow period, the species of interest (e.g., a fluorescent product of the reaction ) is formed and measured.The advantage of the stopped-flow method in biotechnological applications is that it allows continuous monitoring of a reaction rate, a parameter of great interest, without an excessive consumption of reagent (cf., continuous flow methodologies).When addressing mixing in microfabricated structures (and in many FIA systems), consideration must be given to the behavior of fluids at micrometer dimensions if one wishes to design a functional FIA-based assay. In this context, much emphasis has been placed on the Reynolds number (Re), which can be described as the ratio of inertial to viscous forces, and is useful for describing whether a flowing liquid is turbulent (Re > 2000) or laminar (Re << 2000).18 This can be readily seen by inspection of a hydraulic equation used to calculate Re: Re = ( ) Deq av n r m (1) where Deq is the effective diameter of the structure through which the liquid is flowed, vav is the average linear velocity of the liquid, r its density and m its viscosity. Assuming the liquid is Newtonian and incompressible and low volumetric flow rates are employed ( < 1 ml min21), the Deq term dominates at micrometer dimensions and low Res result.Therefore, microfabricated structures are being designed to exploit diffusionbased mass transport since convective mixing is thought to be difficult to achieve. The Einstein–Smoluchowski equation is useful for estimating the time required for a species to diffuse a given distance: t x D = ( ) D 2 2 (2) where .(Dx)2� is the average square of the displacement in the xdirection through which diffusion will occur and D is the diffusion coefficient of a molecule in a given phase (D is typically 1025 cm2 s21 for molecules in liquids at STP).With this in mind, researchers have constructed elaborate devices for promoting diffusional mixing in liquids by creating conditions where .(Dx)2� is minimized. For example, a microfabricated multi-stage, multi-layer lamination assembly was designed to optimize diffusional mixing of species by splitting and rejoining continuously flowing sample and reagent streams through a complex three-dimensional network.19 However, owing to their small channel widths, such designs may have inherent drawbacks with handling biological samples where the presence of particulate is typical.Therefore, this work exploited a different path, viz., the use of stopped-flow FIA methodology in combination with a sheath flow technique to facilitate mixing in the MFC and allow the determination of a species of interest in a biochemical reaction. In terms of traditional wet chemistry, the laminar sheath flow in the MFC can be thought of as a means to confine sample and reagent in their separate beakers until the experimenter decides that the conditions are optimum to measure a chemical reaction.By simultaneously stopping the core and sheath flows, the sheath–core boundary collapses and mixing of the now stationary sample and reagent streams occurs [Fig. 3(b)–(d)]. It is important to note that the sheath flow serves as a type of multi-lamination technique, where the width of the streams (i.e., ‘channels’ of sample and reagent) can be set by changing the ratio of sheath to core flow rates.This is advantageous for experiments where cell suspensions are used in order to avoid clogging effects while at the same time allowing mixing to occur on a convienent time scale when the sheath and core flows have stopped: the dimensions of the channel, subdivided by the three streams, promotes mixing of sample and reagent species, since they need only displace a short distance (about 200 mm) before penetrating each other.Substitution of typical experimental run parameters into eqn. (2) yields 20 s as the required time for a substrate molecule on the periphery of the core– sheath boundary to diffuse to the other side of the enzyme stream. Furthermore, experiments also showed that mixing after stopping had an additional component to mixing: the abrupt Fig. 3 Snapshots of the FIA–sheath flow principle.Images taken from center channel of MFC. Core syringe loaded with fluorescein tracer and sheath syringe loaded with buffer. The time elapsed (in seconds) from the start of the FIA assay is given at the bottom. As the sheath and core syringes flow, laminar flow is observed at t = 25 s (a), just before stopping the flow at t = 26 s. Upon stopping of syringes, mixing of the core stream into the sheath stream is recorded at successive time intervals from t = 30 s (b), 35 s (c) and 55 s (d) after the beginning of the assay.Analyst, September 1997, Vol. 122 885stopping of flow caused an expansion of the core stream into the sheath stream. This phenomenon, which lasted of the order of < 1 s, promoted reproducible mixing of the sheath and core streams. This could result from artifactual processes, such as the syringes not stopping at precisely the same time once a command is given to stop. Also, it is not possible to describe this behavior in terms of Re.Once the flow is stopped, the Reynolds conditions do not apply. The utility of the stopped-floproach for mixing slowly diffusing species (such as microbeads with cells) will be investigated further. To run an enzymatic assay (Fig. 4), the core and sheath syringes were loaded with 33.5 ml of TRIS buffer and 67.0 ml of substrate solution, respectively. While the syringes were loading, the downstream injector valve was turned to the load position and 20 ml of enzyme standard solution were loaded into the sample loop. After the syringes had been loaded, the injector valve was turned to the inject position and the contents of the core and sheath syringes were dispensed to the MFC.To allow rapid delivery of sample and reagent to the MFC, the volumetric flow rates were chosen to be 250 ml min21 for the sheath stream and 125 ml min21 for the core stream. Note that this choice of flow rates resulted in laminar flow (Re = 2) of three equally sized streams (about 200 mm width) when viewed under the microscope.Once the syringes had finished delivering their loaded volumes, they stopped simultaneously for 50 s. This cessation was coordinated so that the FIA peak of the injected enzyme solution was trapped in the MFC. Upon stopping, the increase in fluorescence with time was measured, corresponding to the increase in concentration of released AMC fluorophore from the AAF–AMC complex. After the measuring time, the syringes were loaded again (sheath 250 ml, core 125 ml) to rinse the MFC.This was followed by an additional rinse step with 250 ml of buffer by the core syringe. The assay was then repeated for each standard. The completeness of mixing was of some concern in preliminary experiments prior to discovering the effectiveness of the stopped-flow approach. Since preliminary experiments with tracer solutions suggested that agitation-based mixing schemes would be beneficial in decreasing the mixing time, this was investigated.However, experiments incorporating these mixing schemes with the enzyme–substrate system gave similar results to the stopped flow approach, and therefore the agitationbased mixing scheme was not explored further. To increase the sensitivity of the assay at low enzyme activity, it was found that increasing the measuring time was possible until enough fluorescent product was generated by enzymatic action to be detected. Therefore, a measuring time of 50 s was chosen as a convenient time scale in order to exploit the dynamic range of the system while at the same time quantify enzyme activities of analytical interest (0.1–0.5 mKNPU ml21).20 The total volumes of substrate and buffer solution volumes were, 317 and 408 ml per assay, respectively, exceeding manyfold the volume of the flow system.The majority of this volume comes from the washing steps of the FIA routine. This was found necessary to insure that the hydrophilic protease did not adhere to the walls of the flow chamber where it would give residual activity, resulting in a carryover which was especially conspicuous if higher activity samples were assayed in succession.Although other schemes were considered, such as acidic washes to deactivate residual enzyme activity or double injection of sample and reagent, they were not explored with the FIA–MFC system for reasons of fluidic simplicity. Otherwise, a further reduction in volumes would be facilitated by replacing the traditional FIA tubing used in this research with micrometer- diameter fused silica capillary tubing typically used in capillary electrophoresis.The recent introduction of commercially available connectors21 suitable for mating fused silica tubing to injector valves, splitting tees, etc., will facilitate the further minimization of sample and reagent volumes consumed per assay. Data Analysis Fluorescence spectra were collected and analyzed using the software utility that came with the spectrometer.The window of time (Dt) used for calculation of the slope of each spectra was 50–90 s. This window was found sufficient to allow mixing to occur and enough product to be detected. Since the slope of the linear portion of each spectrum is directly proportional to the enzyme activity in pseudo-zero-order kinetics, a linear equation should result if the slope of the linear portion of each spectra is plotted versus the enzyme activity of each of the enzyme standards.As can be seen in the spectra, the 8 and 10 mKNPU ml21 standards show non-linearity of the fluorescence signal (measured in cps, photon counts per second) over the selected time window, as the fluorescent signal measured at these activities, especially with the 10 mKNPU ml21 standard, exceeds the linear range of the photomultiplier tube detector. However, a linear calibration curve ([Dcps Dt21] = 2266 [mKNPU ml21] + 93.328, r2 = 0.9997, n = 8) was calculated for the range of enzyme activities from 0.08 to 8 mKNPU ml21.With the points used, the FIA–MFC technique was linear over a two decade range. Reproducibility of injection and mixing of the sheath flow–stopped-flow technique was tested by repeating the assay with a 0.8 mKNPU ml21 standard. The reproducibility was found to be 1.5% (n = 3). Conclusion Stopped flow of streams brought together in a microfabricated flow chamber that permits monitoring of reaction products in real time is a novel approach to mixing that combines aspects of flow cytometry and FIA to allow reaction rate measurements on biochemical and biological systems.Both the design of the MFC and the choice of the model enzyme chemistry were based on the employment of a fluorescence microscope platform as a detector. The use of the fluorescence microscope as a detector reflects its importance and utility in routine biological studies and its ability to image cellular processes.The hybrid mFIA system presented in this paper takes advantage of a highprecision syringe pump FIA apparatus, microfabrication technology and fluorescence microscopy to allow both mixing and subsequent chemical analysis of small volumes of fluid in an automated, highly reproducible fashion. The MFC itself can be thought of as both a nanoliter volume mixing chamber and Fig. 4 Fluorescence spectra of 10 successive Savinase enzymatic assays in the MFC–FIA system, 0–10 mKNPU ml21.Each assay is started by injecting contents of core and sheath syringes into the MFC. Upon stopped flow, an increase in fluorescence corresponding to the generation of fluorescent product is observed. Enzyme is then rinsed out at the end of the assay. 886 Analyst, September 1997, Vol. 122microscale flow cell suitable for performing a variety of fluorescence-based assays. Since hydrodynamically based pumping schemes are employed, the FIA–MFC system is insensitive to the use of different solvents, allowing greater flexibility in the choice of reagent chemistries (cf., electroosmotic- based pumping schemes).The microfabricated structure incorporates no moving parts and is therefore robust. However, in order to access even smaller volumes of sample and reagent per assay, further optimization of the system hardware is necessary. The study of mixing and reaction kinetics carried out in this mFIA format is a stepping stone towards the development of a stopped-flow cytometer designed to study reaction kinetics in cellular suspensions.Future research will be directed towards minimizing volumes of sample and reagent used for analysis and also applying the FIA–MFC system to cell studies where the sheath flow will contain certain fluorescent probes, agonists or antagonists, while the core stream will contain cell suspensions of interest. The authors thank L. Hallgren of Novo Nordisk Industries, Denmark for providing the Savinase enzyme. In addition, acknowledgment is made to Professor G. Christian for his thoughtful discussions, Dr. D. Holman for his help in constructing the computer program to power the Cavro syringes and Dr. L. Scampavia for his role in the inital designs of MFC prototypes. This research was supported by NIHGMS grant RO1 GM45260. References 1 Borman, S., Chem. Eng. News, 1995, 73, 37. 2 Manz, A., Graber, N., and Widmer, H. M., Sens. Actuators B, 1990, 1, 244. 3 Liang, Z., Chiem, N., Ocvirk, G., Tang, T., Fluri, K., and Harrison, J. D., Anal. Chem., 1996, 68, 1040. 4 Effenhauser, C., Manz, A., Widmer, H. M., Anal. Chem., 1995, 67, 2284. 5 Reay, R. J., Flannery, A. F., Storment, C. W., Kounaves, S. P., and Kovacs, G. T. A., Sens. Actuators B, 1996, 34, 450. 6 Ruzicka, J., and Hansen, E. H., Flow Injection Analysis, Wiley, New York, 2nd edn., 1988. 7 Haswell, S. J., Analyst, 1997, 122, 1R. 8 Dasgupta, P. K., and Liu, S., Anal. Chem., 1994, 66, 1792. 9 Zengerle, R., Stehr, M., Freygang, M., Haffner, H., Messner, S., Rossberg, R., and Sandmaier, H., in AMI Special Issue mTAS ’96, ed. Widmer, H. M., Verpoorte, E., and Barnard, S., AMI Basle, 1996, pp. 91–93. 10 Ruzicka, J., and G�ubeli, T., Anal. Chem., 1991, 63, 1680. 11 Savinase Product Sheet, Novo Nordisk Industries, Copenhagen, Detergent Enzyme Division. 12 Kanaoka, Y., Takahashi, T., Nakayama, H., and Tanizawa, K., Chem. Pharm. Bull., 1985, 33, 1721. 13 Hansen, E. H., and Jensen, A., Talanta, 1993, 40, 1891. 14 Kovacs, G. T. A., Petersen, K., and Albin, M., Anal. Chem., 1996, 68, 407A. 15 Kachel, V., Fellner-Feldegg, H., and Menke, E., in Flow Cytometry and Sorting, ed. Melamed, M. R., Lindmo, T., and Mendelsohn, M. L., Wiley-Liss, New York, 2nd edn., 1990, ch. 3. 16 Blankenstein, G., Scampavia, L., Branebjerg, J., Larsen, U. D., and Ruzicka, J., in AMI Special Issue mTAS ’96, ed. Widmer, H. M., Verpoorte, E., and Barnard, S., AMI, Basle, 1996, pp. 82–84. 17 Christian, G. D., and Ruzicka, J., Anal. Chim. Acta., 1992, 261, 11. 18 Bird, R. B., Stewart, W. E., and Lightfoot, E. N., Transport Phenomena, Wiley, New York, 1960, ch. 1–2. 19 Branebjerg, J., Gravesen, P., Krog, J. P., and Nielsen, C. R., in Proceedings of the 9th International Workshop on Micro Electro Mechanical Systems, San Diego, CA, 1996, pp. 441–446. 20 Analytical Method AF 220/1-GB, Novo Nordisk Industries, Copenhagen, Enzymes Division, 1986. 21 Catalog of Chromatography and Fluid Transfer Fittings, Upchurch Scientific, Oak Harbor, WA, 1997, pp. 25–26. Paper 7/01750B Received March 12, 1997 Accepted May 22, 1997 Analyst, September 1997, Vol. 122 8
ISSN:0003-2654
DOI:10.1039/a701750b
出版商:RSC
年代:1997
数据来源: RSC
|
5. |
Determination of Phenoxy Acid Herbicides From Aqueous Samples byImproved Clean-up on Polymeric Pre-columns at High pH |
|
Analyst,
Volume 122,
Issue 9,
1997,
Page 889-894
René B. Geerdink,
Preview
|
|
摘要:
Determination of Phenoxy Acid Herbicides From Aqueous Samples by Improved Clean-up on Polymeric Pre-columns at High pH Ren�e B. Geerdink*a, Sylvia van Tol-Wildenburga, Wilfried M. A. Niessenb and Udo A. Th. Brinkmanb a RIZA, P.O. Box 17, 8200 AA Lelystad, The Netherlands b Department of Analytical Chemistry, Free University, De Boelelaan 1083, 1081 HV Amsterdam, The Netherlands An improved procedure for the determination of phenoxy acid herbicides in environmental water samples is reported.The procedure consists in solid-phase extraction (SPE) of 60 ml water samples on a polymeric pre-column at pH 2.8, a clean-up step at high pH and subsequent desorption and ion-pair LC separation at pH 8.8. The main improvements are in the basic clean-up step and in the LC eluent composition. The release of compounds which are electrostatically bound to the precolumn is favoured by a washing step with 0.1 mol l21 sodium hydroxide solution. As regards LC, gradient elution is applied using solvents with low buffer and ion-pairing concentrations.The detection limits for the phenoxy acids (UV detection at 232 nm) are 5–20 ng l21 for tap water samples. At the 0.1 mg l21 spiking level, the RSD is 6% (n = 7) and the recoveries are better than 83% for all analytes. The long-term reproducibility typically has RSD values of 5% (n = 7). The method was successfully tested on water samples from various origins, and the results obtained with the present on-line SPE–LC–UV procedure were found to compare well with those obtained with procedures involving SPE combined off-line with GC–MS or flow injection MS–MS.Keywords: Phenoxy acid herbicides; solid-phase extraction–liquid chromatography–UV detection; high-pH clean-up; aqueous samples In our laboratory, the determination of phenoxy acid herbicides (for names and structures, see Table 1) in surface water is performed according to a previously described method.1,2 The method essentially consists in on-line analyte enrichment from a small sample volume at pH 3, pre-column clean-up and an isocratic ion-pair LC separation on a polymeric column.However, over the years some of the initial conditions were changed because of maintenance problems. As an example, the high buffer salt concentration and the high pH of the washing solvent caused rapid deterioration of the LC pumps and valves, and were therefore modified. In addition, in order to simplify the procedure, heart-cutting of the pre-column desorption solvent2 was omitted.All modifications were evaluated by testing spiked tap water samples and comparing the results obtained with the initial conditions. In general, the modified procedures could be used successfully for analyses of tap water and water from the main Dutch rivers, the Rhine and Meuse. However, surprisingly, the analysis of spiked surface water samples showed large differences in analyte recoveries depending on the origin of the sample.Problems were experienced especially with samples from rural areas and were expected to be due to the presence of relatively large amounts of humic material. In order to improve the reliability of the procedure and to enhance the recovery of the analytes in, especially, the ruralarea samples, the total analytical procedure has now been further modified and optimized. As a check, the results of the final solid-phase extraction (SPE)–LC–UV procedure were compared with those obtained by means of flow injection analysis (FIA)– MS–MS and GC–MS.Experimental Reagents All chemicals and solvents were of analytical-reagent grade. HPLC-grade methanol, acetonitrile, sodium hydroxide, ammo- Table 1 Structures of phenoxy acid herbicides and bentazone Analyte Analyte structure Abbreviation Peak No. pKa 2 4 5 R CAS No. MCPA 1 3.1 CH3 Cl H CH2COOH 94-74-6 MCPP 2 3.6 Cl Cl H CH2COOH 94-75-7 2,4-D 3 2.8 CH3 Cl H CH(CH3)COOH 7085-19-0 2,4-DP 4 2.7 Cl Cl H CH(CH3)COOH 120-36-5 Bentazone 5 3.2 25057-89-0 MCPB 6 4.6 Cl Cl Cl CH2COOH 93-76-5 2,4-DB 7 4.4 CH3 Cl H (CH2)3COH 94-81-5 2,4,5-T 8 2.8 Cl Cl H (CH2)3COH 94-82-6 2,4,5-TP 9 3.6 Cl Cl Cl CH(CH3)COOH 93-72-1 Analyst, September 1997, Vol. 122 (889–893) 889nium acetate, perchloric acid and water were obtained from J. T. Baker (Deventer, The Netherlands) and tetrabutylammonium hydrogensulfate from Fluka (Buchs, Switzerland). Chlorophenoxycarboxylic acids were obtained from Riedel-de Ha�en (Hannover, Germany) and Promochem (Wesel, Germany) and bentazone from Promochem. Equipment A Waters (Millipore, Bedford, MA, USA) Model 600E pump was used to deliver the eluent.A fully automated Prospekt (Spark, Emmen, The Netherlands) cartridge exchange system with two additional valves was used to control the flow scheme during analysis. The solvent delivery unit (SDU) of the Prospekt was provided with two additional six-port solvent selection valves for the wetting and conditioning of the pre-columns, the washing solvents and the aqueous samples.A Model 785A programmable absorbance detector (Applied Biosystems, H. I. Ambacht, The Netherlands) was operated at 232 nm. The column thermostat was a Mistral from Spark. Chromatograms were recorded and integrated by a Millipore data station using Baseline 810 or Millenium 2020 software. For MS detection, a Finnigan TSQ-700 mass spectrometer (Finnigan MAT, San Jose, CA, USA) equipped with a Finnigan thermospray interface (TSP-2) was used; the experimental conditions were as in previous work.3 Procedures Stock standard solutions of the chlorophenoxy acids and bentazone were prepared by weighing approximately 5 mg of each component and dissolution in 50 ml of methanol; they were stored at 220 °C.These solutions were diluted with ultrapure water and acidified to pH 2.8 with 0.1 mol l21 HClO4, to obtain individual and mixed working standard solutions which were stored at 4 °C.Spiked sample solutions were prepared by diluting the stock standard solutions with tap or surface water and acidifying them to pH 2.8 with 0.1 mol l21 HClO4. Pre-columns (10 3 2 mm id), pre-packed with 15–25 mm PLRP-S styrene–divinylbenzene copolymer (Polymer Laboratories, Church Stretton, UK), were wetted with 2 ml of methanol (2 ml min21) and activated with 2 ml of 0.001 mol l21 HClO4 (1 ml min21) prior to use. Preconcentration of the samples (60 ml) was performed at 2 ml min21.Details of the clean-up procedure are described below. LC separations were carried out at 50 °C on a 2503 4.6 mm id column pre-packed with 5 mm, 100 Å PLRP-S using an acetonitrile–water gradient (for details, see Table 2 and later text). Results and Discussion The trace-level determination of phenoxy acids in surface water with an isocratic LC separation often creates problems because most analyte peaks are eluted on a sharply decreasing baseline caused by a high humic acid hump in the early part of the chromatogram.This is demonstrated in Fig. 1, where the chromatogram of a standard solution is compared with that of a surface water sample. It will be obvious that the presence of the hump will cause difficulties during peak integration. Moreover, changing the sample volume applied to the pre-column causes a shift in the retention time of the MCPB and 2,4-DB (peaks 7 and 8), which are co-eluted in the system.This is demonstrated in Fig. 2, in which three tap water samples acidified to pH 3, of 30, 60 and 150 ml were applied to the pre-column and subsequently subjected to separation by LC. One should also note that, even with tap water samples, the baseline is not straight because of the presence of the dissolved organic carbon (DOC) which is not removed during tap water preparation from surface water. As is to be expected, the peak areas of all compounds of interest increased with increasing sample volume, i.e., they were twoand five-fold higher for the 60 and 150 ml samples, respectively, compared with the 30 ml sample.However, with the 150 ml sample, the peaks of MCPB and 2,4-DB shifted to longer retention times and were now co-eluted with 2,4,5-TP (peak 9). One explanation of the shift is that it is due to the amount of acid applied to the pre-column. Obviousa larger volume of highpH solution will be necessary to ionize the compounds among which MCPB and 2,4-DB have the highest pKa values.Desorption with the ion-pair-containing acetonitrile–water (30 + 70 v/v) eluent at pH 8.8 does not occur sufficiently rapidly for these two compounds, although the pH used ensures complete dissociation of all compounds with pKa 2.7–4.6. This Table 2 Gradient elution procedure D Time/ Flow rate/ min ml min21 A* (%) B* (%) 0 1.0 70 30 5 70 30 25 30 70 26 70 30 30 70 30 60 0.05 70 30 * Solvent A: acetonitrile–water (20 + 80 v/v) with 0.0125 mol l21 borate buffer in aqueous phase and 0.003 mol l21 TBA.Solvent B: acetonitrile–water (40 + 60 v/v) with 0.0125 mol l21 borate buffer in aqueous phase and 0.003 mol l21 TBA. Fig. 1 SPE–LC–UV of acid herbicides after preconcentration of (A) surface water and (B) a standard solution. Conditions: 60 ml sample, spike 1.0 mg l21; washing solvent, acetonitrile–water (4 + 96 v/v), pH 3, 1 ml; system, 10 3 2 mm id pre-column packed with 20 mm PLRP-S, 250 3 4.6 mm id analytical column packed with 5 mm PLRP-S; eluent, acetonitrile– water (30 + 70 v/v), containing 0.0125 mol l21 TB and 0.003 mol l21 borate buffer, pH 8.8, flow-rate 1.0 ml min21; UV detection, 232 nm.Peaks: 1, MCPA; 2, 2,4-D; 3, MCPP; 4, 2,4-DP; 5, bentazone; 6, 2,4,5-T; 7/8, MCPB/ 2,4-DB; 9, 2,4,5-TP. Fig. 2 SPE–LC–UV of acid herbicides after preconcentration of (A) 30, (B) 60 and (C) 150 ml of tap water spiked with 0.25 mg l21 of each component. For conditions, see Fig. 1. At asterisk: MCPB/2,4-DB. 890 Analyst, September 1997, Vol. 122suggests that the buffer strength may be an important parameter. In order to improve the LC separation, several conditions of the procedure were re-examined. Analytical LC Separation In Table 3 the conditions from refs. 1 and 2 (columns A and B, respectively) and the initial conditions used in this work (column C) are summarized. As discussed in the Introduction, the conditions of column C are the result of earlier improvements.First, the influence of the pH used during sorption, i.e., the pH of the sample, and the pH used during desorption, i.e., the pH of the eluent, were studied in the ranges 2.0–3.5 and 7.5–9.5, respectively. The results showed that a low sample pH (2.0) gave less reproducible results than higher pHs: the RSD values of the retention times of the test analytes were about 5% and 0.1% at pH 2.0 and 2.8, respectively (n = 7). On the other hand, varying the eluent pH over the range indicated had no significant effect.Next, gradient elution was introduced and the ion-pair reagent (TBA) and the borate buffer salt concentrations were optimized in order to obtain a low and non-fluctuating baseline during elution of the analytes of interest. The TBA concentration was studied in the range 0.001–0.006 mol l21 and the borate concentration in the range 0.005–0.025 mol l21. The proper eluent composition for separation turned out to be a linear gradient for 20 min from 26 + 74 to 34 + 66 v/v acetonitrile– water with constant concentrations of 0.003 mol l21 TBA and 0.0125 mol l21 borate buffer of pH 8.8 (Table 3, column D).With these conditions, chromatograms of tap, surface and rural water samples all showed the desired baseline during analyte elution, even though there still was a baseline offset due to matrix constituents with surface and rural water. Influence of Matrix Constituents The trace-level determination of phenoxy acids generally requires extensive clean-up to remove humic constituents.These interfering compounds with their many acidic and phenolic substituents are difficult to separate from the similar phenoxy acids. In actual practice, pre-treatment techniques such as ultrafiltration,2 on-line dialysis with (hollow-fibre) membranes4 or a dual pre-column procedure (PLRP-S and anion exchanger)5 are not sufficient for the analysis of surface water with concentrations down to 0.1 mg l21.With tap and groundwater, detection limits of @0.1 mg l21 are more easily achieved with off-line SPE on C18-bonded silica at neutral pH using an ion-pairing mechanism.6 In another study on the preconcentration of a broad range of acidic, basic and neutral compounds from surface water adjusted to pH 13,7 the selectivity did improve considerably, but only 2–5 ml samples could be handled without breakthrough. This resulted in poor detection limits of 0.5–1.3 mg l21. As an alternative, in a previous paper2 pre-column clean-up with up to 7 ml of 0.1 mol l21 aqueous sodium hydroxide was described. At that time, the LC eluent contained a commercial ion-pair reagent, low-UV PIC A (Waters), with a high buffer salt concentration that caused rapid deterioration of the pump heads.The target analytes were, admittedly, quantitatively recovered from the pre-column, probably owing to the additional ‘mass effect’ of the buffer salt; however, the humic acid hump in the chromatogram was still pronounced.In addition, owing to the high maintenance costs, the robustness of the procedure was not satisfactory. As regards alternatives, the use of a clean-up procedure with a solution containing 1–5% of acetonitrile did not cause breakthrough of the analytes. However, the recoveries of the analytes (spiked at 0.25 and 1.0 mg l21) from rural samples were not constant and often much lower than those from tap water. Typical examples are presented in Table 4.In several instances, the analyte recoveries were below 60%; however, duplicate runs gave the same result, indicating that the nature of the sample and not the analytical procedure was the main cause of the problem. Several of the above samples were also subjected to analysis by FIA–MS–MS, a rapid screening procedure which was described in detail in a previous paper.3 To illustrate the problems then encountered, in one sample 3 mg l21 of bentazone was found with FIA–MS–MS, whereas there was no peak at the retention time of bentazone in the LC–UV trace.This, and other, false-negative results indicate that the phenoxy acids are not, or incompletely, sorbed from these samples on the pre-column or, more likely, that they are sorbed on the pre-column but not desorbed by the LC eluent. With the latter explanation, the surface of the polymeric sorbent is considered to become coated by humic substances during the preconcentration process, which now acts as a (modified) stationary phase extracting the analytes of interest.Since bonding to this humic phase is stronger than bonding to the polymeric phase, analyte recoveries will be low(er). Moreover, in such a situation conventional clean-up procedures will not be successful because, with the release of the matrix constituents, the analytes will also be released and recoveries will be poor. Therefore, a washing step with water containing a few per cent of acetonitrile will only be efficient if relatively clean samples such as tap and ground Table 3 Schematic representation of methods used to determine phenoxy acids and bentazone Procedure Parameter A B C D LC Eluent Acetonitrile–water (30 +70) Acetonitrile–water (30 +70) Acetonitrile–water (30 +70) A: acetonitrile–water (20 + 80) B: acetonitrile–water (40 + 60) Elution Isocratic Isocratic Isocratic Gradient (see Table 2) Buffer No Low-UV PIC A reagent 0.0125 mol l21 borate 0.0125 mol l21 borate pH 11 8.3 8.8 8.8 Ion pair 0.01 mol l21 TBA 0.005 mol l21 TBA 0.003 mol l21 TBA 0.003 mol l21 TBA Sample pH 3 3 2.8 2.8 Heart-cut No Yes No No Clean-up 100 ml acetonitrile–water 1 ml 0.1 mol l21 sodium 1 ml acetonitrile–water 1 ml sodium hydroxide (30 + 70), pH 3 hydroxide solution, (4 + 96), pH 3 solution, pH 11 pH 12.5 Analyst, September 1997, Vol. 122 891water are analysed and matrix constituents are not dominant. Rural water samples, however, need more drastic clean-up procedures such as high-pH washing.High-pH Clean-up The pre-column washing with up to 7 ml of 0.1 mol l21 sodium hydroxide solution referred to above was followed by heartcutting of the pre-column eluent; this caused a considerable decrease in the matrix interferences. The explanation then provided was that the humic substances are highly polar molecules which were easily eluted from the pre-column without dragging along the analytes of interest. However, this explanation needs further extension.Application of a high-pH solvent to the pre-column will result in a charged humic phase. This phase can still interact with the, also charged, phenoxy acids and the polymeric sorbent, the interaction being due to charge-transfer complexation between the aromatic styrene– divinylbenzene copolymer with its p-electron pairs and the electrophilic humic and phenoxy acids.8 The binding energies of these interactions on polymeric pre-columns are higher than hydrophobic interactions,9 which are the main retention forces if C18-bonded silica is applied for the analysis of water for these compounds. This may also explain why such large volumes of high-pH washing solvents (7–10 ml) can be used without analyte breakthrough.The explanation also supports the results of Aiken et al.,10 who reported that the order of breakthrough of fulvic acids on polymeric resins at pH 13 is the same as that at pH 2 (note that C18-type phases show a different retention mechanism with humic-bound compounds passing through and non-bound compounds being retained11,12).Desorption of the, now very polar, pre-column with the aqueous pH 8.8 eluent resulted in the quantitative and instantaneous release of all phenoxy compounds and the remaining humic material. In summary, application of 1 ml of sodium hydroxide solution (pH 11) for clean-up and the linear LC gradient of procedure D for separation gave excellent baselines and good recoveries for all types of water samples studied.Fig. 3 shows a typical example of a tap and a rural-area water sample. Table 5 shows that the recoveries from the rural samples determined with gradient elution (procedure D) are much higher than those with the isocratic procedure C. The recoveries are essentially quantitative (80–110%) for all but one analyte (bentazone, 70–80%). In other words, the desired rapid release of the analytes is indeed effected by the introduction of the high-pH washing step. Analytical Data The limits of detection, set at three times the standard deviation (n = 7), which were determined using tap water spiked at the 0.01–0.02 mg l21 level, were 5–20 ng l21 for all nine analytes (60 ml samples).The repeatability and recovery of the procedure were determined with tap water spiked at the 0.1 mg l21 level. The recoveries invariably were > 83% with RSD values of 3–7% (n = 7). The reproducibility of the procedure was tested by analysing one freshly prepared sample each week (spiking level 0.1 mg l21), for seven weeks; the RSD values were 2–10%.The method was also tested by spiking several samples from a rural area at three levels, 0.25, 0.5 and 1.0 mg l21. The samples were analysed in our laboratory with the optimized SPE–LC– UV procedure D and with FIA–MS–MS. The sample set was also analysed by a contract agency which used a GC–MS procedure. Table 6 shows that the present procedure works well at all levels with mean recoveries > 80%.At the 0.25 mg l21 level, the recovery for some of the compounds is lower; however, these recoveries are comparable with the results of the other procedures. The precision of the method (reference material from the contract laboratory at 0.2 mg l21) is excellent with recoveries of 97–109% and RSDs of 4–6% (n = 5). In general, the SPE–LC–UV results compare well with those from the FIA–MS–MS procedure (in which an internal standard is used) at all levels tested, whereas the GC–MS results were Table 4 Analyte recoveries (%) from rural-area samples spiked at the 0.25 and 1.0 mg l21 levels and determined with procedure C.Values corrected for blank values Sample 1 Sample 2 Sample 3 Analyte +0.25 +1.0 +0.25 +1.0 +0.25 +1.0 MCPA 68 74 60 53 62 64 MCPP 52 73 56 64 57 76 2,4-D 60 67 46 62 53 62 2,4-DP 56 75 47 64 52 75 MCPB/2,4-DB 76 85 58 82 61 80 2,4,5-T 42 107 64 63 102 62 2,4,5-TP 32 65 40 77 48 66 Bentazone 76 87 60 86 60 82 Fig. 3 SPE–LC–UV of acid herbicides after preconcentration of 60 ml of rural (top) and tap (bottom) water spiked with 0.25 mg l21 of each component, using procedure D. System: 10 3 2 mm id pre-column packed with 20 mm PLRP-S, 250 3 4.6 mm id analytical column packed with 5 mm PLRP-S; UV detection, 232 nm. For gradient elution conditions, see Table 2. Peaks: 1, MCPA; 2, 2,4-D; 3, MCPP; 4, 2,4-DP; 5, bentazone; 6, 2,4,5-T; 7/8, MCPB/2,4-DB; 9, 2,4,5-TP. Table 5 Analyte recoveries (%) from rural-area samples spiked at the 1.0 mg l21 level and determined with procedures C and D.Values corrected for blank values Sample No. Analyte Procedure 1 2 3 4 5 MCPA C 50 66 72 64 76 D 85 79 83 91 78 MCPP C 85 91 84 88 73 D 106 83 103 98 95 2,4-D C 62 59 54 61 54 D 91 78 86 95 83 2,4-DP C 79 85 80 80 76 D 105 80 112 101 101 MCPB/2,4-DB C 89 87 64 87 86 D 95 76 100 95 76 2,4,5-T C 86 75 67 85 –* D 87 92 75 87 89 2,4,5-TP C 88 90 –* 88 81 D 110 81 –* 108 120 Bentazone C 58 65 55 65 51 D 70 84 77 71 77 * – Not determined. 892 Analyst, September 1997, Vol. 122sometimes disappointing, probably owing to low conversion yields during derivatization. Routine Procedure D has been in use for about 2 years and is one of the ISO 9001/9002 methods in our laboratory. About 100 samples are analysed each year and the robustness of the procedure turned out to be excellent. Fig. 4 shows an example of an SPE–LC–UV trace from a rural water sample.In this sample, four compounds are detected at concentrations of 0.05–0.1 mg l21. From Fig. 4, it is clear that at this level these compounds are readily detected, owing to the improved clean-up of the pre-column. Conclusions Two previously reported procedures for the trace-level determination of phenoxy acid herbicides and bentazone were evaluated and the conditions changed in order to improve the practicability for a variety of water samples. The introduction of an LC gradient instead of isocratic elution caused a less fluctuating baseline, which resulted in improved detection limits.Lowering the buffer salt and ion-pair concentrations helped to prolong the lifetime of the LC system. Washing the pre-column with water containing 4% acetonitrile turned out to be effective and reliable only if relatively clean samples were analysed. Washing the pre-column with a high-pH solution, which induces the formation of chargetransfer complexes between the humic phase and the polymeric sorbent, turned out to be a much more powerful option.The compounds of interest and the remaining humic constituents are instantaneously released from this phase on introducing the LC eluent. The analyte recovery is now satisfactory ( > 70%) even for samples with a high humic content. The repeatability and reproducibility of the total analytical procedure are good and detection limits for 60 ml samples are as low as 5–20 ng l21.The results obtained for spiked rural-area samples with the present SPE–LC–UV procedure turned out to be essentially the same as those with a (much more expensive) FIA–MS–MS procedure and at least as good as those with a conventional GC–MS procedure, which has the disadvantage of requiring derivatization. References 1 Geerdink, R. B., van Balkom, C. A. A., and Brouwer, H.-J., J. Chromatogr., 1989, 481, 275. 2 Geerdink, R. B., Graumans, A. M. B. C., and Viveen, J., J. Chromatogr., 1991, 547, 478. 3 Geerdink, R. B., Kienhuis, P. G. M., and Brinkman, U. A. Th., Chromatographia, 1994, 39, 311. 4 van de Merbel, N. C. F., Lagerwerf, M., Lingeman, H., and Brinkman, U. A. Th., Int. J. Environ. Anal. Chem., 1994, 54, 105. 5 Vera-Avila, L. E., Padilla, P. C., Hernandez, M. G., and Meraz, J. L. L., J. Chromatogr. A, 1996, 731, 115. 6 Ballinova, A., J. Chromatogr. A, 1996, 728, 319. 7 Pijlman, L., Technical Report, RIZA, Lelystad, 1991. 8 LeCloire, P., Lelacheur, R.M., Johnson, J. D., and Christman, R. F., Water Res., 1990, 24, 1151. 9 McDowell, R. D., LC–GC Int., 1994, 7, 638. 10 Aiken, G. R., Thurman, E. M., Malcolm, R. L., and Walton, H. F., Anal. Chem., 1979, 51, 1799. 11 Gremm, T., Huber, S., and Frimmel, F. H., Vom Wasser, 1993, 80, 109. 12 Landrum, P. F., Nihart, S. R., Eadle, B. J., and Gardner, W. S., Environ. Sci. Technol., 1984, 18, 187. Paper 7/02338C Received April 7, 1997 Accepted May 30, 1997 Table 6 Recovery and precision, expressed as RSD, at 0.2 mg l21 and mean results (mg l21) for phenoxy acids and bentazone spiked at 0.25, 0.5 and 1.0 mg l21 in four rural-area samples and determined with FIA–MS–MS (A), SPE–LC–UV (B) and GC–MS (C).Values corrected for blank values Spike 1.0 mg l21 Spike 0.5 mg l21 Spike 0.25 mg l21 Recovery RSD Analyte (%) (%) A B C A B C A B C MCPA 104 5.9 0.79 0.94 0.78 0.39 0.42 0.39 0.19 0.18 0.21 MCPP 101 3.3 0.79 0.73 0.82 0.43 0.42 0.45 0.21 0.20 0.22 2,4-D 102 6.2 0.98 0.90 0.58 0.50 0.49 0.36 0.23 0.20 0.18 MCPB –* – 0.63 – 0.42 0.34 – 0.24 0.17 – 0.14 2,4-DP 97 4.3 0.83 1.03 0.71 0.45 0.50 0.36 0.30 0.26 0.21 2,4-DB 109 4.9 0.78 0.94 0.32 0.46 0.49 0.18 0.14† 0.20 0.15‡ 2,4,5-T 99 4.1 1.05 0.84 0.37 0.50 0.44 0.23 0.23 0.19 0.16 2,4,5-TP 104 5.8 0.79 0.87 0.52 0.42 0.50 0.31 0.32 0.28‡ 0.19 Bentazone 109 4.0 0.72 0.66 0.24 0.41 0.43 0.14 0.20 0.17 0.13‡ * Not determined.† Not detected in two samples; average for two samples. ‡ Not detected in one sample; average for two samples.Fig. 4 SPE–LC–UV of 60 ml of rural water using procedure D. For conditions, see Fig. 3. Analyst, September 1997, Vol. 122 893 Determination of Phenoxy Acid Herbicides From Aqueous Samples by Improved Clean-up on Polymeric Pre-columns at High pH Ren�e B. Geerdink*a, Sylvia van Tol-Wildenburga, Wilfried M. A. Niessenb and Udo A. Th. Brinkmanb a RIZA, P.O. Box 17, 8200 AA Lelystad, The Netherlands b Department of Analytical Chemistry, Free University, De Boelelaan 1083, 1081 HV Amsterdam, The Netherlands An improved procedure for the determination of phenoxy acid herbicides in environmental water samples is reported. The procedure consists in solid-phase extraction (SPE) of 60 ml water samples on a polymeric pre-column at pH 2.8, a clean-up step at high pH and subsequent desorption and ion-pair LC separation at pH 8.8.The main improvements are in the basic clean-up step and in the LC eluent composition.The release of compounds which are electrostatically bound to the precolumn is favoured by a washing step with 0.1 mol l21 sodium hydroxide solution. As regards LC, gradient elution is applied using solvents with low buffer and ion-pairing concentrations. The detection limits for the phenoxy acids (UV detection at 232 nm) are 5–20 ng l21 for tap water samples. At the 0.1 mg l21 spiking level, the RSD is 6% (n = 7) and the recoveries are better than 83% for all analytes.The long-term reproducibility typically has RSD values of 5% (n = 7). The method was successfully tested on water samples from various origins, and the results obtained with the present on-line SPE–LC–UV procedure were found to compare well with those obtained with procedures involving SPE combined off-line with GC–MS or flow injection MS–MS. Keywords: Phenoxy acid herbicides; solid-phase extraction–liquid chromatography–UV detection; high-pH clean-up; aqueous samples In our laboratory, the determination of phenoxy acid herbicides (for names and structures, see Table 1) in surface water is performed according to a previously described method.1,2 The method essentially consists in on-line analyte enrichment from a small sample volume at pH 3, pre-column clean-up and an isocratic ion-pair LC separation on a polymeric column.However, over the years some of the initial conditions were changed because of maintenance problems. As an example, the high buffer salt concentration and the high pH of the washing solvent caused rapid deterioration of the LC pumps and valves, and were therefore modified. In addition, in order to simplify the procedure, heart-cutting of the pre-column desorption solvent2 was omitted.All modifications were evaluated by testing spiked tap water samples and comparing the results obtained with the initial conditions. In general, the modified procedures could be used successfully for analyses of tap water and water from the main Dutch rivers, the Rhine and Meuse.However, surprisingly, the analysis of spiked surface water samples showed large differences in analyte recoveries depending on the origin of the sample. Problems were experienced especially with samples from rural areas and were expected to be due to the presence of relatively large amounts of humic material. In order to improve the reliability of the procedure and to enhance the recovery of the analytes in, especially, the ruralarea samples, the total analytical procedure has now been further modified and optimized.As a check, the results of the final solid-phase extraction (SPE)–LC–UV procedure were compared with those obtained by means of flow injection analysis (FIA)– MS–MS and GC–MS. Experimental Reagents All chemicals and solvents were of analytical-reagent grade. HPLC-grade methanol, acetonitrile, sodium hydroxide, ammo- Table 1 Structures of phenoxy acid herbicides and bentazone Analyte Analyte structure Abbreviation Peak No.pKa 2 4 5 R CAS No. MCPA 1 3.1 CH3 Cl H CH2COOH 94-74-6 MCPP 2 3.6 Cl Cl H CH2COOH 94-75-7 2,4-D 3 2.8 CH3 Cl H CH(CH3)COOH 7085-19-0 2,4-DP 4 2.7 Cl Cl H CH(CH3)COOH 120-36-5 Bentazone 5 3.2 25057-89-0 MCPB 6 4.6 Cl Cl Cl CH2COOH 93-76-5 2,4-DB 7 4.4 CH3 Cl H (CH2)3COH 94-81-5 2,4,5-T 8 2.8 Cl Cl H (CH2)3COH 94-82-6 2,4,5-TP 9 3.6 Cl Cl Cl CH(CH3)COOH 93-72-1 Analyst, September 1997, Vol. 122 (889–893) 889nium acetate, perchloric acid and water were obtained from J.T. Baker (Deventer, The Netherlands) and tetrabutylammonium hydrogensulfate from Fluka (Buchs, Switzerland). Chlorophenoxycarboxylic acids were obtained from Riedel-de Ha�en (Hannover, Germany) and Promochem (Wesel, Germany) and bentazone from Promochem. Equipment A Waters (Millipore, Bedford, MA, USA) Model 600E pump was used to deliver the eluent. A fully automated Prospekt (Spark, Emmen, The Netherlands) cartridge exchange system with two additional valves was used to control the flow scheme during analysis. The solvent delivery unit (SDU) of the Prospekt was provided with two additional six-port solvent selection valves for the wetting and conditioning of the pre-columns, the washing solvents and the aqueous samples.A Model 785A programmable absorbance detector (Applied Biosystems, H. I. Ambacht, The Netherlands) was operated at 232 nm. The column thermostat was a Mistral from Spark. Chromatograms were recorded and integrated by a Millipore data station using Baseline 810 or Millenium 2020 software.For MS detection, a Finnigan TSQ-700 mass spectrometer (Finnigan MAT, San Jose, CA, USA) equipped with a Finnigan thermospray interface (TSP-2) was used; the experimental conditions were as in previous work.3 Procedures Stock standard solutions of the chlorophenoxy acids and bentazone were prepared by weighing approximately 5 mg of each component and dissolution in 50 ml of methanol; they were stored at 220 °C.These solutions were diluted with ultrapure water and acidified to pH 2.8 with 0.1 mol l21 HClO4, to obtain individual and mixed working standard solutions which were stored at 4 °C. Spiked sample solutions were prepared by diluting the stock standard solutions with tap or surface water and acidifying them to pH 2.8 with 0.1 mol l21 HClO4. Pre-columns (10 3 2 mm id), pre-packed with 15–25 mm PLRP-S styrene–divinylbenzene copolymer (Polymer Laboratories, Church Stretton, UK), were wetted with 2 ml of methanol (2 ml min21) and activated with 2 ml of 0.001 mol l21 HClO4 (1 ml min21) prior to use.Preconcentration of the samples (60 ml) was performed at 2 ml min21. Details of the clean-up procedure are described below. LC separations were carried out at 50 °C on a 2503 4.6 d column pre-packed with 5 mm, 100 Å PLRP-S using an acetonitrile–water gradient (for details, see Table 2 and later text). Results and Discussion The trace-level determination of phenoxy acids in surface water with an isocratic LC separation often creates problems because most analyte peaks are eluted on a sharply decreasing baseline caused by a high humic acid hump in the early part of the chromatogram.This is demonstrated in Fig. 1, where the chromatogram of a standard solution is compared with that of a surface water sample. It will be obvious that the presence of the hump will cause difficulties during peak integration. Moreover, changing the sample volume applied to the pre-column causes a shift in the retention time of the MCPB and 2,4-DB (peaks 7 and 8), which are co-eluted in the system.This is demonstrated in Fig. 2, in which three tap water samples acidified to pH 3, of 30, 60 and 150 ml were applied to the pre-column and subsequently subjected to separation by LC. One should also note that, even with tap water samples, the baseline is not straight because of the presence of the dissolved organic carbon (DOC) which is not removed during tap water preparation from surface water.As is to be expected, the peak areas of all compounds of interest increased with increasing sample volume, i.e., they were twoand five-fold higher for the 60 and 150 ml samples, respectively, compared with the 30 ml sample. However, with the 150 ml sample, the peaks of MCPB and 2,4-DB shifted to longer retention times and were now co-eluted with 2,4,5-TP (peak 9).One explanation of the shift is that it is due to the amount of acid applied to the pre-column. Obviously, a larger volume of highpH solution will be necessary to ionize the compounds among which MCPB and 2,4-DB have the highest pKa values. Desorption with the ion-pair-containing acetonitrile–water (30 + 70 v/v) eluent at pH 8.8 does not occur sufficiently rapidly for these two compounds, although the pH used ensures complete dissociation of all compounds with pKa 2.7–4.6.This Table 2 Gradient elution procedure D Time/ Flow rate/ min ml min21 A* (%) B* (%) 0 1.0 70 30 5 70 30 25 30 70 26 70 30 30 70 30 60 0.05 70 30 * Solvent A: acetonitrile–water (20 + 80 v/v) with 0.0125 mol l21 borate buffer in aqueous phase and 0.003 mol l21 TBA. Solvent B: acetonitrile–water (40 + 60 v/v) with 0.0125 mol l21 borate buffer in aqueous phase and 0.003 mol l21 TBA. Fig. 1 SPE–LC–UV of acid herbicides after preconcentration of (A) surface water and (B) a standard solution.Conditions: 60 ml sample, spike 1.0 mg l21; washing solvent, acetonitrile–water (4 + 96 v/v), pH 3, 1 ml; system, 10 3 2 mm id pre-column packed with 20 mm PLRP-S, 250 3 4.6 mm id analytical column packed with 5 mm PLRP-S; eluent, acetonitrile– water (30 + 70 v/v), containing 0.0125 mol l21 TB and 0.003 mol l21 borate buffer, pH 8.8, flow-rate 1.0 ml min21; UV detection, 232 nm. Peaks: 1, MCPA; 2, 2,4-D; 3, MCPP; 4, 2,4-DP; 5, bentazone; 6, 2,4,5-T; 7/8, MCPB/ 2,4-DB; 9, 2,4,5-TP.Fig. 2 SPE–LC–UV of acid herbicides after preconcentration of (A) 30, (B) 60 and (C) 150 ml of tap water spiked with 0.25 mg l21 of each component. For conditions, see Fig. 1. At asterisk: MCPB/2,4-DB. 890 Analyst, September 1997, Vol. 122suggests that the buffer strength may be an important parameter. In order to improve the LC separation, several conditions of the procedure were re-examined. Analytical LC Separation In Table 3 the conditions from refs. 1 and 2 (columns A and B, respectively) and the initial conditions used in this work (column C) are summarized. As discussed in the Introduction, the conditions of column C are the result of earlier improvements. First, the influence of the pH used during sorption, i.e., the pH of the sample, and the pH used during desorption, i.e., the pH of the eluent, were studied in the ranges 2.0–3.5 and 7.5–9.5, respectively. The results showed that a low sample pH (2.0) gave less reproducible results than higher pHs: the RSD values of the retention times of the test analytes were about 5% and 0.1% at pH 2.0 and 2.8, respectively (n = 7).On the other hand, varying the eluent pH over the range indicated had no significant effect. Next, gradient elution was introduced and the ion-pair reagent (TBA) and the borate buffer salt concentrations were optimized in order to obtain a low and non-fluctuating baseline during elution of the analytes of interest. The TBA concentration was studied in the range 0.001–0.006 mol l21 and the borate concentration in the range 0.005–0.025 mol l21.The proper eluent composition for separation turned out to be a linear gradient for 20 min from 26 + 74 to 34 + 66 v/v acetonitrile– water with constant concentrations of 0.003 mol l21 TBA and 0.0125 mol l21 borate buffer of pH 8.8 (Table 3, column D). With these conditions, chromatograms of tap, surface and rural water samples all showed the desired baseline during analyte elution, even though there still was a baseline offset due to matrix constituents with surface and rural water.Influence of Matrix Constituents The trace-level determination of phenoxy acids generally requires extensive clean-up to remove humic constituents. These interfering compounds with their many acidic and phenolic substituents are difficult to separate from the similar phenoxy acids. In actual practice, pre-treatment techniques such as ultrafiltration,2 on-line dialysis with (hollow-fibre) membranes4 or a dual pre-column procedure (PLRP-S and anion exchanger)5 are not sufficient for the analysis of surface water with concentrations down to 0.1 mg l21.With tap and groundwater, detection limits of @0.1 mg l21 are more easily achieved with off-line SPE on C18-bonded silica at neutral pH using an ion-pairing mechanism.6 In another study on the preconcentration of a broad range of acidic, basic and neutral compounds from surface water adjusted to pH 13,7 the selectivity did improve considerably, but only 2–5 ml samples could be handled without breakthrough.This resulted in poor detection limits of 0.5–1.3 mg l21. As an alternative, in a previous paper2 pre-column clean-up with up to 7 ml of 0.1 mol l21 aqueous sodium hydroxide was described. At that time, the LC eluent contained a commercial ion-pair reagent, low-UV PIC A (Waters), with a high buffer salt concentration that caused rapid deterioration of the pump heads.The target analytes were, admittedly, quantitatively recovered from the pre-column, probably owing to the additional ‘mass effect’ of the buffer salt; however, the humic acid hump in the chromatogram was still pronounced. In addition, owing to the high maintenance costs, the robustness of the procedure was not satisfactory. As regards alternatives, the use of a clean-up procedure with a solution containing 1–5% of acetonitrile did not cause breakthrough of the analytes.However, the recoveries of the analytes (spiked at 0.25 and 1.0 mg l21) from rural samples were not constant and often much lower than those from tap water. Typical examples are presented in Table 4. In several instances, the analyte recoveries were below 60%; however, duplicate runs gave the same result, indicating that the nature of the sample and not the analytical procedure was the main cause of the problem. Several of the above samples were also subjected to analysis by FIA–MS–MS, a rapid screening procedure which was described in detail in a previous paper.3 To illustrate the problems then encountered, in one sample 3 mg l21 of bentazone was found with FIA–MS–MS, whereas there was no peak at the retention time of bentazone in the LC–UV trace.This, and other, false-negative results indicate that the phenoxy acids are not, or incompletely, sorbed from these samples on the pre-column or, more likely, that they are sorbed on the pre-column but not desorbed by the LC eluent.With the latter explanation, the surface of the polymeric sorbent is considered to become coated by humic substances during the preconcentration process, which now acts as a (modified) stationary phase extracting the analytes of interest. Since bonding to this humic phase is stronger than bonding to the polymeric phase, analyte recoveries will be low(er). Moreover, in such a situation conventional clean-up procedures will not be successful because, with the release of the matrix constituents, the analytes will also be released and recoveries will be poor.Therefore, a washing step with water containing a few per cent of acetonitrile will only be efficient if relatively clean samples such as tap and ground Table 3 Schematic representation of methods used to determine phenoxy acids and bentazone Procedure Parameter A B C D LC Eluent Acetonitrile–water (30 +70) Acetonitrile–water (30 +70) Acetonitrile–water (30 +70) A: acetonitrile–water (20 + 80) B: acetonitrile–water (40 + 60) Elution Isocratic Isocratic Isocratic Gradient (see Table 2) Buffer No Low-UV PIC A reagent 0.0125 mol l21 borate 0.0125 mol l21 borate pH 11 8.3 8.8 8.8 Ion pair 0.01 mol l21 TBA 0.005 mol l21 TBA 0.003 mol l21 TBA 0.003 mol l21 TBA Sample pH 3 3 2.8 2.8 Heart-cut No Yes No No Clean-up 100 ml acetonitrile–water 1 ml 0.1 mol l21 sodium 1 ml acetonitrile–water 1 ml sodium hydroxide (30 + 70), pH 3 hydroxide solution, (4 + 96), pH 3 solution, pH 11 pH 12.5 Analyst, September 1997, Vol. 122 891water are analysed and matrix constituents are not dominant. Rural water samples, however, need more drastic clean-up procedures such as high-pH washing. High-pH Clean-up The pre-column washing with up to 7 ml of 0.1 mol l21 sodium hydroxide solution referred to above was followed by heartcutting of the pre-column eluent; this caused a considerable decrease in the matrix interferences. The explanation then provided was that the humic substances are highly polar molecules which were easily eluted from the pre-column without dragging along the analytes of interest.However, this explanation needs further extension. Application of a high-pH solvent to the pre-column will result in a charged humic phase. This phase can still interact with the, also charged, phenoxy acids and the polymeric sorbent, the interaction being due to charge-transfer complexation between the aromatic styrene– divinylbenzene copolymer with its p-electron pairs and the electrophilic humic and phenoxy acids.8 The binding energies of these interactions on polymeric pre-columns are higher than hydrophobic interactions,9 which are the main retention forces if C18-bonded silica is applied for the analysis of water for these compounds.This may also explain why such large volumes of high-pH washing solvents (7–10 ml) can be used without analyte breakthrough. The explanation also supports the results of Aiken et al.,10 who reported that the order of breakthrough of fulvic acids on polymeric resins at pH 13 is the same as that at pH 2 (note that C18-type phases show a different retention mechanism with humic-bound compounds passing through and non-bound compounds being retained11,12).Desorption of the, now very polar, pre-column with the aqueous pH 8.8 eluent resulted in the quantitative and instantaneous release of all phenoxy compounds and the remaining humic material.In summary, application of 1 ml of sodium hydroxide solution (pH 11) for clean-up and the linear LC gradient of procedure D for separation gave excellent baselines and good recoveries for all types of water samples studied. Fig. 3 shows a typical example of a tap and a rural-area water sample. Table 5 shows that the recoveries from the rural samples determined with gradient elution (procedure D) are much higher than those with the isocratic procedure C.The recoveries are essentially quantitative (80–110%) for all but one analyte (bentazone, 70–80%). In other words, the desired rapid release of the analytes is indeed effected by the introduction of the high-pH washing step. Analytical Data The limits of detection, set at three times the standard deviation (n = 7), which were determined using tap water spiked at the 0.01–0.02 mg l21 level, were 5–20 ng l21 for all nine analytes (60 ml samples). The repeatability and recovery of the procedure were determined with tap water spiked at the 0.1 mg l21 level. The recoveries invariably were > 83% with RSD values of 3–7% (n = 7).The reproducibility of the procedure was tested by analysing one freshly prepared sample each week (spiking level 0.1 mg l21), for seven weeks; the RSD values were 2–10%. The method was also tested by spiking several samples from a rural area at three levels, 0.25, 0.5 and 1.0 mg l21. The samples were analysed in our laboratory with the optimized SPE–LC– UV procedure D and with FIA–MS–MS.The sample set was also analysed by a contract agency which used a GC–MS procedure. Table 6 shows that the present procedure works well at all levels with mean recoveries > 80%. At the 0.25 mg l21 level, the recovery for some of the compounds is lower; however, these recoveries are comparable with the results of the other procedures. The precision of the method (reference material from the contract laboratory at 0.2 mg l21) is excellent with recoveries of 97–109% and RSDs of 4–6% (n = 5).In general, the SPE–LC–UV results compare well with those from the FIA–MS–MS procedure (in which an internal standard is used) at all levels tested, whereas the GC–MS results were Table 4 Analyte recoveries (%) from rural-area samples spiked at the 0.25 and 1.0 mg l21 levels and determined with procedure C. Values corrected for blank values Sample 1 Sample 2 Sample 3 Analyte +0.25 +1.0 +0.25 +1.0 +0.25 +1.0 MCPA 68 74 60 53 62 64 MCPP 52 73 56 64 57 76 2,4-D 60 67 46 62 53 62 2,4-DP 56 75 47 64 52 75 MCPB/2,4-DB 76 85 58 82 61 80 2,4,5-T 42 107 64 63 102 62 2,4,5-TP 32 65 40 77 48 66 Bentazone 76 87 60 86 60 82 Fig. 3 SPE–LC–UV of acid herbicides after preconcentration of 60 ml of rural (top) and tap (bottom) water spiked with 0.25 mg l21 of each component, using procedure D. System: 10 3 2 mm id pre-column packed with 20 mm PLRP-S, 250 3 4.6 mm id analytical column packed with 5 mm PLRP-S; UV detection, 232 nm.For gradient elution conditions, see Table 2. Peaks: 1, MCPA; 2, 2,4-D; 3, MCPP; 4, 2,4-DP; 5, bentazone; 6, 2,4,5-T; 7/8, MCPB/2,4-DB; 9, 2,4,5-TP. Table 5 Analyte recoveries (%) from rural-area samples spiked at the 1.0 mg l21 level and determined with procedures C and D. Values corrected for blank values Sample No. Analyte Procedure 1 2 3 4 5 MCPA C 50 66 72 64 76 D 85 79 83 91 78 MCPP C 85 91 84 88 73 D 106 83 103 98 95 2,4-D C 62 59 54 61 54 D 91 78 86 95 83 2,4-DP C 79 85 80 80 76 D 105 80 112 101 101 MCPB/2,4-DB C 89 87 64 87 86 D 95 76 100 95 76 2,4,5-T C 86 75 67 85 –* D 87 92 75 87 89 2,4,5-TP C 88 90 –* 88 81 D 110 81 –* 108 120 Bentazone C 58 65 55 65 51 D 70 84 77 71 77 * – Not determined. 892 Analyst, September 1997, Vol. 122sometimes disappointing, probably owing to low conversion yields during derivatization. Routine Procedure D has been in use for about 2 years and is one of the ISO 9001/9002 methods in our laboratory.About 100 samples are analysed each year and the robustness of the procedure turned out to be excellent. Fig. 4 shows an example of an SPE–LC–UV trace from a rural water sample. In this sample, four compounds are detected at concentrations of 0.05–0.1 mg l21. From Fig. 4, it is clear that at this level these compounds are readily detected, owing to the improved clean-up of the pre-column. Conclusions Two previously reported procedures for the trace-level determination of phenoxy acid herbicides and bentazone were evaluated and the conditions changed in order to improve the practicability for a variety of water samples.The introduction of an LC gradient instead of isocratic elution caused a less fluctuating baseline, which resulted in improved detection limits. Lowering the buffer salt and ion-pair concentrations helped to prolong the lifetime of the LC system. Washing the pre-column with water containing 4% acetonitrile turned out to be effective and reliable only if relatively clean samples were analysed.Washing the pre-column with a high-pH solution, which induces the formation of chargetransfer complexes between the humic phase and the polymeric sorbent, turned out to be a much more powerful option. The compounds of interest and the remaining humic constituents are instantaneously released from this phase on introducing the LC eluent. The analyte recovery is now satisfactory ( > 70%) even for samples with a high humic content.The repeatability and reproducibility of the total analytical procedure are good and detection limits for 60 ml samples are as low as 5–20 ng l21. The results obtained for spiked rural-area samples with the present SPE–LC–UV procedure turned out to be essentially the same as those with a (much more expensive) FIA–MS–MS procedure and at least as good as those with a conventional GC–MS procedure, which has the disadvantage of requiring derivatization. References 1 Geerdink, R. B., van Balkom, C. A. A., and Brouwer, H.-J., J. Chromatogr., 1989, 481, 275. 2 Geerdink, R. B., Graumans, A. M. B. C., and Viveen, J., J. Chromatogr., 1991, 547, 478. 3 Geerdink, R. B., Kienhuis, P. G. M., and Brinkman, U. A. Th., Chromatographia, 1994, 39, 311. 4 van de Merbel, N. C. F., Lagerwerf, M., Lingeman, H., and Brinkman, U. A. Th., Int. J. Environ. Anal. Chem., 1994, 54, 105. 5 Vera-Avila, L. E., Padilla, P. C., Hernandez, M. G., and Meraz, J. L. L., J. Chromatogr. A, 1996, 731, 115. 6 Ballinova, A., J. Chromatogr. A, 1996, 728, 319. 7 Pijlman, L., Technical Report, RIZA, Lelystad, 1991. 8 LeCloire, P., Lelacheur, R. M., Johnson, J. D., and Christman, R. F., Water Res., 1990, 24, 1151. 9 McDowell, R. D., LC–GC Int., 1994, 7, 638. 10 Aiken, G. R., Thurman, E. M., Malcolm, R. L., and Walton, H. F., Anal. Chem., 1979, 51, 1799. 11 Gremm, T., Huber, S., and Frimmel, F. H., Vom Wasser, 1993, 80, 109. 12 Landrum, P. F., Nihart, S. R., Eadle, B. J., and Gardner, W. S., Environ. Sci. Technol., 1984, 18, 187. Paper 7/02338C Received April 7, 1997 Accepted May 30, 1997 Table 6 Recovery and precision, expressed as RSD, at 0.2 mg l21 and mean results (mg l21) for phenoxy acids and bentazone spiked at 0.25, 0.5 and 1.0 mg l21 in four rural-area samples and determined with FIA–MS–MS (A), SPE–LC–UV (B) and GC–MS (C). Values corrected for blank values Spike 1.0 mg l21 Spike 0.5 mg l21 Spike 0.25 mg l21 Recovery RSD Analyte (%) (%) A B C A B C A B C MCPA 104 5.9 0.79 0.94 0.78 0.39 0.42 0.39 0.19 0.18 0.21 MCPP 101 3.3 0.79 0.73 0.82 0.43 0.42 0.45 0.21 0.20 0.22 2,4-D 102 6.2 0.98 0.90 0.58 0.50 0.49 0.36 0.23 0.20 0.18 MCPB –* – 0.63 – 0.42 0.34 – 0.24 0.17 – 0.14 2,4-DP 97 4.3 0.83 1.03 0.71 0.45 0.50 0.36 0.30 0.26 0.21 2,4-DB 109 4.9 0.78 0.94 0.32 0.46 0.49 0.18 0.14† 0.20 0.15‡ 2,4,5-T 99 4.1 1.05 0.84 0.37 0.50 0.44 0.23 0.23 0.19 0.16 2,4,5-TP 104 5.8 0.79 0.87 0.52 0.42 0.50 0.31 0.32 0.28‡ 0.19 Bentazone 109 4.0 0.72 0.66 0.24 0.41 0.43 0.14 0.20 0.17 0.13‡ * Not determined. † Not detected in two samples; average for two samples. ‡ Not detected in one sample; average for two samples. Fig. 4 SPE–LC–UV of 60 ml of rural water using procedure D. For conditions, see Fig. 3. Analyst, September 1997, Vol. 122 893
ISSN:0003-2654
DOI:10.1039/a702338c
出版商:RSC
年代:1997
数据来源: RSC
|
6. |
Extraction of Hexaconazole From Weathered Soils: a ComparisonBetween Soxhlet Extraction, Microwave-assisted Extraction, SupercriticalFluid Extraction and Accelerated Solvent Extraction |
|
Analyst,
Volume 122,
Issue 9,
1997,
Page 895-898
S. P. Frost,
Preview
|
|
摘要:
Extraction of Hexaconazole From Weathered Soils: a Comparison Between Soxhlet Extraction, Microwave-assisted Extraction, Supercritical Fluid Extraction and Accelerated Solvent Extraction S. P. Frosta, J. R. Dean*a, K. P. Evansb, K. Harradinec, C. Caryc and M. H. I. Comberd a Department of Chemical and Life Sciences, University of Northumbria at Newcastle, Ellison Building, Newcastle upon Tyne, UK NE1 8ST b Zeneca Specialties, P.O. Box 42, Hexagon House, Blackley, Manchester, UK M9 8ZS c Zeneca Agrochemicals, Jealott’s Hill Research Station, Bracknell, Berkshire, UK RG12 6EY d Brixham Environmental Laboratory, Zeneca Ltd., Freshwater Quarry, Brixham, Devon, UK TQ5 8BA Extraction of hexaconazole residues which had been weathered over periods of time varying between 0 and 52 weeks was undertaken from two characterized soils.The various extraction techniques considered were Soxhlet extraction, microwave-assisted extraction, supercritical fluid extraction and accelerated solvent extraction.The results indicated that the physico-chemical properties of the soil are an important factor in the extraction of weathered residues of hexaconazole from soil. Overall the results obtained by accelerated solvent extraction were comparable to those obtained by Soxhlet extraction. Some matrix dependence of supercritical fluid extraction and microwave-assisted extraction was noted for the aged (52 week) sampled sandy loam soil. This effect was not evident for any of the extraction techniques using the sandy clay soil.The best precision was obtained for the automated accelerated solvent extraction system. Keywords: Supercritical fluid extraction; microwave-assisted extraction; accelerated solvent extraction; hexaconazole; weathered soils The need to change the nature of the solvent, the amount of solvent used and the time required to undertake an extraction from a matrix, whether it is soil, textiles or organic matter, has been the driving force behind the development of various techniques in recent years to challenge the long standing and proven method of Soxhlet extraction.1–4 Substantial progress has been made towards developing improved techniques such as supercritical fluid extraction (SFE), microwave-assisted extraction (MAE) and accelerated solvent extraction (ASE).This paper reports the results found in a comparative study between these techniques for the extraction of the broad spectrum systemic triazole fungicide hexaconazole from weathered soil samples.Supercritical fluid extraction has been the focus of much research for the preparation of samples from solid matrices. It is a popular method owing to the nature of the supercritical fluid CO2, which is commonly used. Its critical parameters are low and it is inexpensive compared with HPLC-grade solvents, chemically inert and non-toxic. Many workers have used it for the extraction of non-polar and moderately polar environmental contaminants such as polycyclic aromatic hydrocarbons (PAHs),5,6 although it has generally been found that the addition of a polar modifier, commonly methanol, is required for the extraction of more polar compounds and also to improve the recoveries of apolar compounds.This work has been extended to take into account the importance of increased temperature in the extraction process. Yang et al.7 found that increased temperatures, much higher than those normally encountered in SFE experiments, gave improved recoveries of PAHs from environmental samples.The use of microwave energy to extract organic compounds from contaminated soil was first reported in 1986–87.8,9 Microwave energy is non-ionizing radiation that causes molecular motion by migration of ions and rotation of dipoles, but does not induce changes in molecular structure. Rapid heating within an organic sample with a permanent dipole moment is brought about by alignment of the molecules, followed by their rapid return to disorder.Therefore, for the extraction process to be efficient, the sample obviously has to be contained in a solvent with a permanent dipole moment. In experiments carried out by several workers,10–12 mainly on the extraction of PAHs, a solvent mixture of hexane–acetone (1 + 1) was used. Like SFE, MAE has created interest for the recovery of organic pollutants from environmental matrices because of the necessity to reduce organic solvent consumption and the speed of extraction.Typically, 30–40 ml of the organic solvent are used with extraction times of 5–20 min per sample. Since the analyte is typically contained within a closed vessel during extraction, the temperature of the solvent can be raised above its boiling point whilst under pressure. In addition, multiple sample extractions can be undertaken simultaneously under the same conditions. More recently, ASE has emerged and is gaining in popularity. 13,14 This technique uses conventional liquid organic solvents at elevated temperatures (50–200 °C) and pressures [1500–2000 psi (1 psi = 6894.76 Pa)] to extract solid samples quickly and with smaller solvent volumes, typically 10–20 ml.The sample is enclosed in a high-pressure stainless-steel cell and, after an initial heating process, the sample is allowed to interact statically with the pressurized solvent for a predetermined time period, after which it is purged from the cell by compressed nitrogen into a collection vial.The whole process takes < 15 min per sample. As the system is fully automated, samples can be run sequentially. Experimental Weathered Canadian soil samples were supplied by Zeneca Agrochemicals (Bracknell, Berkshire, UK). These consisted of soil samples which had been collected at various times after hexaconazole application. Two soil types were evaluated. Each soil sample was air dried for 48 h and sieved through a 2–3.5 mm sieve prior to storage at 218 °C.Each soil was collected at various times after hexaconazole application and stored until Analyst, September 1997, Vol. 122 (895–898) 895required. Soil A was collected over 0, 1 and 52 weeks, corresponding to soils A1, A2 and A3, respectively, and soil B was collected over 1 and 52 weeks, corresponding to soils B2 and B3, respectively. Extraction Procedures The levels of hexaconazole in each soil were quantified initially by Soxhlet extraction.Amounts of 40 g of soil were weighed into a 25 3 100 mm extraction thimble, 80 ml of acetonitrile– water 1 + 1 were placed in a round-bottomed flask and the extraction was carried out for 6 h. Each extract was diluted with 2 ml of glass distilled water and partitioned into 5 ml of dichloromethane, reduced to dryness and re-diluted in 1 ml of dichloromethane. Samples were transfered on to dichloromethane (DCM) pre-wetted silica ISOLUTE columns (International Sorbent Technology, Hengoed, Mid-Glamorgan, UK), washed with DCM–MeOH (95 + 5) (2 ml), and eluted with DCM–MeOH (95 + 5) (2.5 ml).The eluate evaporated to dryness and re-diluted in acetone–hexane (20 + 80) for subsequent analysis. Supercritical fluid extraction was carried out on a Jasco (Tokyo, Japan) SFE instrument which possesses a second pump for modifier addition and a back-pressure regulator (BPR), which allows independent control of the extraction pressure.SFC-grade CO2 purchased from Air Products (Sunderland, UK) was combined with HPLC-grade methanol (HiPerSolv, BDH Laboratory Supplies, Poole, Dorset, UK). The optimum conditions for extraction were as follows: pressure, 250 kg cm22 (1 kg cm22 = 98 066 Pa); temperature, 55 °C; proportion of methanol modifier, 20%; and extraction time, 20 min. These conditions were determined using laboratory slurry spiked soils.15 However, as slurry spiked samples may not relate to weathered soil samples, the effects of extraction time were evaluated further.The flow rate was maintained at 2 ml min21 and the pump head was cooled by means of a re-circulating coolant. A sample of approximately 4 g was accurately weighed into a stainless-steel extraction cell and placed inside the oven, where it was allowed to equilibrate to the set temperature for 5 min. Next the cell was filled with the supercritical CO2 and methanol modifier via the pumps and allowed to interact statically with the analyte–matrix for a further 5 min with the pumps switched off. After this, the pumps were restarted and the dynamic period was initiated for the determined period.The extract was collected by inserting the end of the BPR through a PTFE-lined septum into a collection vial containing a small volume of methanol (5 ml). Depressurized samples were vented through a cyano-bonded solid-phase extraction (SPE) cartridge (Varian, Walton-on-Thames, UK), pre-wetted with methanol.Upon completion of the extraction, the cartridge was backflushed with 2 3 2 ml of methanol, then 1 ml of 1 ppm internal standard (UK-46,253, Pfizer Central Research, Kent, UK) was added to each extract. Soil samples (A1–A3) were subjected to a pre-chromatographic clean-up using C18 (Phase Separations, Queensferry, Clwyd, UK) SPE cartridges, pre-wetted with methanol, to remove largely non-polar contaminants. The cartridge was eluted with 5 ml of methanol and filtered through 0.2 mm Acrodiscs (Phase Separations).Samples B2 and B3 were filtered using 0.2 mm Acrodiscs only. Each sample was heated to 45 °C in a water-bath and reduced in volume to 1 ml under a stream of compressed air. Microwave-assisted extraction was carried out using a Milestone 1200 Mega unit (Milestone, Rome, Italy). Soil samples (5 g) were accurately weighed into the Teflon liner of the extraction vessel and 30 ml of acetone (analytical-reagent grade, Rathburn, Walkerburn, UK) added.The vessels were sealed and extractions were carried out at 115 °C at 1000 W for 15 min. After extraction, the vessels were allowed to cool to 40 °C, typically taking 30 min, before opening. The contents were filtered through a size 4 sintered glass filter and 1 ml of internal standard (in acetone) was added. All soil samples were subjected to a pre-chromatographic clean-up using C18 (Phase Separations) cartridges, pre-wetted with methanol, to remove largely non-polar contaminants.The cartridge was eluted with 5 ml of methanol and then filtered through 0.2 mm Acrodiscs (Phase Separations). Finally, the filtered sample was reduced in volume to 1 ml. Accelerated solvent extraction was carried out using an ASE 200 system (Dionex, Camberley, Sussex, UK). Acetone (AnalaR) was used as the solvent to implement the extractions. Samples of 5 g of the soils were weighed into the stainless-steel extraction cells (11 ml capacity) and the dead volume was filled with pre-cleaned sand (Dionex).After temperature equilibration (5 min), extraction was carried out at 2000 psi at 100 °C for 10 min then 1 ml of 5 ppm internal standard was added and the Table 1 Extraction of hexaconazole from aged soil samples.* All values in mg kg21 No. of weeks Supercritical Microwave- Accelerated after application Soxhlet fluid assisted solvent Soil of hexaconazole extraction† extraction extraction extraction A1, sandy 0 0.14 Mean 0.074 Mean 0.080 Mean 0.124 loam soil RSD 7.4% RSD 8.1% RSD 3.8% n = 5 n = 4 n = 5 A2, sandy 1 0.12 Mean 0.087 Mean 0.104 Mean 0.127 loam soil RSD 5.4% RSD 7.8% RSD 4.4% n = 4 n = 4 n = 6 A3, sandy 52 0.07 Mean 0.034 Mean 0.035 Mean 0.093 loam soil RSD 7.2% RSD 23.0% RSD 1.8% n = 4 n = 4 n = 6 B2, sandy clay 1 0.14 Mean 0.119 Mean 0.134 Mean 0.136 soil RSD 10.1% RSD 14.7% RSD 1.9% n = 4 n = 4 n = 6 B3, sandy clay 52 0.08 Mean 0.072 Mean 0.073 Mean 0.070 soil RSD 8.4% RSD 4.6% RSD 3.4% n = 4 n = 4 n = 6 * All soils taken from a depth of 0–5 cm.Sandy loam soil: 5.7% organic matter; 81% sand; 8% silt; 11% clay and pH 7.4. Sandy clay soil: 1.5% organic matter; 53% sand; 25% silt; 22% clay and pH 8.3. † Typical RSD < 15%. 896 Analyst, September 1997, Vol. 122samples were reduced to 1 ml before being filtered through 0.2 mm Acrodiscs only. Chromatographic Analysis MAE and ASE samples were analysed using a Fisons (Crawley, Sussex, UK) Model 8000 gas chromatograph equipped with an electron capture detector. The following chromatographic conditions were found to be suitable for the analysis: a 1 ml sample was injected, on-column, on to a 60 m 3 0.25 mm id DB1 column with a 0.25 mm film thickness.The oven temperature was ramped from 70 to 280 °C at 35 °C min21 and held at 280 °C for 14 min, with a helium flow rate of 2.0 ml min21. Calibration curves were prepared with standards of 0.05, 0.1, 0.2, 0.5 and 1.0 ppm, giving correlation coefficients of 0.9966 (ASE) and 0.9985 (MAE).SFE samples were analysed with a Varian Star 3400 gas chromatograph equipped with an electron capture detector. Volumes of 1 ml were injected via a split–splitless injector on to a 60 m 3 0.25 mm id DB5 column with a 0.25 mm film thickness. The oven temperature was ramped from 80 to 270 °C at 40 °C min21 and held at 270 °C for 14 min. A calibration curve was prepared with standards as above, with a correlation coefficient of 0.9996.Results and Discussion Time studies were carried out for SFE extractions over 20, 40 and 60 min using soil A3. The mean recoveries (n = 2) were 0.045, 0.047 and 0.044 mg kg21 for 20, 40 and 60 min extraction, respectively. As similar recoveries were achieved over all time periods, the extraction time was maintained at 20 min for speed of analysis. The results from multiple extractions from ASE, SFE and MAE were compared with those obtained using Soxhlet extraction (Table 1).For the sandy clay soil with an organic matter content of 1.5% (B2 and B3) good recoveries were achieved for all three methods compared with the recoveries obtained using Soxhlet extraction. However, consideration of the RSD values reveals that ASE generally produced far superior results on both the 1 and 52 week samples. This is probably due to the use of a fully automated system where sample handling is kept to a minimum. However, the results obtained for the soil with a higher organic matter content (5.7%), the sandy loam soil (soil samples A1–A3), showed greater variation between recoveries.SFE gave recoveries approximately 50% of those achieved by Soxhlet extraction, although the RSD remained within acceptable levels (5 and 7%). Similar results were achieved with MAE, but the RSD for the 52 week sample was 23.0%, indicating the poor repeatability. ASE, however, gave recovery data in good agreement with the Soxhlet extraction data at all levels of weathering with RSD values ranging between 1 and 4%. In terms of the actual extract produced from each technique, ASE produced the cleanest chromatographic sample, with no interfering or co-eluting peaks.In contrast, MAE samples produced a far ‘dirtier’ extract, which required pre-chromatographic clean-up. The interference was a single large tailing peak (retention time tR = 7.0 min) which co-eluted close to the peak of interest (hexaconazole, tR = 8.5 min) and produced rapid degeneration of the column efficiency (see, for example, Fig. 1). SFE extracted samples required the pre-chromatographic clean-up of the high organic content soils only, to allow the analyte signal to be resolved, as many peaks were eluted from these samples. Conclusions The nature of the analyte–matrix interaction is dependent on the composition of the soil, although its ability to be recovered is dependent upon the extraction conditions and the efficiency of the system.MAE can handle several samples simultaneously and is rapid, although long cool-down periods are required and poor recoveries were achieved for high organic content soils. SFE gave results similar to MAE without the need for several sample handling stages, although actual extraction periods were longer with this method. ASE gave good recoveries for all sample types used in this study, with rapid extraction times, and was the only technique to give results comparable to those obtained using the conventional Soxhlet extraction technique.S.P.F. and J.R.D. gratefully acknowledge the financial support of Zeneca and the University of Northumbria at Newcastle. The authors also acknowledge Dionex for the loan of the ASE system. References 1 van der Velde, E. G., de Haan, W., and Liem, A. K. D., J. Chromatogr., 1992, 626, 135. 2 Lopez-Avila, V., Young, R., and Teplitsky, N., J. AOAC Int., 1996, 79, 142. Fig. 1 Gas chromatogram of hexaconazole after microwave-assisted extraction: (a) prior to and (b) after C18 SPE clean-up.Analyst, September 1997, Vol. 122 8973 David, M. D., and Seiber, J. N., Anal. Chem., 1996, 68, 3038. 4 Bowadt, S., Johansson, B., Wunerli, S., Zennegg, M., de Alencastro, L. F., and Grandjean, D., Anal. Chem., 1995, 67, 2424. 5 Barnabas, I. J., Dean, J. R., Hitchin, S. M., and Owen, S. P., Anal. Chim. Acta, 1994, 291, 261. 6 Barnabas, I. J., Dean, J. R., Hitchin, S. M., and Owen, S. P., J.Chromatogr. Sci., 1994, 32, 547. 7 Yang, Y., Gharaibeh, A., Hawthorne, S. B., and Miller, D. J., Anal. Chem., 1995, 67, 641. 8 Ganzler, K., Salgo, A., and Valko, K., J. Chromatogr., 1986, 371, 299. 9 Ganzler, K., and Salgo, A., Z. Lebensm.-Unters. Forsch., 1987, 184, 274. 10 Barnabas, I. J., Dean, J. R., Fowlis, I. A., and Owen, S. P., Analyst, 1995, 120, 1897. 11 Lopez-Avila, V., Benedicto, J., Charan, C., and Young, R., Environ. Sci. Technol., 1995, 29, 2709. 12 Lopez-Avila, V., Young, R., Benedicto, J., Ho, P., and Kim, R., Anal.Chem., 1995, 67, 2096. 13 Dean, J. R., Anal. Comm., 1996, 33, 191. 14 Richter, B. E., Jones, B. A., Ezzel, J. L., and Porter, N. L., Anal. Chem., 1996, 68, 1033. 15 Frost, S. P., Dean, J. R., Evans, K. P., Harradine, K., Cary, C., and Comber, M. H. I., in preparation. Paper 7/02688I Received April 21, 1997 Accepted May 27, 1997 898 Analyst, September 1997, Vol. 122 Extraction of Hexaconazole From Weathered Soils: a Comparison Between Soxhlet Extraction, Microwave-assisted Extraction, Supercritical Fluid Extraction and Accelerated Solvent Extraction S.P. Frosta, J. R. Dean*a, K. P. Evansb, K. Harradinec, C. Caryc and M. H. I. Comberd a Department of Chemical and Life Sciences, University of Northumbria at Newcastle, Ellison Building, Newcastle upon Tyne, UK NE1 8ST b Zeneca Specialties, P.O. Box 42, Hexagon House, Blackley, Manchester, UK M9 8ZS c Zeneca Agrochemicals, Jealott’s Hill Research Station, Bracknell, Berkshire, UK RG12 6EY d Brixham Environmental Laboratory, Zeneca Ltd., Freshwater Quarry, Brixham, Devon, UK TQ5 8BA Extraction of hexaconazole residues which had been weathered over periods of time varying between 0 and 52 weeks was undertaken from two characterized soils.The various extraction techniques considered were Soxhlet extraction, microwave-assisted extraction, supercritical fluid extraction and accelerated solvent extraction.The results indicated that the physico-chemical properties of the soil are an important factor in the extraction of weathered residues of hexaconazole from soil. Overall the results obtained by accelerated solvent extraction were comparable to those obtained by Soxhlet extraction. Some matrix dependence of supercritical fluid extraction and microwave-assisted extraction was noted for the aged (52 week) sampled sandy loam soil. This effect was not evident for any of the extraction techniques using the sandy clay soil.The best precision was obtained for the automated accelerated solvent extraction system. Keywords: Supercritical fluid extraction; microwave-assisted extraction; accelerated solvent extraction; hexaconazole; weathered soils The need to change the nature of the solvent, the amount of solvent used and the time required to undertake an extraction from a matrix, whether it is soil, textiles or organic matter, has been the driving force behind the development of various techniques in recent years to challenge the long standing and proven method of Soxhlet extraction.1–4 Substantial progress has been made towards developing improved techniques such as supercritical fluid extraction (SFE), microwave-assisted extraction (MAE) and accelerated solvent extraction (ASE).This paper reports the results found in a comparative study between these techniques for the extraction of the broad spectrum systemic triazole fungicide hexaconazole from weathered soil samples.Supercritical fluid extraction has been the focus of much research for the preparation of samples from solid matrices. It is a popular method owing to the nature of the supercritical fluid CO2, which is commonly used. Its critical parameters are low and it is inexpensive compared with HPLC-grade solvents, chemically inert and non-toxic. Many workers have used it for the extraction of non-polar and moderately polar environmental contaminants such as polycyclic aromatic hydrocarbons (PAHs),5,6 although it has generally been found that the addition of a polar modifier, commonly methanol, is required for the extraction of more polar compounds and also to improve the recoveries of apolar compounds.This work has been extended to take into account the importance of increased temperature in the extraction process. Yang et al.7 found that increased temperatures, much higher than those normally encountered in SFE experiments, gave improved recoveries of PAHs from environmental samples.The use of microwave energy to extract organic compounds from contaminated soil was first reported in 1986–87.8,9 Microwave energy is non-ionizing radiation that causes molecular motion by migration of ions and rotation of dipoles, but does not induce changes in molecular structure. Rapid heating within an organic sample with a permanent dipole moment is brought about by alignment of the molecules, followed by their rapid return to disorder. Therefore, for the extraction process to be efficient, the sample obviously has to be contained in a solvent with a permanent dipole moment.In experiments carried out by several workers,10–12 mainly on the extraction of PAHs, a solvent mixture of hexane–acetone (1 + 1) was used. Like SFE, MAE has created interest for the recovery of organic pollutants from environmental matrices because of the necessity to reduce organic solvent consumption and the speed of extraction.Typically, 30–40 ml of the organic solvent are used with extraction times of 5–20 min per sample. Since the analyte is typically contained within a closed vessel during extraction, the temperature of the solvent can be raised above its boiling point whilst under pressure. In addition, multiple sample extractions can be undertaken simultaneously under the same conditions. More recently, ASE has emerged and is gaining in popularity. 13,14 This technique uses conventional liquid organic solvents at elevated temperatures (50–200 °C) and pressures [1500–2000 psi (1 psi = 6894.76 Pa)] to extract solid samples quickly and with smaller solvent volumes, typically 10–20 ml.The sample is enclosed in a high-pressure stainless-steel cell and, after an initial heating process, the sample is allowed to interact statically with the pressurized solvent for a predetermined time period, after which it is purged from the cell by compressed nitrogen into a collection vial.The whole process takes < 15 min per sample. As the system is fully automated, samples can be run sequentially. Experimental Weathered Canadian soil samples were supplied by Zeneca Agrochemicals (Bracknell, Berkshire, UK). These consisted of soil samples which had been collected at various times after hexaconazole application. Two soil types were evaluated. Each soil sample was air dried for 48 h and sieved through a 2–3.5 mm sieve prior to storage at 218 °C.Each soil was collected at various times after hexaconazole application and stored until Analyst, September 1997, Vol. 122 (895–898) 895required. Soil A was collected over 0, 1 and 52 weeks, corresponding to soils A1, A2 and A3, respectively, and soil B was collected over 1 and 52 weeks, corresponding to soils B2 and B3, respectively. Extraction Procedures The levels of hexaconazole in each soil were quantified initially by Soxhlet extraction.Amounts of 40 g of soil were weighed into a 25 3 100 mm extraction thimble, 80 ml of acetonitrile– water 1 + 1 were placed in a round-bottomed flask and the extraction was carried out for 6 h. Each extract was diluted with 2 ml of glass distilled water and partitioned into 5 ml of dichloromethane, reduced to dryness and re-diluted in 1 ml of dichloromethane. Samples were transfered on to dichloromethane (DCM) pre-wetted silica ISOLUTE columns (International Sorbent Technology, Hengoed, Mid-Glamorgan, UK), washed with DCM–MeOH (95 + 5) (2 ml), and eluted with DCM–MeOH (95 + 5) (2.5 ml).The eluate evaporated to dryness and re-diluted in acetone–hexane (20 + 80) for subsequent analysis. Supercritical fluid extraction was carried out on a Jasco (Tokyo, Japan) SFE instrument which possesses a second pump for modifier addition and a back-pressure regulator (BPR), which allows independent control of the extraction pressure. SFC-grade CO2 purchased from Air Products (Sunderland, UK) was combined with HPLC-grade methanol (HiPerSolv, BDH Laboratory Supplies, Poole, Dorset, UK).The optimum conditions for extraction were as follows: pressure, 250 kg cm22 (1 kg cm22 = 98 066 Pa); temperature, 55 °C; proportion of methanol modifier, 20%; and extraction time, 20 min. These conditions were determined using laboratory slurry spiked soils.15 However, as slurry spiked samples may not relate to weathered soil samples, the effects of extraction time were evaluated further.The flow rate was maintained at 2 ml min21 and the pump head was cooled by means of a re-circulating coolant. A sample of approximately 4 g was accurately weighed into a stainless-steel extraction cell and placed inside the oven, where it was allowed to equilibrate to the set temperature for 5 min. Next the cell was filled with the supercritical CO2 and methanol modifier via the pumps and allowed to interact statically with the analyte–matrix for a further 5 min with the pumps switched off.After this, the pumps were restarted and the dynamic period was initiated for the determined period. The extract was collected by inserting the end of the BPR through a PTFE-lined septum into a collection vial containing a small volume of methanol (5 ml). Depressurized samples were vented through a cyano-bonded solid-phase extraction (SPE) cartridge (Varian, Walton-on-Thames, UK), pre-wetted with methanol. Upon completion of the extraction, the cartridge was backflushed with 2 3 2 ml of methanol, then 1 ml of 1 ppm internal standard (UK-46,253, Pfizer Central Research, Kent, UK) was added to each extract.Soil samples (A1–A3) were subjected to a pre-chromatographic clean-up using C18 (Phase Separations, Queensferry, Clwyd, UK) SPE cartridges, pre-wetted with methanol, to remove largely non-polar contaminants. The cartridge was eluted with 5 ml of methanol and filtered through 0.2 mm Acrodiscs (Phase Separations).Samples B2 and B3 were filtered using 0.2 mm Acrodiscs only. Each sample was heated to 45 °C in a water-bath and reduced in volume to 1 ml under a stream of compressed air. Microwave-assisted extraction was carried out using a Milestone 1200 Mega unit (Milestone, Rome, Italy). Soil samples (5 g) were accurately weighed into the Teflon liner of the extraction vessel and 30 ml of acetone (analytical-reagent grade, Rathburn, Walkerburn, UK) added. The vessels were sealed and extractions were carried out at 115 °C at 1000 W for 15 min.After extraction, the vessels were allowed to cool to 40 °C, typically taking 30 min, before opening. The contents were filtered through a size 4 sintered glass filter and 1 ml of internal standard (in acetone) was added. All soil samples were subjected to a pre-chromatographic clean-up using C18 (Phase Separations) cartridges, pre-wetted with methanol, to remove largely non-polar contaminants. The cartridge was eluted with 5 ml of methanol and then filtered through 0.2 mm Acrodiscs (Phase Separations). Finally, the filtered sample was reduced in volume to 1 ml.Accelerated solvent extraction was carried out using an ASE 200 system (Dionex, Camberley, Sussex, UK). Acetone (AnalaR) was used as the solvent to implement the extractions. Samples of 5 g of the soils were weighed into the stainless-steel extraction cells (11 ml capacity) and the dead volume was filled with pre-cleaned sand (Dionex). After temperature equilibration (5 min), extraction was carried out at 2000 psi at 100 °C for 10 min then 1 ml of 5 ppm internal standard was added and the Table 1 Extraction of hexaconazole from aged soil samples.* All values in mg kg21 No.of weeks Supercritical Microwave- Accelerated after application Soxhlet fluid assisted solvent Soil of hexaconazole extraction† extraction extraction extraction A1, sandy 0 0.14 Mean 0.074 Mean 0.080 Mean 0.124 loam soil RSD 7.4% RSD 8.1% RSD 3.8% n = 5 n = 4 n = 5 A2, sandy 1 0.12 Mean 0.087 Mean 0.104 Mean 0.127 loam soil RSD 5.4% RSD 7.8% RSD 4.4% n = 4 n = 4 n = 6 A3, sandy 52 0.07 Mean 0.034 Mean 0.035 Mean 0.093 loam soil RSD 7.2% RSD 23.0% RSD 1.8% n = 4 n = 4 n = 6 B2, sandy clay 1 0.14 Mean 0.119 Mean 0.134 Mean 0.136 soil RSD 10.1% RSD 14.7% RSD 1.9% n = 4 n = 4 n = 6 B3, sandy clay 52 0.08 Mean 0.072 Mean 0.073 Mean 0.070 soil RSD 8.4% RSD 4.6% RSD 3.4% n = 4 n = 4 n = 6 * All soils taken from a depth of 0–5 cm.Sandy loam soil: 5.7% organic matter; 81% sand; 8% silt; 11% clay and pH 7.4. Sandy clay soil: 1.5% organic matter; 53% sand; 25% silt; 22% clay and pH 8.3. † Typical RSD < 15%. 896 Analyst, September 1997, Vol. 122samples were reduced to 1 ml before being filtered through 0.2 mm Acrodiscs only. Chromatographic Analysis MAE and ASE samples were analysed using a Fisons (Crawley, Sussex, UK) Model 8000 gas chromatograph equipped with an electron capture detector.The following chromatographic conditions were found to be suitable for the analysis: a 1 ml sample was injected, on-column, on to a 60 m 3 0.25 mm id DB1 column with a 0.25 mm film thickness. The oven temperature was ramped from 70 to 280 °C at 35 °C min21 and held at 280 °C for 14 min, with a helium flow rate of 2.0 ml min21. Calibration curves were prepared with standards of 0.05, 0.1, 0.2, 0.5 and 1.0 ppm, giving correlation coefficients of 0.9966 (ASE) and 0.9985 (MAE).SFE samples were analysed with a Varian Star 3400 gas chromatograph equipped with an electron capture detector. Volumes of 1 ml were injected via a split–splitless injector on to a 60 m 3 0.25 mm id DB5 column with a 0.25 mm film thickness. The oven temperature was ramped from 80 to 270 °C at 40 °C min21 and held at 270 °C for 14 min. A calibration curve was prepared with standards as above, with a correlation coefficient of 0.9996. Results and Discussion Time studies were carried out for SFE extractions over 20, 40 and 60 min using soil A3.The mean recoveries (n = 2) were 0.045, 0.047 and 0.044 mg kg21 for 20, 40 and 60 min extraction, respectively. As similar recoveries were achieved over all time periods, the extraction time was maintained at 20 min for speed of analysis. The results from multiple extractions from ASE, SFE and MAE were compared with those obtained using Soxhlet extraction (Table 1). For the sandy clay soil with an organic matter content of 1.5% (B2 and B3) good recoveries were achieved for all three methods compared with the recoveries obtained using Soxhlet extraction. However, consideration of the RSD values reveals that ASE generally produced far superior results on both the 1 and 52 week samples.This is probably due to the use of a fully automated system where sample handling is kept to a minimum. However, the results obtained for the soil with a higher organic matter content (5.7%), the sandy loam soil (soil samples A1–A3), showed greater variation between recoveries.SFE gave recoveries approximately 50% of those achieved by Soxhlet extraction, although the RSD remained within acceptable levels (5 and 7%). Similar results were achieved with MAE, but the RSD for the 52 week sample was 23.0%, indicating the poor repeatability. ASE, however, gave recovery data in good agreement with the Soxhlet extraction data at all levels of weathering with RSD values ranging between 1 and 4%.In terms of the actual extract produced from each technique, ASE produced the cleanest chromatographic sample, with no interfering or co-eluting peaks. In contrast, MAE samples produced a far ‘dirtier’ extract, which required pre-chromatographic clean-up. The interference was a single large tailing peak (retention time tR = 7.0 min) which co-eluted close to the peak of interest (hexaconazole, tR = 8.5 min) and produced rapid degeneration of the column efficiency (see, for example, Fig. 1). SFE extracted samples required the pre-chromatographic clean-up of the high organic content soils only, to allow the analyte signal to be resolved, as many peaks were eluted from these samples. Conclusions The nature of the analyte–matrix interaction is dependent on the composition of the soil, although its ability to be recovered is dependent upon the extraction conditions and the efficiency of the system. MAE can handle several samples simultaneously and is rapid, although long cool-down periods are required and poor recoveries were achieved for high organic content soils.SFE gave results similar to MAE without the need for several sample handling stages, although actual extraction periods were longer with this method. ASE gave good recoveries for all sample types used in this study, with rapid extraction times, and was the only technique to give results comparable to those obtained using the conventional Soxhlet extraction technique. S.P.F. and J.R.D. gratefully acknowledge the financial support of Zeneca and the University of Northumbria at Newcastle. The authors also acknowledge Dionex for the loan of the ASE system. References 1 van der Velde, E. G., de Haan, W., and Liem, A. K. D., J. Chromatogr., 1992, 626, 135. 2 Lopez-Avila, V., Young, R., and Teplitsky, N., J. AOAC Int., 1996, 79, 142. Fig. 1 Gas chromatogram of hexaconazole after microwave-assisted extraction: (a) prior to and (b) after C18 SPE clean-up. Analyst, September 1997, Vol. 122 8973 David, M. D., and Seiber, J. N., Anal. Chem., 1996, 68, 3038. 4 Bowadt, S., Johansson, B., Wunerli, S., Zennegg, M., de Alencastro, L. F., and Grandjean, D., Anal. Chem., 1995, 67, 2424. 5 Barnabas, I. J., Dean, J. R., Hitchin, S. M., and Owen, S. P., Anal. Chim. Acta, 1994, 291, 261. 6 Barnabas, I. J., Dean, J. R., Hitchin, S. M., and Owen, S. P., J. Chromatogr. Sci., 1994, 32, 547. 7 Yang, Y., Gharaibeh, A., Hawthorne, S. B., and Miller, D. J., Anal. Chem., 1995, 67, 641. 8 Ganzler, K., Salgo, A., and Valko, K., J. Chromatogr., 1986, 371, 299. 9 Ganzler, K., and Salgo, A., Z. Lebensm.-Unters. Forsch., 1987, 184, 274. 10 Barnabas, I. J., Dean, J. R., Fowlis, I. A., and Owen, S. P., Analyst, 1995, 120, 1897. 11 Lopez-Avila, V., Benedicto, J., Charan, C., and Young, R., Environ. Sci. Technol., 1995, 29, 2709. 12 Lopez-Avila, V., Young, R., Benedicto, J., Ho, P., and Kim, R., Anal. Chem., 1995, 67, 2096. 13 Dean, J. R., Anal. Comm., 1996, 33, 191. 14 Richter, B. E., Jones, B. A., Ezzel, J. L., and Porter, N. L., Anal. Chem., 1996, 68, 1033. 15 Frost, S. P., Dean, J. R., Evans, K. P., Harradine, K., Cary, C., and Comber, M. H. I., in preparation. Paper 7/02688I Received April 21, 1997 Accepted May 27, 1997 898 Analyst, September 1997, Vol. 122
ISSN:0003-2654
DOI:10.1039/a702688i
出版商:RSC
年代:1997
数据来源: RSC
|
7. |
Speciation Analysis of Chromium Using Cryptand Ethers |
|
Analyst,
Volume 122,
Issue 9,
1997,
Page 899-902
Elena Andrés García,
Preview
|
|
摘要:
Speciation Analysis of Chromium Using Cryptand Ethers Elena Andr�es Garc�ýa* and Domingo Blanco Gomis Departamento de Qu�ýmica F�ýsica y Anal�ýtica, Facultad de Qu�ýmica, Universidad de Oviedo, Oviedo, Spain The speciation analysis of chromium was studied using sequential solvent extraction of CrIII and CrVI (previously reduced to CrIII) with cryptand 2.2.1 as the ligand and eosin as a counter ion, in combination with spectrofluorimetry. A linear working range from the detection limit (1 ng ml21) to 150 ng ml21 of chromium was obtained. The proposed method was successfully applied to the determination of CrIII and CrVI in sea-water and chromium in certified coal samples.Keywords: Chromium; speciation; cryptand ether; ion-pair extraction; spectrofluorimetry Chromium is one of the essential trace elements in the human body; it appears to play a role in the metabolism of glucose and certain lipids, mainly cholesterol.1–3 However, nowadays the main reason for studying this element lies in its toxic nature.The biochemical toxicology of different compounds varies considerably with the chemical form and entrance route into the body. Thus, CrVI compounds are approximately 100 times more toxic than CrIII salts, owing to their high oxidation potential and the ease with which they penetrate biological membranes. Consequently, it is of great importance in environmental studies and studies of food contamination to determine the level and oxidation state of chromium since excessive amounts of this element, particularly in the more toxic CrVI form, are detrimental to health as this element may be involved in the pathogenesis of certain diseases such as lung and gastrointestinal cancers.Speciation studies of chromium have been carried out in natural waters,4 crayfish,5 galvanic waste waters,6 drilling fluid wastes,7 activated sludge,8 tannery effluent,9 welding dust,10 waste waters,11 etc. In the environment, air chromium particulates play a role in the oxidation of sulfur dioxide and the formation of acidic aerosols involved in global acid rain.12 The most common techniques for the determination of chromium are AAS employing flame or electrothermal atomization and ICP-AES, and more recently ICP-MS, although by themselves they only yield information on total concentrations.This is the reason why speciation analysis of Cr has already been investigated by means of the application of different techniques, such as precipitation, adsorption, solvent extraction and chromatography.6,13–15 One approach to this problem is to use macrobicyclic compounds (cryptand ethers) with cavities of the right size to accommodate the chromium metal cation.16 Cryptand ether 2.2.1 was used here.Although the product is not fluorescent, direct fluorimetric determination of the cation is possible by solvent extraction of the ion-pair formed by the cryptate complex and a highly fluorescent organic anion.In this paper, we describe the sequential extraction and spectrofluorimetric determination of trace amounts of CrIII and CrVI (previously reduced to CrIII with ethanol) by using cryptand 2.2.1 and eosin as a counter ion. The proposed method was successfully applied to the speciation analysis of chromium in sea-water and for the determination of chromium in samples of coal certified by the Community Bureau of Reference. Experimental Reagents All reagents were of analytical-reagent grade.Doubly distilled and de-mineralized water was used throughout. Stock standard solutions of CrIII and CrVI (1 g l21) were prepared by dissolving chromium sulfate and potassium dichromate in acidified water. All working standard solutions were freshly prepared by dilution of the appropriate stock standard solution with acidified water. Solutions of cryptand 2.2.1 (1024 m) were prepared by dissolving the commercial product (Kryptofix; Merck, Darmstadt, Germany) in water to which perchloric acid had been added and through which argon had been passed in order to remove carbon dioxide and to avoid carbonation of the cryptand.The solutions were stored in polyethylene flasks. Acidic eosin solutions (1024 m) were prepared by dissolving pure eosin (synthesized by reaction of Br2–BrO32 with fluorescein in acidified aqueous acetone17) in alkaline water (pH 8–9). Buffer solutions were prepared with 0.5 or 0.1 m TRIS-HCl.Apparatus Fluorescence intensity measurements and spectra were obtained with a Perkin-Elmer (Norwalk, CT, USA) LS-5 spectrofluorimeter equipped with a Model 3600 data station. The excitation and emission slit widths were both 2.5 nm and standard 1 cm silica cells were used. The temperature of the sample cell was kept constant within ±1 °C by using a Julabo Paratherm III thermostat system. A WTW-D812 Model 319, pH meter, calibrated against Radiometer (Copenhagen, Denmark) buffers, was used for pH measurements of the aqueous phase.Procedure Pipette standard CRIII and CRVI solutions into a 10 ml centrifuge tube, add 0.3 ml of the cryptand 2.2.1 stock standard solution, 1 ml of buffer solution and 0.3 ml of eosin solution and dilute to 5 ml with water (final pH 9 ± 0.1). After mixing, add 5 ml of 1,2-dichloroethane and extract the CrIII complex by shaking for 10 min. Allow the phases to separate and transfer 4 ml of the aqueous phase into a centrifuge tube, add 0.3 ml of ethanol to reduce CrVI to CrIII and 0.3 ml of the cryptand 2.2.1 stock standard solution and dilute to 5 ml with water.After mixing, add 5 ml of 1,2-dichloroethane and extract the new CrIII complex by shaking for 10 min. The organic phases were previously equilibrated with buffered aqueous phase. Measure the fluorescence intensity, If, of the 1,2-dichloroethane phases at 549 nm (excitation wavelength 534 nm). Run a reagent blank in the same way and subtract its fluorescence from that of the sample.For the analysis of coal, 2 g of the sample were first pretreated by adding a small volume of NaOH. Combustion then Analyst, September 1997, Vol. 122 (899–902) 899took place in an oxygen bomb, and finally the resulting sample was diluted to 50 ml with water. Results and Discussion Optimization of Extraction Conditions Various acidic fluorescent dyes of the fluorescein group [fluorescein, dichlorofluorescein, tetrabromofluorescein (eosin), tetraiodofluorescein (erythrosin) and tetrachlorotetrabromofluorescein (Rose Bengal)] as counter ions in different extraction solvents (toluene, chlorobenzene, carbon tetrachloride, chloroform, 1,2-dichloroethane and dichloromethane) were tested.The results showed that the system eosin– 1,2-dichloroethane gave the best fluorescence signal. Fig. 1 shows the excitation and emission spectra of the blank and the complex extracted by following the general procedure.The excitation spectrum has a maximum at 534 nm and the emission maximum is at 549 nm. The spectra were not corrected for variations in the emission characteristics of the lamp or for the response characteristics of the photomultiplier. A spectral bandpass of 2.5 nm was used for both absorption and fluorescence. The ion-pair extraction of chromium has a complicated dependence on pH, owing to the basic nature of the cryptand, the dissociation of eosin and hydrolysis of the cation.Taking into account the pKa values of mono- and diprotonated cryptand 2.2.1,18 the stability constant of the chromium hydroxide complex19 and the pKa values of eosin,20 it can be inferred that the ion pair would be extractable from an alkaline medium. Fig. 2 shows the effect of pH in the range 6–12 on the fluorescence intensity for 0.1 mg ml21 of chromium. The fluorescence is maximum with extraction at pH 8–10 and pH 9 (TRIS buffer) was selected for subsequent assays.Reagent Concentrations The effect of the variation of the cryptand 2.2.1 and eosin concentrations on the fluorescence signal of the extract was studied for a fixed amount (0.1 mg) of chromium and a single extraction step. Fig. 3 shows that the optimum concentrations are not less than 3.0 3 1026 m cryptand 2.2.1 (for a fixed eosin concentration of 3.8 3 1025 m) and 3.0 3 1026 m eosin (with a fixed cryptacentration of 3.8 3 1025 m ). The molar ratio of cryptand 2.2.1 to chromium was 2 and that of eosin to chromium was 1.5.Therefore, a stoichiometry 2 : 4 : 3 (metal : ligand : counter ion) was established, where the metal cation was located between two macrobicyclic rings with a sandwich type structure. The eosin to chromium ratio must be 3 : 2 so as to neutralize the charge. Rate of Extraction and Stability of the Extract The first extraction of CrIII is maximal after a 3 min shaking time. The fluorescence produced in the organic layer remains constant for at least 24 h under normal laboratory conditions.The rate of the second extraction is slower owing to the reduction of CrVI to CrIII by ethanol. In fact, the fluorescence signal reaches a maximum after a 10 min shaking time. In the extraction procedure, it is important to add the cryptand 2.2.1 before pH adjustment. The recommended order is chromium, cryptand 2.2.1, eosin and TRIS buffer solution. Effect of Ionic Strength The influence of ionic strength, I (from 0.01 to 0.25 m, adjusted with TRIS), is shown in Fig. 4. The fluorescence intensity of the organic phase, and therefore the extraction of the chromium ion pair, decrease quickly at I > 0.1. Preliminary results showed that increasing the ionic strength promotes poorer association of the metal–cryptand complex with the eosinate anion and hence extraction. Calibration Graph, Detection Limit and Precision The calibration graph was linear from the detection limit up to 150 ng ml21 of chromium.The detection limit (evaluated as the Fig. 1 (a) Excitation and (b) emission spectra of the chromium ionassociation complex and blank. Fig. 2 Variation of the fluorescence intensity of the chromium ionassociation complex with pH. Fig. 3 Influence of the concentration of (a) cryptand 2.2.1 and (b) eosin on fluorescence intensity. 900 Analyst, September 1997, Vol. 122concentration corresponding to three times the standard deviation of the blank signal) was 1 ng ml21.The RSD, evaluated by repeated analysis of standards containing 100 ng ml21, was 1.6%. Effect of Foreign Ions To evaluate the selectivity attainable, chromium was extracted in the presence of a number of foreign ions capable of forming stable complexes with cryptand 2.2.1 and certain common anions. The tolerance limit was set as the concentration of foreign ions that produced a variation in the apparent recovery of 100 ng ml21 of chromium greater than three times the RSD (i.e., variations > ±5%).The results are summarized in Table 1. It can be seen that large amounts of alkali metal ions do not interfere. Among the rest of the cations tested, thallium can be tolerated up to a 10-fold molar excess, silver, zinc and lead up to five-fold, calcium and mercury up to two-fold and cadmium and strontium only up to one-fold. The interference of cadmium and strontium cations could be reduced by increasing the ionic strength above 0.1 m since, in previous work,21,22 we have observed a strong decrease in the extraction of strontium and cadmium with increase in the ionic strength of the aqueous phase.The common anions chloride, sulfate, phosphate and nitrate had a negligible effect even for amounts above 104 times that of chromium. Determination of Chromium in Real Samples The proposed method was tested by applying it to the determination of chromium in sea-water and coal (obtained from the combustion of pulverized coal).The coal samples had been certified by the Community Bureau of Reference (BCR). The effect of sea-water matrices, collected at two points (near and far from a river mouth) on the beach at Gij�on (northern Spain), on the determination of low levels of chromium by the proposed method was investigated by recovery studies. Seawater samples were spiked with 10, 20 and 40 mg l21 of CrIII and CrVI and analysed using the proposed method. As can be seen in Table 2, neither of the samples is polluted with CrVI.However, both samples are polluted with CrIII, and the sea-water sample collected near the river mouth has a greater amount (25 mg l21) of CrIII. The recoveries ranged between 91.4 and 106.4%, showing acceptable precision. For the analysis of coal, a portion of the coal sample solution (2.5 ml) was extracted according to the proposed procedure. The results, given in Table 3, are in good agreement with the certified value. Conclusion The sensitivity, selectivity and applicability that can be achieved by using cryptands in the speciation of chromium have been demonstrated. Sequential ion-pair extraction was applied to the separation of the species in real samples, so that the Fig. 4 Effect of ionic strength on fluorescence intensity. Table 2 Recovery studies of CrIII and CrVI added to two sea-water samples Sea-water Chromium Concentration in Concentration Found/ Recovery sample species water/mg l21 added/mg l21 mg l21 (%) 10 34.5 95.0 A CrIII 25.0 20 46.3 106.4 40 66.3 104.9 10 24.7 96.0 B 15.1 20 35.7 103.2 40 54.8 99.3 10 9.1 91.4 A CrVI Not detected 20 20.1 100.5 40 40.7 101.6 10 9.8 98.0 B Not detected 20 20.0 100.0 40 40.4 101.0 Table 1 Effect of foreign ions (M) on the determination of 100 mg l21 of chromium Cation or M: Cr Apparent anion molar ratio recovery (%) Li+ 1000 100.3 Na+ 1000 98.2 K+ 1000 102.1 Cs+ 1000 99.6 NH4 + 500 105.6 Tl+ 10 95.7 Ag+ 5 98.2 Mg2+ 1000 96.4 Ca2+ 2 99.1 Sr2+ 1 94.8 Ba2+ 500 97.7 Cu2+ 10 99.3 Pb2+ 5 103.4 Cd2+ 1 105.3 Hg2+ 2 95.6 Ni2+ 500 96.9 Zn2+ 5 94.9 Co2+ 1000 100.6 Fe3+ 100 100.1 Al3+ 1000 98.8 NO32 20 000 101.7 Cl2 20 000 100.6 PO4 32 10 000 98.6 SO4 22 10 000 95.7 Analyst, September 1997, Vol. 122 901procedure described here can be used for speciation analysis of other elements if the extraction system is chosen adequately, since selectivity depends on the stability constants of the cryptate complexes and the nature of the counter ion and the solvent used for extraction.References 1 Anderson, R. A., Clin. Physiol. Biochem., 1986, 4, 31. 2 Merian, E., and Geldmacher, R., Metalle in der Umwelt, Verteilung, Analytik und Biologische Relevanz, VCH,Weinheim, 1984. 3 Versieck, J., and Cornelis, R., Trace Elements in Human Plasma or Serum, CRC Press, Boca Raton, FL, 1989. 4 Crespon Romero, R. M., Yebra Biurrun, M. C., and Bermejo Barrera, M. P., Anal. Chim. Acta, 1996, 327, 37. 5 Bundy, K., Millet, L., Bollinger, J., and Anderson, M., FASEB J., 1966, 10, 1007. 6 Andrle, C.M., Jakubowski, N., and Broekaert, J. A. C., Spectrochim. Acta, Part B, 1997, 52, 189. 7 Ghode, R., Muley, R., and Sarin, R., Chem. Speciation Bioavailability, 1995, 7, 133. 8 Imai, A., and Gloyna, E. F., Water Environ. Res., 1996, 68, 301. 9 Walsh, A. R., and Ohalloran, J., Water Res., 1996, 30, 2393. 10 Girard, L., and Huber, J., Talanta, 1966, 43, 1965. 11 Pantsarkallio, M., and Manninen, P.K. G., J. Chromatogr., 1996, 750, 89. 12 Fifield, F. W., and Haines, P. J., Environmental Analytical Chemistry, Blackie, Glasgow, 1995. 13 Lopez, A., Rotunno, T., Palmisano, F., Passino, R., Tiravanti, G., and Zambonin, G., Environ. Sci. Technol., 1991, 25, 1262. 14 Hill, S. J., Bloxham, M. J., and Worsfold, P. J., J. Anal. At. Spectrom., 1993, 8, 499. 15 Jakubowski, N., Jepkens, B., Stuewer, D., and Berndt, H., J. Anal. At. Spectrom., 1994, 9, 193. 16 Lehn, J. M., and Sauvage, J.P., J. Am. Chem. Soc., 1975, 97, 6700. 17 Fompeydie, D., Onur, F., and Levillain, Bull. Soc. Chim. Fr., 1982, II- 5. 18 Dietrich, B., Lehn, J. M., and Sauvage, J. P., Tetrahedron Lett., 1969, 2885. 19 Ringbom, A., Complexation in Analytical Chemistry, Alhambra, Madrid, 1979, p. 347. (Spanish translation). 20 Levillain, P., and Fompeydie D., Anal. Chem., 1985, 57, 2561. 21 Gomis, D. B., Alonso, E. F., Garc�ýa, E. A., and Abrodo, P. A., Talanta, 1989, 36, 1237. 22 Blanco, D., Fuente, E., and Arias, P., Microchim.Acta, 1989, 14, 59. Paper 7/01961K Received March 20, 1997 Accepted May 29, 1997 Table 3 Determination of CrIII in BCR coal certified reference material CrIII concentration/ mg g21 Coal Certified Proposed sample value method 1 31.3 ± 2.0 34.3 ± 1.8 2 32.7 ± 1.7 35.4 ± 1.5 902 Analyst, September 1997, Vol. 122 Speciation Analysis of Chromium Using Cryptathers Elena Andr�es Garc�ýa* and Domingo Blanco Gomis Departamento de Qu�ýmica F�ýsica y Anal�ýtica, Facultad de Qu�ýmica, Universidad de Oviedo, Oviedo, Spain The speciation analysis of chromium was studied using sequential solvent extraction of CrIII and CrVI (previously reduced to CrIII) with cryptand 2.2.1 as the ligand and eosin as a counter ion, in combination with spectrofluorimetry.A linear working range from the detection limit (1 ng ml21) to 150 ng ml21 of chromium was obtained. The proposed method was successfully applied to the determination of CrIII and CrVI in sea-water and chromium in certified coal samples.Keywords: Chromium; speciation; cryptand ether; ion-pair extraction; spectrofluorimetry Chromium is one of the essential trace elements in the human body; it appears to play a role in the metabolism of glucose and certain lipids, mainly cholesterol.1–3 However, nowadays the main reason for studying this element lies in its toxic nature. The biochemical toxicology of different compounds varies considerably with the chemical form and entrance route into the body.Thus, CrVI compounds are approximately 100 times more toxic than CrIII salts, owing to their high oxidation potential and the ease with which they penetrate biological membranes. Consequently, it is of great importance in environmental studies and studies of food contamination to determine the level and oxidation state of chromium since excessive amounts of this element, particularly in the more toxic CrVI form, are detrimental to health as this element may be involved in the pathogenesis of certain diseases such as lung and gastrointestinal cancers. Speciation studies of chromium have been carried out in natural waters,4 crayfish,5 galvanic waste waters,6 drilling fluid wastes,7 activated sludge,8 tannery effluent,9 welding dust,10 waste waters,11 etc.In the environment, air chromium particulates play a role in the oxidation of sulfur dioxide and the formation of acidic aerosols involved in global acid rain.12 The most common techniques for the determination of chromium are AAS employing flame or electrothermal atomization and ICP-AES, and more recently ICP-MS, although by themselves they only yield information on total concentrations.This is the reason why speciation analysis of Cr has already been investigated by means of the application of different techniques, such as precipitation, adsorption, solvent extraction and chromatography.6,13–15 One approach to this problem is to use macrobicyclic compounds (cryptand ethers) with cavities of the right size to accommodate the chromium metal cation.16 Cryptand ether 2.2.1 was used here.Although the product is not fluorescent, direct fluorimetric determination of the cation is possible by solvent extraction of the ion-pair formed by the cryptate complex and a highly fluorescent organic anion. In this paper, we describe the sequential extraction and spectrofluorimetric determination of trace amounts of CrIII and CrVI (previously reduced to CrIII with ethanol) by using cryptand 2.2.1 and eosin as a counter ion.The proposed method was successfully applied to the speciation analysis of chromium in sea-water and for the determination of chromium in samples of coal certified by the Community Bureau of Reference. Experimental Reagents All reagents were of analytical-reagent grade. Doubly distilled and de-mineralized water was used throughout. Stock standard solutions of CrIII and CrVI (1 g l21) were prepared by dissolving chromium sulfate and potassium dichromate in acidified water.All working standard solutions were freshly prepared by dilution of the appropriate stock standard solution with acidified water. Solutions of cryptand 2.2.1 (1024 m) were prepared by dissolving the commercial product (Kryptofix; Merck, Darmstadt, Germany) in water to which perchloric acid had been added and through which argon had been passed in order to remove carbon dioxide and to avoid carbonation of the cryptand. The solutions were stored in polyethylene flasks.Acidic eosin solutions (1024 m) were prepared by dissolving pure eosin (synthesized by reaction of Br2–BrO32 with fluorescein in acidified aqueous acetone17) in alkaline water (pH 8–9). Buffer solutions were prepared with 0.5 or 0.1 m TRIS-HCl. Apparatus Fluorescence intensity measurements and spectra were obtained with a Perkin-Elmer (Norwalk, CT, USA) LS-5 spectrofluorimeter equipped with a Model 3600 data station. The excitation and emission slit widths were both 2.5 nm and standard 1 cm silica cells were used.The temperature of the sample cell was kept constant within ±1 °C by using a Julabo Paratherm III thermostat system. A WTW-D812 Model 319, pH meter, calibrated against Radiometer (Copenhagen, Denmark) buffers, was used for pH measurements of the aqueous phase. Procedure Pipette standard CRIII and CRVI solutions into a 10 ml centrifuge tube, add 0.3 ml of the cryptand 2.2.1 stock standard solution, 1 ml of buffer solution and 0.3 ml of eosin solution and dilute to 5 ml with water (final pH 9 ± 0.1).After mixing, add 5 ml of 1,2-dichloroethane and extract the CrIII complex by shaking for 10 min. Allow the phases to separate and transfer 4 ml of the aqueous phase into a centrifuge tube, add 0.3 ml of ethanol to reduce CrVI to CrIII and 0.3 ml of the cryptand 2.2.1 stock standard solution and dilute to 5 ml with water.After mixing, add 5 ml of 1,2-dichloroethane and extract the new CrIII complex by shaking for 10 min. The organic phases were previously equilibrated with buffered aqueous phase. Measure the fluorescence intensity, If, of the 1,2-dichloroethane phases at 549 nm (excitation wavelength 534 nm). Run a reagent blank in the same way and subtract its fluorescence from that of the sample. For the analysis of coal, 2 g of the sample were first pretreated by adding a small volume of NaOH.Combustion then Analyst, September 1997, Vol. 122 (899–902) 899took place in an oxygen bomb, and finally the resulting sample was diluted to 50 ml with water. Results and Discussion Optimization of Extraction Conditions Various acidic fluorescent dyes of the fluorescein group [fluorescein, dichlorofluorescein, tetrabromofluorescein (eosin), tetraiodofluorescein (erythrosin) and tetrachlorotetrabromofluorescein (Rose Bengal)] as counter ions in different extraction solvents (toluene, chlorobenzene, carbon tetrachloride, chloroform, 1,2-dichloroethane and dichloromethane) were tested.The results showed that the system eosin– 1,2-dichloroethane gave the best fluorescence signal. Fig. 1 shows the excitation and emission spectra of the blank and the complex extracted by following the general procedure. The excitation spectrum has a maximum at 534 nm and the emission maximum is at 549 nm. The spectra were not corrected for variations in the emission characteristics of the lamp or for the response characteristics of the photomultiplier.A spectral bandpass of 2.5 nm was used for both absorption and fluorescence. The ion-pair extraction of chromium has a complicated dependence on pH, owing to the basic nature of the cryptand, the dissociation of eosin and hydrolysis of the cation. Taking into account the pKa values of mono- and diprotonated cryptand 2.2.1,18 the stability constant of the chromium hydroxide complex19 and the pKa values of eosin,20 it can be inferred that the ion pair would be extractable from an alkaline medium. Fig. 2 shows the effect of pH in the range 6–12 on the fluorescence intensity for 0.1 mg ml21 of chromium. The fluorescence is maximum with extraction at pH 8–10 and pH 9 (TRIS buffer) was selected for subsequent assays. Reagent Concentrations The effect of the variation of the cryptand 2.2.1 and eosin concentrations on the fluorescence signal of the extract was studied for a fixed amount (0.1 mg) of chromium and a single extraction step.Fig. 3 shows that the optimum concentrations are not less than 3.0 3 1026 m cryptand 2.2.1 (for a fixed eosin concentration of 3.8 3 1025 m) and 3.0 3 1026 m eosin (with a fixed cryptand 2.2.1 concentration of 3.8 3 1025 m ). The mo cryptand 2.2.1 to chromium was 2 and that of eosin to chromium was 1.5. Therefore, a stoichiometry 2 : 4 : 3 (metal : ligand : counter ion) was established, where the metal cation was located between two macrobicyclic rings with a sandwich type structure.The eosin to chromium ratio must be 3 : 2 so as to neutralize the charge. Rate of Extraction and Stability of the Extract The first extraction of CrIII is maximal after a 3 min shaking time. The fluorescence produced in the organic layer remains constant for at least 24 h under normal laboratory conditions. The rate of the second extraction is slower owing to the reduction of CrVI to CrIII by ethanol.In fact, the fluorescence signal reaches a maximum after a 10 min shaking time. In the extraction procedure, it is important to add the cryptand 2.2.1 before pH adjustment. The recommended order is chromium, cryptand 2.2.1, eosin and TRIS buffer solution. Effect of Ionic Strength The influence of ionic strength, I (from 0.01 to 0.25 m, adjusted with TRIS), is shown in Fig. 4. The fluorescence intensity of the organic phase, and therefore the extraction of the chromium ion pair, decrease quickly at I > 0.1.Preliminary results showed that increasing the ionic strength promotes poorer association of the metal–cryptand complex with the eosinate anion and hence extraction. Calibration Graph, Detection Limit and Precision The calibration graph was linear from the detection limit up to 150 ng ml21 of chromium. The detection limit (evaluated as the Fig. 1 (a) Excitation and (b) emission spectra of the chromium ionassociation complex and blank.Fig. 2 Variation of the fluorescence intensity of the chromium ionassociation complex with pH. Fig. 3 Influence of the concentration of (a) cryptand 2.2.1 and (b) eosin on fluorescence intensity. 900 Analyst, September 1997, Vol. 122concentration corresponding to three times the standard deviation of the blank signal) was 1 ng ml21. The RSD, evaluated by repeated analysis of standards containing 100 ng ml21, was 1.6%. Effect of Foreign Ions To evaluate the selectivity attainable, chromium was extracted in the presence of a number of foreign ions capable of forming stable complexes with cryptand 2.2.1 and certain common anions.The tolerance limit was set as the concentration of foreign ions that produced a variation in the apparent recovery of 100 ng ml21 of chromium greater than three times the RSD (i.e., variations > ±5%). The results are summarized in Table 1. It can be seen that large amounts of alkali metal ions do not interfere.Among the rest of the cations tested, thallium can be tolerated up to a 10-fold molar excess, silver, zinc and lead up to five-fold, calcium and mercury up to two-fold and cadmium and strontium only up to one-fold. The interference of cadmium and strontium cations could be reduced by increasing the ionic strength above 0.1 m since, in previous work,21,22 we have observed a strong decrease in the extraction of strontium and cadmium with increase in the ionic strength of the aqueous phase.The common anions chloride, sulfate, phosphate and nitrate had a negligible effect even for amounts above 104 times that of chromium. Determination of Chromium in Real Samples The proposed method was tested by applying it to the determination of chromium in sea-water and coal (obtained from the combustion of pulverized coal). The coal samples had been certified by the Community Bureau of Reference (BCR). The effect of sea-water matrices, collected at two points (near and far from a river mouth) on the beach at Gij�on (northern Spain), on the determination of low levels of chromium by the proposed method was investigated by recovery studies.Seawater samples were spiked with 10, 20 and 40 mg l21 of CrIII and CrVI and analysed using the proposed method. As can be seen in Table 2, neither of the samples is polluted with CrVI. However, both samples are polluted with CrIII, and the sea-water sample collected near the river mouth has a greater amount (25 mg l21) of CrIII.The recoveries ranged between 91.4 and 106.4%, showing acceptable precision. For the analysis of coal, a portion of the coal sample solution (2.5 ml) was extracted according to the proposed procedure. The results, given in Table 3, are in good agreement with the certified value. Conclusion The sensitivity, selectivity and applicability that can be achieved by using cryptands in the speciation of chromium have been demonstrated.Sequential ion-pair extraction was applied to the separation of the species in real samples, so that the Fig. 4 Effect of ionic strength on fluorescence intensity. Table 2 Recovery studies of CrIII and CrVI added to two sea-water samples Sea-water Chromium Concentration in Concentration Found/ Recovery sample species water/mg l21 added/mg l21 mg l21 (%) 10 34.5 95.0 A CrIII 25.0 20 46.3 106.4 40 66.3 104.9 10 24.7 96.0 B 15.1 20 35.7 103.2 40 54.8 99.3 10 9.1 91.4 A CrVI Not detected 20 20.1 100.5 40 40.7 101.6 10 9.8 98.0 B Not detected 20 20.0 100.0 40 40.4 101.0 Table 1 Effect of foreign ions (M) on the determination of 100 mg l21 of chromium Cation or M: Cr Apparent anion molar ratio recovery (%) Li+ 1000 100.3 Na+ 1000 98.2 K+ 1000 102.1 Cs+ 1000 99.6 NH4 + 500 105.6 Tl+ 10 95.7 Ag+ 5 98.2 Mg2+ 1000 96.4 Ca2+ 2 99.1 Sr2+ 1 94.8 Ba2+ 500 97.7 Cu2+ 10 99.3 Pb2+ 5 103.4 Cd2+ 1 105.3 Hg2+ 2 95.6 Ni2+ 500 96.9 Zn2+ 5 94.9 Co2+ 1000 100.6 Fe3+ 100 100.1 Al3+ 1000 98.8 NO32 20 000 101.7 Cl2 20 000 100.6 PO4 32 10 000 98.6 SO4 22 10 000 95.7 Analyst, September 1997, Vol. 122 901procedure described here can be used for speciation analysis of other elements if the extraction system is chosen adequately, since selectivity depends on the stability constants of the cryptate complexes and the nature of the counter ion and the solvent used for extraction. References 1 Anderson, R. A., Clin. Physiol. Biochem., 1986, 4, 31. 2 Merian, E., and Geldmacher, R., Metalle in der Umwelt, Verteilung, Analytik und Biologische Relevanz, VCH,Weinheim, 1984. 3 Versieck, J., and Cornelis, R., Trace Elements in Human Plasma or Serum, CRC Press, Boca Raton, FL, 1989. 4 Crespon Romero, R. M., Yebra Biurrun, M. C., and Bermejo Barrera, M. P., Anal. Chim. Acta, 1996, 327, 37. 5 Bundy, K., Millet, L., Bollinger, J., and Anderson, M., FASEB J., 1966, 10, 1007. 6 Andrle, C. M., Jakubowski, N., and Broekaert, J. A. C., Spectrochim. Acta, Part B, 1997, 52, 189. 7 Ghode, R., Muley, R., and Sarin, R., Chem. Speciation Bioavailability, 1995, 7, 133. 8 Imai, A., and Gloyna, E. F., Water Environ. Res., 1996, 68, 301. 9 Walsh, A. R., and Ohalloran, J., Water Res., 1996, 30, 2393. 10 Girard, L., and Huber, J., Talanta, 1966, 43, 1965. 11 Pantsarkallio, M., and Manninen, P. K. G., J. Chromatogr., 1996, 750, 89. 12 Fifield, F. W., and Haines, P. J., Environmental Analytical Chemistry, Blackie, Glasgow, 1995. 13 Lopez, A., Rotunno, T., Palmisano, F., Passino, R., Tiravanti, G., and Zambonin, G., Environ. Sci. Technol., 1991, 25, 1262. 14 Hill, S. J., Bloxham, M. J., and Worsfold, P. J., J. Anal. At. Spectrom., 1993, 8, 499. 15 Jakubowski, N., Jepkens, B., Stuewer, D., and Berndt, H., J. Anal. At. Spectrom., 1994, 9, 193. 16 Lehn, J. M., and Sauvage, J. P., J. Am. Chem. Soc., 1975, 97, 6700. 17 Fompeydie, D., Onur, F., and Levillain, Bull. Soc. Chim. Fr., 1982, II- 5. 18 Dietrich, B., Lehn, J. M., and Sauvage, J. P., Tetrahedron Lett., 1969, 2885. 19 Ringbom, A., Complexation in Analytical Chemistry, Alhambra, Madrid, 1979, p. 347. (Spanish translation). 20 Levillain, P., and Fompeydie D., Anal. Chem., 1985, 57, 2561. 21 Gomis, D. B., Alonso, E. F., Garc�ýa, E. A., and Abrodo, P. A., Talanta, 1989, 36, 1237. 22 Blanco, D., Fuente, E., and Arias, P., Microchim. Acta, 1989, 14, 59. Paper 7/01961K Received March 20, 1997 Accepted May 29, 1997 Table 3 Determination of CrIII in BCR coal certified reference material CrIII concentration/ mg g21 Coal Certified Proposed sample value method 1 31.3 ± 2.0 34.3 ± 1.8 2 32.7 ± 1.7 35.4 ± 1.5 902 Analyst, September 1997, Vol.
ISSN:0003-2654
DOI:10.1039/a701961k
出版商:RSC
年代:1997
数据来源: RSC
|
8. |
Correlation Between Inclusion Formation Constant and DistributionCoefficient in a Liquid–Liquid Extraction System Consisting ofHydrocarbon Solvents and Aqueous Dimethyl Sulfoxide Solutions ofβ-Cyclodextrin |
|
Analyst,
Volume 122,
Issue 9,
1997,
Page 903-910
Masaki Tachibana,
Preview
|
|
摘要:
Correlation Between Inclusion Formation Constant and Distribution Coefficient in a Liquid–Liquid Extraction System Consisting of Hydrocarbon Solvents and Aqueous Dimethyl Sulfoxide Solutions of b-Cyclodextrin Masaki Tachibana* and Nobutoshi Kiba Department of Applied Chemistry and Biotechnology, Faculty of Engineering, Yamanashi University, Kofu, Yamanashi 400, Japan Carbazole was used as a fluorescence probe for the investigation of the correlation between the apparent formation constant (Kf) for an inclusion complex of b-cyclodextrin (CD) and the distribution coefficient (Kd) on liquid–liquid extraction.Single batch extractions of carbazole were performed at 25 ± 1 °C from three types of hydrocarbon solvents into an aqueous dimethyl sulfoxide (DMSO) phase containing CD or nothing. Both the Kf and Kd values for carbazole can be determined simultaneously under the same conditions on the basis of the extraction data. A plot of log Kf versus log Kd gives a straight line in both extraction systems of heptane and dodecane, with good correlations (0.9996 and 0.9992) and similar slopes (20.675 and 20.672).The experimental results obtained from the cyclohexane–aqueous DMSO system suggest that the hydrocarbon used as the organic phase competes with carbazole for complexation with CD. The linear relationship is embodied in the form of an empirical equation, log KfOB = 20.673 log Kd + log (KfCR/KfHY), where KfOB is the observed Kf value for a 1 : 1 carbazole–CD complex in the aqueous DMSO medium and KfCR and KfHY are the absolute Kf values in water for the carbazole– and hydrocarbon–CD complexes, respectively.Keywords: Correlation; inclusion formation constant; distribution coefficient; liquid–liquid extraction; carbazole–b-cyclodextrin complexation Cyclodextrins (CDs) are cyclic oligosaccharides joined via a- 1,4-linkages of glucopyranose units into a cone-shaped torus. The most commonly used CDs are those containing six, seven and eight glucopyranose units, and are referred to as a-, b- and g-cyclodextrins, respectively.These compounds and the related derivatives are well known for their ability to form guest–host inclusion complexes with a variety of molecules on the basis of spatial fitting. Such a stereoselective interaction plays an important role in analytical chemistry. The knowledge of the formation constant (Kf) for a given complex is necessary to predict and understand better the interaction between cyclodextrin and the guest.Therefore, the Kf values for various cyclodextrin complexes have been evaluated by a wide variety of analytical techniques, including spectroscopy,1,2 chromatography, 3,4 capillary electrophoresis,5,6 pulse polarography7 and other methods.8–11 Several applications of CDs in extraction processes have also been reported.12–14 In such cases, CDs are normally added to the aqueous phase on liquid–liquid extraction in order to enhance partitioning of organic compounds from the organic phase. Size selectivity of CDs is often observed in complex samples, especially among polycyclic aromatic compounds (PACs).12,15 However, the effect of the additions has been little discussed in connection with the Kf values, and further with basic parameters on solvent extraction, such as the distribution ratio (D) and the distribution coefficient (Kd).The reason is probably the difficulty in determining these parameter values at the same time.In most studies, the application of CDs to extraction processes resulted in undesirable precipitation from the aqueous phase. A previous investigation in our laboratory showed that a few PACs could be selectively extracted from a hydrocarbon phase into a water–dimethyl sulfoxide (DMSO) solution of b- cyclodextrin (CD) without the precipitation process.16 In addition, the extraction technique was applied to the determination of the apparent Kf value for a 1 : 1 carbazole–CD complex in the aqueous DMSO medium.17 We noted that in this technique not only Kf but also Kd can be measured at the same time and under the same conditions.Both values of Kf and Kd will provide significant information regarding equilibrium of an analyte in a liquid–liquid extraction system containing CDs. There have been a few reports relating the Kf value to the Kd value in order to elucidate the binding forces contributing to the interaction of CDs with organic substances.18,19 However, the correlation between these two values has remained obscure because the Kd values were independently derived from extraction data without the interaction.In the present paper, a reliable correlation between Kf and Kd was studied in three types of extraction systems consisting of three different hydrocarbons and aqueous DMSO media, using carbazole as an appropriate probe. Although a good linear relationship between log Kf and log Kd for carbazole in a heptane–aqueous DMSO extraction system was observed previously,17 it was not within the scope of the previous study to confirm the relationship.In this work, the linearity of the log Kf versus log Kd plot was clearly demonstrated on the basis of extraction data for carbazole from both dodecane and heptane. Furthermore, the results from the cyclohexane system suggested the competitive CD complexation of carbazole with the hydrocarbon solvent. A plausible relationship between Kf and Kd was consequently proposed in the form of an empirical equation, based on some assumptions associated with these experimental results.We expect to find such a correlation between Kf and Kd for other species and/or extraction systems, although the findings obtained in our work may be limited to carbazole in hydrocarbon–aqueous DMSO two-phase systems. It will be worthwhile for the selective separation of complicated PAC samples to correlate the two equilibrium constants relating to the CD interaction and liquid–liquid distribution.Experimental Apparatus Synchronous spectrofluorimetric measurements were performed on a Hitachi (Tokyo, Japan) Model 650-40 spectrofluorimeter equipped with a Model 056 recorder. A 150 W xenon Analyst, September 1997, Vol. 122 (903–909) 903arc lamp was used as the excitation source and a 1.00 cm pathlength quartz cell as the measurement vessel. Both excitation and emission bandwidths were always set at 2.5 nm.All batch extractions were carried out using a Irie (Tokyo, Japan) TS shaker in an air-conditioned room at 25 ± 1 °C. Either 50 cm3 separating funnels or 15 3 160 mm test-tubes with well fitting ground glass stoppers were used for the extractions. Chemicals Carbazole was purified by zone melting after the removal of anthracene present as a main impurity with pre-treatments based on the Diels–Alder reaction with maleic anhydride.20 b- Cyclodextrin of guaranteed-reagent grade was purchased from Tokyo Kasei Kogyo (Tokyo, Japan) and used as received.Dimethyl sulfoxide was purified by repetition of the sequential operations of freezing partially, discarding the remaining liquid and re-melting the available solid.21 Heptane and cyclohexane were of analytical-reagent grade for fluorimetry (Kanto Chemical, Tokyo, Japan). Commercially available dodecane of guaranteed-reagent grade was further shaken with a mixture of concentrated sulfuric and nitric acid, washed with water, dried over anhydrous calcium chloride and finally passed through an activated alumina open column.Distilled water and ethanol were of analytical-reagent grade for HPLC (Kanto Chemical) and used as received. Extraction Procedure Three different stock standard solutions of carbazole were first prepared at a concentration of 500 mmol m23 in three hydrocarbon solvents, heptane, dodecane and cyclohexane. Working standard solutions were obtained by dilution of these stock standard solutions with the corresponding hydrocarbons at three different carbazole concentrations, 5.0, 10 and 25 mmol m23.Water–DMSO solutions of CD were prepared at eight consecutive concentrations in the range 0–10 mol m23. In these preparations, CD was initially dissolved in a suitable amount of DMSO and the DMSO solution was diluted with distilled water to give a 10 mol m23 CD concentration. The other CD solutions were then produced by diluting the 10 mol m23 solution with the corresponding water–DMSO medium.The DMSO concentrations in these media ranged from 10 to 60% v/v. Single batch extraction of carbazole was performed by shaking an aliquot (2–10 cm3) of the working solution with a suitable portion (1–16 cm3) of the CD solution for 3 min mechanically, either in a separating funnel or a test-tube with a tightly fitting glass stopper. In this extraction system the concentrations of carbazole in the two phases were almost unchanged after shaking for more than 1.5 min. Therefore, a shaking time of 3 min was adopted for rapid batch extractions.The separating funnels were used when the sum of the solution volumes was more than 11 cm3, but for smaller volumes the test-tubes were convenient. In the latter case, only the upper hydrocarbon layer to be measured was pipetted off, without discarding the bottom aqueous layer after the separation of the two phases. All extraction procedures were carried out in an airconditioned room at 25 ± 1 °C after solutions and glassware had been placed in it for at least 30 min to ensure temperature equilibrium.Synchronous Spectrofluorimetric Measurements The synchronous scanning technique was applied to the fluorimetric measurements of carbazole in hydrocarbon solvents before and after extractions. On extraction from 5.0 mmol m23 working standard solutions into various water– DMSO solutions of CD, synchronous fluorescence spectra of the separated hydrocarbon layers were measured directly with a wavelength interval (Dl) of 7 nm.The peak intensities at 337 nm in the spectra were obtained by the baseline method and the relative values were calculated by use of the intensity of a carbazole standard solution. The relative intensities were regarded as the fluorescence of carbazole remaining in the hydrocarbon phase and employed for calculating the distribution ratio (D) or distribution coefficient (Kd) for carbazole in the two-phase system.In such cases, a 5.0 mmol m23 hydrocarbon solution of carbazole saturated with the corresponding water– DMSO medium was used for the measurement of the initial intensity, instead of the original 5.0 mmol m23 solution. This is because the synchronous fluorescence intensities of carbazole in pure hydrocarbons decrease to a discernible extent owing to saturation with a trace amount of the water–DMSO medium. The medium-saturated hydrocarbon solutions of carbazole were prepared by adding 0.5 cm3 of the aqueous medium to 100 cm3 of the original solution, stirring the immiscible mixture vigorously and allowing it to clear on standing.The prepared 5.0 mmol m23 carbazole solution was also used as the abovementioned standard for evaluation of the relative synchronous fluorescence intensities. When extractions were performed at higher carbazole concentrations, the separated hydrocarbon phases were diluted with ethanol prior to the measurements in order to ensure a linear relationship.In these cases, an aliquot (3 cm3 at 10 mmol m23 and 2 cm3 at 25 mmol m23 carbazole concentrations) of the hydrocarbon layer was pipetted into a calibrated flask (5 and 10 cm3, respectively) and diluted to the mark with ethanol. Synchronous spectral intensities of the resulting hydrocarbon– ethanol mixtures were measured at a peak at 343 nm with Dl = 7 nm. As the standards, 6.0 and 5.0 mmol m23 carbazole solutions in 40 and 80% v/v ethanol mixtures, respectively, with the corresponding hydrocarbons were used.These solutions were readily prepared by diluting the carbazole working standard solutions with ethanol. The calibration graphs for carbazole were linear in the concentration ranges 0–5.0 and 0–6.0 mmol m23 using the three types of hydrocarbons and various hydrocarbon–ethanol mixtures. Determination of Kf and Kd Values Application of solvent extraction to the determination of the apparent formation constant for a 1 : 1 complex of carbazole with CD was detailed in a previous paper.17 On extraction of carbazole from a hydrocarbon solvent into an aqueous CD medium, according to the proposed theory, the following two equations are applicable at the same time: Kf = [CR -CD]a [CR]a [CD]a (1) D = + - [CR] [CR CD] [CR] a a h (2) where Kf and D are the apparent formation constant for the 1 : 1 carbazole–CD complex formed in the aqueous phase and the distribution ratio for carbazole in the two-phase extraction system, respectively, [CR], [CD] and [CR–CD] are the equilibrium concentrations of the free carbazole, free CD and complex and the subscripts a and h denote the aqueous and hydrocarbon phases, respectively. In the absence of CD, on the other hand, the distribution coefficient (Kd) for carbazole between the two phases is simply defined according to the basic theory for solvent extraction by Kd = [CR]a/[CR]h (3) Combination of eqns.(1), (2) and (3) and rearrangment yield 904 Analyst, September 1997, Vol. 122D = Kd + Kd Kf [CD]a (4) When the initial concentration of CD in the water–DMSO medium ([CD]a0) is in large excess compared with the concentration of the complex, the reasonable simplification [CD]a = [CD]a0 can be made. Eqn. (4) can be therefore expressed as D = Kd + Kd Kf [CD]a0 (5) In eqn. (5), the D and Kd values can be obtained experimentally by measuring the fluorescence intensities of carbazole present in the hydrocarbon phase before and after extraction. Also, a plot of D versus [CD]a0 would be found to be linear if the assumption of 1 : 1 stoichiometry is applicable to the inclusion complex.The value of Kf can consequently be determined from the slope and intercept of the linear plot according to eqn. (5). In our work, D and Kd were originally determined by synchronous fluorimetric measurements of hydrocarbon phases after extraction with and without CDs, respectively.The apparent Kf value was subsequently calculated by dividing the slope of the D versus [CD]a0 plot by the observed Kd value, not from the intercept of the plot. Since the D versus [CD]a0 plots tend to show upwards concave curvature at high CD concentrations, it seems reasonable to calculate in this way. Results and Discussion Heptane–Aqueous DMSO Extraction System Single batch extraction of carbazole from heptane into various water–DMSO media were performed 384 times at 25 ± 1 °C under partly different conditions, that is, solution volumes of hydrocarbon 2–10 cm3 and aqueous DMSO 1–16 cm3 and concentrations of carbazole 5.0, 10 and 25 mmol m23, CD 0–10 mol m23 and DMSO 10, 20, 30, 40, 50 and 60% v/v.The distribution ratio for carbazole determined from these extractions is plotted as a function of the initial concentration of CD (Fig. 1), together with the distribution coefficient, Kd = D when [CD]a0 = 0.Also, a few values accessible for the correlation between Kf and Kd were calculated from the extraction data. The parameter values given in Table 1 are grouped into six columns based on differences in the DMSO concentration. In each column, Kd was determined by averaging a series of values available from the eight different extractions with only the Fig. 1 Plot of the distribution ratio for carbazole versus the initial concentration of CD observed in a liquid–liquid extraction system consisting of heptane and aqueous DMSO (10–60% v/v).Analyst, September 1997, Vol. 122 905water–DMSO medium. The RSD was then calculated for the Kd, in order to assess the precision. Each D versus [CD]a0 plot is composed of the 64 data points measured at the same DMSO concentration. As can be seen from Fig. 1, deviation from linearity of the plot is observed at higher [CD]a0 and lower DMSO concentrations. In order to determine the Kf value according to eqn.(5), therefore, a linear regression analysis for the D versus [CD]a0 plot was applied to only 40 data points in the lower [CD]a0 range of 0–6 mol m23. The slope of the linear plot is given in Table 1, together with the correlation coefficient. Finally, the apparent Kf value for the carbazole–CD complex in the water–DMSO medium was calculated from dividing the slope by the observed Kd value as mentioned above. Table 1 indicates that the increase in Kd becomes pronounced with increasing DMSO concentration, having good RSD values at the higher concentrations.This is to be expected owing to the high solvent power of DMSO for carbazole. This power will result in a large variation in concentration of carbazole in the heptane phase before and after extraction. On the other hand, the slope of the D versus [CD]a0 plots does not increase as much as Kd, and the correlation coefficient goes through a maximum value at a DMSO concentration of 30% v/v. The smaller correlation at lower concentrations was attributable to the large deviation from the linear plots, and that at higher concentrations to the slight difference between the amounts of carbazole extracted with CD-free and -containing solutions.As can be seen from Table 1, the apparent Kf value given by the slope–Kd correlation decreases successively from 480 dm3 mol21 in 10% to 13.0 dm3 mol21 in 60% v/v DMSO medium. When the third component is dissolved in an aqueous CD medium as the organic modifier, other workers have suggested competitive complexation of the modifier compounds22,23 or the presence of a ternary complex.24,25 The decrease in Kf in this system may also be interpreted in terms of the assumption that the presence of DMSO in water breaks up the carbazole–CD complex competitively and assists carbazole in leaving the CD cavity.Dodecane–Aqueous DMSO Extraction System Carbazole in dodecane was extracted with water–DMSO media under the same conditions as described for the heptane–aqueous DMSO system, and the D versus [CD]a0 plot was constructed in the same manner (Fig. 2). Table 2 gives the parameter values for carbazole in this extraction system. Comparison of Table 2 with Table 1 reveals that there is little difference in the parameter values for carbazole between dodecane and heptane. This similarity should be predictable to some extent from the similar structures of these two straight-chain hydrocarbons. On closer inspection, however, it is found that there are some differences on comparing the dodecane with the heptane solvent.The values of Kd and the slope in the dodecane system are slightly smaller than the corresponding values in the heptane system, except for the slope at a 50% v/v DMSO concentration. The slight decrease in Kd can be ascribed to the difference in solubility of carbazole between the two hydrocarbons, but it is difficult to interpret the decrease in the slope of the D versus [CD]a0 plot.Further, the correlation for the D versus [CD]a0 plot in the dodecane system is poor compared with that in the heptane system, although it is better at a 10% v/v DMSO concentration. Probably this is due to the lower reproducibility of the synchronous spectrofluorimetric measurements of carbazole in dodecane. Cyclohexane–Aqueous DMSO Extraction System Extractions of carbazole from cyclohexane into CD solutions were tried under the same conditions as described in the above two systems.At DMSO concentrations < 30% v/v, however, it was impossible to estimate valid values of D and Kd for carbazole owing to precipitation from the aqueous DMSO solutions. Although the precipitation was observed even at 30% v/v DMSO at higher CD concentrations, the D values were measurable for [CD]a0 lower than approximately 6 mol m23. A possible explanation for this phenomenon has already been offered by Sanemasa et al.,26 who reported26 that CD is easily precipitated from its aqueous solution upon introducing cyclohexane vapour into the solution, coprecipitating PACs at the same time.Therefore, such precipitation observed in the less concentrated DMSO media may be regarded as due to a complicated complex of CD with cyclohexane (and in part with carbazole). Fig. 3 and Table 3 provide the D versus [CD]a0 plot and the parameter values for carbazole, respectively, for DMSO concentrations ranging only from 30 to 50% v/v.The data at 60% v/v DMSO concentration were not evaluated because of the insensitivity to the extraction of carbazole. In this cyclohexane system, an appreciably poorer correlation for the linear plots of D versus [CD]a0 is revealed by Fig. 3 and/or the correlation coefficients in Table 3. This poor correlation is due to the nature of the carbazole extraction, that is, a slight increase in the amount of carbazole with a large increase in the amount of CD. There may be some uncertainty in this system regarding the determination of the values of both the slope and the related Kf.However, the results for the cyclic hydrocarbon should be useful for comparison with those for the above straight-chain hydrocarbons. Table 3 indicates that the apparent Kf values at 30 and 40% v/v DMSO concentrations can be clearly distinguished from those in the heptane and dodecane systems, in spite of the similar decreases in Kd values. Anigbogu et al.3 pointed out that differences in the experimental conditions and probing techniques could result in significant differences in Kf.In this extraction technique, however, large differences in Kf values were observed under the same conditions and using the same method, although there is a difference in type of hydrocarbon. This implies that the hydrocarbons used as the organic phase compete with carbazole for CD complexation. It may therefore be concluded that the estimated Kf for the carbazole–CD Table 1 Parameter values obtained from data on solvent extraction of carbazole from heptane into various aqueous DMSO solutions of CD Concentration of DMSO (% v/v) Value 10 20 30 40 50 60 Mean Kd for 8 measurements 0.0216 0.0651 0.180 0.490 1.50 4.75 RSD of Kd measurements (%) 21.1 7.7 4.5 4.5 3.1 2.7 Slope of D versus [CD]a0 plot*/dm3 mol21 10.4 14.7 19.7 27.4 38.4 61.6 Correlation coefficient for the plot* 0.9600 0.9687 0.9808 0.9536 0.9311 0.7658 Kf (slope/Kd)/dm3 mol21 480 226 109 55.9 25.6 13.0 * Linear regression analysis was applied to 40 data points in the [CD]a0 range 0–6 mol m23 in the plot. 906 Analyst, September 1997, Vol. 122complex must be considered as a relative value to that for the hydrocarbon–CD complex. Linear Relationship Between log Kf and log Kd The apparent formation constant is a measure of the complexation of carbazole with CD in a given water–DMSO medium. On the other hand, the distribution coefficient is a measure of the distribution of carbazole between a hydrocarbon solvent and the medium in the absence of CD.Therefore, these two equilibrium constants, Kf and Kd, should be essentially independent of each other even if the hydrocarbon has a competitive effect on the carbazole–CD complexation. Application of linear regression analysis to each of the six data from the heptane and dodecane extraction systems led to the equations log Kf = 20.675 log Kd + 1.547 (r = 20.9996) and log Kf = 20.672 log Kd + 1.511 (r = 20.9992), respectively.Because of the small number of data points (n = 3), regression analysis was not applied to the results from the cyclohexane system. As can be seen from the above equations, Fig. 2 Plot of the distribution ratio for carbazole versus the initial concentration of CD observed in a liquid–liquid extraction system consisting of dodecane and aqueous DMSO (10–60% v/v). Table 2 Parameter values obtained from data on solvent extraction of carbazole from dodecane into various aqueous DMSO solutions of CD Concentration of DMSO (% v/v) Value 10 20 30 40 50 60 Mean Kd for 8 measurements 0.0201 0.0593 0.157 0.412 1.27 4.19 RSD of Kd measurements (%) 16.4 13.6 2.9 4.4 2.6 2.7 Slope of D versus [CD]a0 plot*/dm3 mol21 9.06 12.9 17.1 23.6 38.6 49.5 Correlation coefficient of the plot* 0.9744 0.9635 0.9581 0.9535 0.90006 0.6395 Kf (slope/Kd)/dm3 mol21 451 217 109 57.2 30.5 11.8 * Linear rergression analysis was applied to 40 data points in the [CD]a0 range 0–6 mol m23 in the plot.Analyst, September 1997, Vol. 122 907the plot of log Kf versus log Kd gives straight lines with good correlations (0.9996 and 0.9992) and similar slopes (20.675 and 20.672). These results suggest that the slope of the log Kf versus log Kd plot remains unaltered with changes to the extraction system. Fig. 4 illustrates log–log plots of the two equilibrium constants in all three hydrocarbon–aqueous DMSO extraction systems.The slopes of the three lines are identical, being the mean of the two values obtained from the heptane and dodecane systems. It is found that the above assumption is satisfactorily applicable to the case of the cyclohexane system. From these results, therefore, we conclude that there is a linear correlation between log Kf and log Kd for carbazole evaluated by the extraction method. In addition, the slope of the linear plot should be constant in different liquid–liquid extraction systems consisting of hydrocarbons and aqueous DMSO solutions.The reason for the unchanged slope is not clear, but it might arise from an unchanged fitting of the carbazole molecule into the CD cavity in different solvent environments. Interpretation of Empirical Equation On the above supposition, the correlation between Kf and Kd for carbazole in the hydrocarbon–aqueous DMSO systems can be expressed as log Kf = 20.673 log Kd + A (6) where A denotes the intercept of the linear plot and varies according to the hydrocarbon used as the organic phase in extraction.It is found from eqn. (6) that the value of Kf depends only on A when log Kd = 0 (Kd = 1). In addition, the apparent Kf value for carbazole can, in fact, be considered as a relative value resulting in competition with the hydrocarbon solvent. In order to generalize eqn. (6), we shall assume from the above viewpoint that Kf (observed) = Kf (carbazole)/Kf (hydrocarbon). On the basis of this additional assumption, therefore, eqn.(6) can be replaced by log KfOB = 20.673 log Kd + log (KfCR/KfHY) (7) where KfOB is the observed Kf value for a 1 : 1 carbazole–CD complex in a given water–DMSO medium and KfCR and KfHY are the absolute Kf values in neat water for the carbazole– and hydrocarbon–CD complexes, respectively. Eqn. (7) is a plausible empirical equation, but it is difficult to verify the validity at present because of the lack of both KfCR and KfHY values measured at the same time.Fortunately, however, some apparent KfHY values for heptane and cyclohexane have been published and could serve for the inspection of eqn. (7). When eqn. (6) is applied to the two different hydrocarbon solvents, heptane and cyclohexane, the intercepts of the corresponding linear plots are expressed according to eqn. (7) as AHP = log (KfCRKfHP) and ACY = log (KfCR/KfCY), where the subscripts HP and CY represent heptane and cyclohexane, respectively.Consequently, the difference between AHP and ACY can be written as follows: AHP2ACY = log (KfCR/KfHP)2log (KfCR/KfCY) = log (KfCR KfCY/KfHP KfCR) = log (KfCY/KfHP) (8) In this work, the statistically established values of A for heptane, dodecane and cyclohexane were 1.548, 1.510 and 1.238, respectively. Substitution of the two values of interest into eqn. (8) resulted in log (KfCY/KfHP) = 0.310, that is, KfCY/ KfHP = 2.04. Sanemasa and co-workers27,28 studied the KfHY values in an aqueous medium at 25 °C by making use of the volatilization rate of hydrocarbon molecules. The KfHP and KfCY values that they determined were 69 ± 4 and 156 ± 8 dm3 mol21, respectively.Therefore, a simple calculation from their data gives a KfCY/KfHP ratio of 2.26 (with variations in the range 2.03–2.52). On the other hand, Wishnia and Lappi29 presented the dissociation constant; which is the reciprocal of the formation constant (1/Kf) for heptane– and cyclohexane–CD 1 : 1 complexes in the range 0–50 °C.The values given in their table for the heptane and cyclohexane complexes were 0.37 and 0.182 mol m23, respectively, at 20 °C and 0.39 and 0.199 mol m23, respectively, at 30 °C. Further application of their results indicated that KfCY/KfHP = 2.03 at 20 °C and 1.96 at 30 °C. These values of the KfCY/KfHP ratio are in fair agreement with the value calculated in our work. Therefore, these results suggest that there is a reasonable correlation between KfOB and Kd values for carbazole, and also that the correlation can be expressed in the form of the empirical eqn.(7). Fig. 3 Plot of the distribution ratio for carbazole versus the initial concentration of CD observed in a liquid–liquid extraction system consisting of cyclohexane and aqueous DMSO (30–50% v/v). 908 Analyst, September 1997, Vol. 122Conclusion Liquid–liquid extraction of carbazole was performed from three types of hydrocarbon solvents into water–DMSO solutions of CD in order to determine the values of the apparent formation constant and the distribution coefficient at the same time. A linear relationship between the log Kf and log Kd for carbazole was substantiated on the basis of the extraction data.Since the slope of the log Kf versus log Kd plot obtained with the dodecane system was similar to that with the heptane system, it was assumed that the slope does not vary with changes in hydrocarbon phase.The reason for the unchanged slope is not clear at the present stage of our limited study. Further and more extensive studies may be necessary in order to confirm the assumption. From the results with the cyclohexane system, on the other hand, it was concluded that the apparent Kf value observed for a 1 : 1 carbazole–CD complex is a relative value resulting in competition with the hydrocarbon solvent. An empirical equation was finally proposed as a reasonable correlation between the observed Kf and Kd values for carbazole in these hydrocarbon–aqueous DMSO extraction systems.Emphasis was placed on the interpretation of the intercept of the linear log Kf versus log Kd plot. Although it is unknown what the slope of the plot represents, the concept of the relative Kf value as the intercept was pertinent to the empirical equation. Also, it is not clear whether such a correlation can be found or not for species other than carbazole. However, the good linearity of the log Kf versus log Kd plot may be attractive in further analytical investigations, such as for the prediction of selective separations from complicated mixtures and application of CD-containing solvent extraction to practical samples.References 1 Mu�noz de la Pe�na, A., Ndou, T., Zung, J. B., and Warner, I. M., J. Phys. Chem., 1991, 95, 3330. 2 Chokchainarong, S., Fennema, O. R., and Connors, K. A., Carbohydr. Res., 1992, 232, 161. 3 Anigbogu, V. C., Mu�noz de la Pe�na, A., Ndou, T.T., and Warner, I. M., Anal. Chem., 1992, 64, 484. 4 Loukas, Y. L., Antoniadou-Vyza, E., Papadaki-Valiraki, A., and Machera, K. G., J. Agric. Food Chem., 1994, 42, 944. 5 Gareil, P., Pernin, D., Gramond, J.-P., and Guyon, F., J. High Resolut. Chromatogr., 1993, 16, 195. 6 Baumy, Ph., Morin, Ph., Dreux, M., Viaud, M. C., Boye, S., and Guillaumet, G., J. Chromatogr. A, 1995, 707, 311. 7 Choi, H.-S., Chang, C.-J., and Knevel, A. M., Pharm. Res., 1992, 9, 582. 8 Georgiou, M. E., Georgiou, C. A., and Koupparis, M. A., Anal. Chem., 1995, 67, 114. 9 Li, S., and Purdy, W. C., Anal. Chem., 1992, 64, 1405. 10 Valsami, G. N., Koupparis, M. A., and Macheras, P. E., Pharm. Res., 1992, 9, 94. 11 McCormack, S., Russell, N. R., and Cassidy, J. F., Electrochim. Acta, 1992, 37, 1939. 12 Blyshak, L. A., Rossi, T. M., Patonay, G., and Warner, I. M., Anal. Chem., 1988, 60, 2127. 13 Harangi, J., and N�an�asi, P., Anal. Chim. Acta, 1984, 156, 103. 14 Matsunaga, K., Imanaka, M., Ishida, T., and Oda, T., Anal.Chem., 1984, 56, 1980. 15 Elliott, N. B., Prenni, A. J., Ndou, T. T., and Warner, I. M., J. Colloid Interface Sci., 1993, 156, 359. 16 Tachibana, M., and Furusawa, M., Analyst, 1995, 120, 437. 17 Tachibana, M., Furusawa, M., and Kiba, N., J. Inclusion Phenom. Mol. Recognit. Chem., 1995, 22, 313. 18 Matsui, Y., and Mochida, K., Bull. Chem. Soc. Jpn., 1979, 52, 2808. 19 Uekama, K., Hirayama, F., Nasu, S., Matsuo, N., and Irie, T., Chem.Pharm. Bull., 1978, 26, 3477. 20 Furusawa, M., Takeuido, K., and Shimizu, H., Kogyo Kagaku Zasshi, 1963, 66, 1811. 21 Tachibana, M., and Furusawa, M., Bull. Chem. Soc. Jpn., 1988, 61, 2353. 22 Mohseni, R. M., and Hurtubise, R. J., J. Chromatogr., 1990, 499, 395. 23 Mu�noz de la Pe�na, A., Ndou, T. T., Anigbogu, V. C., and Warner, I. M., Anal. Chem., 1991, 63, 1018. 24 Kano, K., Hashimoto, S., Imai, A., and Ogawa, T., J. Inclusion Phenom., 1984, 2, 737. 25 Mu�noz de la Pe�na, A., Ndou, T. T., Zung, J. B., Greene, K. L., Live, D. H., and Warner, I. M., J. Am. Chem. Soc., 1991, 113, 1572. 26 Sanemasa, I., Koga, I., and Deguchi, T., Anal. Sci., 1991, 7, 641. 27 Sanemasa, I., Osajima, T., and Deguchi, T., Bull. Chem. Soc. Jpn., 1990, 63, 2814. 28 Osajima, T., Deguchi, T., and Sanemasa, I., Bull. Chem. Soc. Jpn., 1991, 64, 2705. 29 Wishnia, A., and Lappi, S. J., J. Mol. Biol., 1974, 82, 77. Paper 7/01843F Received March 17, 1997 Accepted May 29, 1997 Table 3 Parameter values obtained from data on solvent extraction of carbazole from cyclohexane into various aqueous DMSO solutions of CD Concentration of DMSO (% v/v) Value 30 40 50 Mean Kd for 8 measurements 0.126 0.365 1.07 RSD of Kd measurements (%) 2.9 4.4 2.6 Slope of D versus [CD]a0 plot*/ dm3 mol21 8.69 9.99 22.3 Correlation coefficient for the plot* 0.8794 0.7755 0.6656 Kf (slope/Kd)/dm3 mol21 68.9 27.4 20.9 * Linear regression analysis was applied to 40 data points in the [CD]a0 range 0–6 mol m23 in the plot.Fig. 4 Correlation between the apparent Kf value for a 1 : 1 carbazole–CD complex present in water–DMSO medium and the Kd value for carbazole on extraction from (2) heptane, (8) dodecane and (Ç) cyclohexane into the medium. Analyst, September 1997, Vol. 122 909 Correlation Between Inclusion Formation Constant and Distribution Coefficient in a Liquid–Liquid Extraction System Consisting of Hydrocarbon Solvents and Aqueous Dimethyl Sulfoxide Solutions of b-Cyclodextrin Masaki Tachibana* and Nobutoshi Kiba Department of Applied Chemistry and Biotechnology, Faculty of Engineering, Yamanashi University, Kofu, Yamanashi 400, Japan Carbazole was used as a fluorescence probe for the investigation of the correlation between the apparent formation constant (Kf) for an inclusion complex of b-cyclodextrin (CD) and the distribution coefficient (Kd) on liquid–liquid extraction. Single batch extractions of carbazole were performed at 25 ± 1 °C from three types of hydrocarbon solvents into an aqueous dimethyl sulfoxide (DMSO) phase containing CD or nothing.Both the Kf and Kd values for carbazole can be determined simultaneously under the same conditions on the basis of the extraction data. A plot of log Kf versus log Kd gives a straight line in both extraction systems of heptane and dodecane, with good correlations (0.9996 and 0.9992) and similar slopes (20.675 and 20.672).The experimental results obtained from the cyclohexane–aqueous DMSO system suggest that the hydrocarbon used as the organic phase competes with carbazole for complexation with CD. The linear relationship is embodied in the form of an empirical equation, log KfOB = 20.673 log Kd + log (KfCR/KfHY), where KfOB is the observed Kf value for a 1 : 1 carbazole–CD complex in the aqueous DMSO medium and KfCR and KfHY are the absolute Kf values in water for the carbazole– and hydrocarbon–CD complexes, respectively. Keywords: Correlation; inclusion formation constant; distribution coefficient; liquid–liquid extraction; carbazole–b-cyclodextrin complexation Cyclodextrins (CDs) are cyclic oligosaccharides joined via a- 1,4-linkages of glucopyranose units into a cone-shaped torus.The most commonly used CDs are those containing six, seven and eight glucopyranose units, and are referred to as a-, b- and g-cyclodextrins, respectively. These compounds and the related derivatives are well known for their ability to form guest–host inclusion complexes with a variety of molecules on the basis of spatial fitting.Such a stereoselective interaction plays an important role in analytical chemistry. The knowledge of the formation constant (Kf) for a given complex is necessary to predict and understand better the interaction between cyclodextrin and the guest. Therefore, the Kf values for various cyclodextrin complexes have been evaluated by a wide variety of analytical techniques, including spectroscopy,1,2 chromatography, 3,4 capillary electrophoresis,5,6 pulse polarography7 and other methods.8–11 Several applications of CDs in extraction processes have also been reported.12–14 In such cases, CDs are normally added to the aqueous phase on liquid–liquid extraction in order to enhance partitioning of organic compounds from the organic phase.Size selectivity of CDs is often observed in complex samples, especially among polycyclic aromatic compounds (PACs).12,15 However, the effect of the additions has been little discussed in connection with the Kf values, and further with basic parameters on solvent extraction, such as the distribution ratio (D) and the distribution coefficient (Kd).The reason is probably the difficulty in determining these parameter values at the same time. In most studies, the application of CDs to extraction processes resulted in undesirable precipitation from the aqueous phase.A previous investigation in our laboratory showed that a few PACs could be selectively extracted from a hydrocarbon phase into a water–dimethyl sulfoxide (DMSO) solution of b- cyclodextrin (CD) without the precipitation process.16 In addition, the extraction technique was applied to the determination of the apparent Kf value for a 1 : 1 carbazole–CD complex in the aqueous DMSO medium.17 We noted that in this technique not only Kf but also Kd can be measured at the same time and under the same conditions. Both values of Kf and Kd will provide significant information regarding equilibrium of an analyte in a liquid–liquid extraction system containing CDs.There have been a few reports relating the Kf value to the Kd value in order to elucidate the binding forces contributing to the interaction of CDs with organic substances.18,19 However, the correlation between these two values has remained obscure because the Kd values were independently derived from extraction data without the interaction.In the present paper, a reliable correlation between Kf and Kd was studied in three types of extraction systems consisting of three different hydrocarbons and aqueous DMSO media, using carbazole as an appropriate probe. Although a good linear relationship between log Kf and log Kd for carbazole in a heptane–aqueous DMSO extraction system was observed previously,17 it was not within the scope of the previous study to confirm the relationship. In this work, the linearity of the log Kf versus log Kd plot was clearly demonstrated on the basis of extraction data for carbazole from both dodecane and heptane.Furthermore, the results from the cyclohexane system suggested the competitive CD complexation of carbazole with the hydrocarbon solvent. A plausible relationship between Kf and Kd was consequently proposed in the form of an empirical equation, based on some assumptions associated with these experimental results.We expect to find such a correlation between Kf and Kd for other species and/or extraction systems, although the findings obtained in our work may be limited to carbazole in hydrocarbon–aqueous DMSO two-phase systems. It will be worthwhile for the selective separation of complicated PAC samples to correlate the two equilibrium constants relating to the CD interaction and liquid–liquid distribution. Experimental Apparatus Synchronous spectrofluorimetric measurements were performed on a Hitachi (Tokyo, Japan) Model 650-40 spectrofluorimeter equipped with a Model 056 recorder.A 150 W xenon Analyst, September 1997, Vol. 122 (903–909) 903arc lamp was used as the excitation source and a 1.00 cm pathlength quartz cell as the measurement vessel. Both excitation and emission bandwidths were always set at 2.5 nm. All batch extractions were carried out using a Irie (Tokyo, Japan) T in an air-conditioned room at 25 ± 1 °C.Either 50 cm3 separating funnels or 15 3 160 mm test-tubes with well fitting ground glass stoppers were used for the extractions. Chemicals Carbazole was purified by zone melting after the removal of anthracene present as a main impurity with pre-treatments based on the Diels–Alder reaction with maleic anhydride.20 b- Cyclodextrin of guaranteed-reagent grade was purchased from Tokyo Kasei Kogyo (Tokyo, Japan) and used as received. Dimethyl sulfoxide was purified by repetition of the sequential operations of freezing partially, discarding the remaining liquid and re-melting the available solid.21 Heptane and cyclohexane were of analytical-reagent grade for fluorimetry (Kanto Chemical, Tokyo, Japan).Commercially available dodecane of guaranteed-reagent grade was further shaken with a mixture of concentrated sulfuric and nitric acid, washed with water, dried over anhydrous calcium chloride and finally passed through an activated alumina open column.Distilled water and ethanol were of analytical-reagent grade for HPLC (Kanto Chemical) and used as received. Extraction Procedure Three different stock standard solutions of carbazole were first prepared at a concentration of 500 mmol m23 in three hydrocarbon solvents, heptane, dodecane and cyclohexane. Working standard solutions were obtained by dilution of these stock standard solutions with the corresponding hydrocarbons at three different carbazole concentrations, 5.0, 10 and 25 mmol m23. Water–DMSO solutions of CD were prepared at eight consecutive concentrations in the range 0–10 mol m23.In these preparations, CD was initially dissolved in a suitable amount of DMSO and the DMSO solution was diluted with distilled water to give a 10 mol m23 CD concentration. The other CD solutions were then produced by diluting the 10 mol m23 solution with the corresponding water–DMSO medium. The DMSO concentrations in these media ranged from 10 to 60% v/v.Single batch extraction of carbazole was performed by shaking an aliquot (2–10 cm3) of the working solution with a suitable portion (1–16 cm3) of the CD solution for 3 min mechanically, either in a separating funnel or a test-tube with a tightly fitting glass stopper. In this extraction system the concentrations of carbazole in the two phases were almost unchanged after shaking for more than 1.5 min. Therefore, a shaking time of 3 min was adopted for rapid batch extractions. The separating funnels were used when the sum of the solution volumes was more than 11 cm3, but for smaller volumes the test-tubes were convenient. In the latter case, only the upper hydrocarbon layer to be measured was pipetted off, without discarding the bottom aqueous layer after the separation of the two phases.All extraction procedures were carried out in an airconditioned room at 25 ± 1 °C after solutions and glassware had been placed in it for at least 30 min to ensure temperature equilibrium.Synchronous Spectrofluorimetric Measurements The synchronous scanning technique was applied to the fluorimetric measurements of carbazole in hydrocarbon solvents before and after extractions. On extraction from 5.0 mmol m23 working standard solutions into various water– DMSO solutions of CD, synchronous fluorescence spectra of the separated hydrocarbon layers were measured directly with a wavelength interval (Dl) of 7 nm. The peak intensities at 337 nm in the spectra were obtained by the baseline method and the relative values were calculated by use of the intensity of a carbazole standard solution.The relative intensities were regarded as the fluorescence of carbazole remaining in the hydrocarbon phase and employed for calculating the distribution ratio (D) or distribution coefficient (Kd) for carbazole in the two-phase system. In such cases, a 5.0 mmol m23 hydrocarbon solution of carbazole saturated with the corresponding water– DMSO medium was used for the measurement of the initial intensity, instead of the original 5.0 mmol m23 solution.This is because the synchronous fluorescence intensities of carbazole in pure hydrocarbons decrease to a discernible extent owing to saturation with a trace amount of the water–DMSO medium. The medium-saturated hydrocarbon solutions of carbazole were prepared by adding 0.5 cm3 of the aqueous medium to 100 cm3 of the original solution, stirring the immiscible mixture vigorously and allowing it to clear on standing.The prepared 5.0 mmol m23 carbazole solution was also used as the abovementioned standard for evaluation of the relative synchronous fluorescence intensities. When extractions were performed at higher carbazole concentrations, the separated hydrocarbon phases were diluted with ethanol prior to the measurements in order to ensure a linear relationship. In these cases, an aliquot (3 cm3 at 10 mmol m23 and 2 cm3 at 25 mmol m23 carbazole concentrations) of the hydrocarbon layer was pipetted into a calibrated flask (5 and 10 cm3, respectively) and diluted to the mark with ethanol. Synchronous spectral intensities of the resulting hydrocarbon– ethanol mixtures were measured at a peak at 343 nm with Dl = 7 nm.As the standards, 6.0 and 5.0 mmol m23 carbazole solutions in 40 and 80% v/v ethanol mixtures, respectively, with the corresponding hydrocarbons were used. These solutions were readily prepared by diluting the carbazole working standard solutions with ethanol. The calibration graphs for carbazole were linear in the concentration ranges 0–5.0 and 0–6.0 mmol m23 using the three types of hydrocarbons and various hydrocarbon–ethanol mixtures.Determination of Kf and Kd Values Application of solvent extraction to the determination of the apparent formation constant for a 1 : 1 complex of carbazole with CD was detailed in a previous paper.17 On extraction of carbazole from a hydrocarbon solvent into an aqueous CD medium, according to the proposed theory, the following two equations are applicable at the same time: Kf = [CR -CD]a [CR]a [CD]a (1) D = + - [CR] [CR CD] [CR] a a h (2) where Kf and D are the apparent formation constant for the 1 : 1 carbazole–CD complex formed in the aqueous phase and the distribution ratio for carbazole in the two-phase extraction system, respectively, [CR], [CD] and [CR–CD] are the equilibrium concentrations of the free carbazole, free CD and complex and the subscripts a and h denote the aqueous and hydrocarbon phases, respectively. In the absence of CD, on the other hand, the distribution coefficient (Kd) for carbazole between the two phases is simply defined according to the basic theory for solvent extraction by Kd = [CR]a/[CR]h (3) Combination of eqns.(1), (2) and (3) and rearrangment yield 904 Analyst, September 1997, Vol. 122D = Kd + Kd Kf [CD]a (4) When the initial concentration of CD in the water–DMSO medium ([CD]a0) is in large excess compared with the concentration of the complex, the reasonable simplification [CD]a = [CD]a0 can be made.Eqn. (4) can be therefore expressed as D = Kd + Kd Kf [CD]a0 (5) In eqn. (5), the D and Kd values can be obtained experimentally by measuring the fluorescence intensities of carbazole present in the hydrocarbon phase before and after extraction. Also, a plot of D versus [CD]a0 would be found to be linear if the assumption of 1 : 1 stoichiometry is applicable to the inclusion complex.The value of Kf can consequently be determined from the slope and intercept of the linear plot according to eqn. (5). In our work, D and Kd were originally determined by synchronous fluorimetric measurements of hydrocarbon phases after extraction with and without CDs, respectively. The apparent Kf value was subsequently calculated by dividing the slope of the D versus [CD]a0 plot by the observed Kd value, not from the intercept of the plot.Since the D versus [CD]a0 plots tend to show upwards concave curvature at high CD concentrations, it seems reasonable to calculate in this way. Results and Discussion Heptane–Aqueous DMSO Extraction System Single batch extraction of carbazole from heptane into various water–DMSO media were performed 384 times at 25 ± 1 °C under partly different conditions, that is, solution volumes of hydrocarbon 2–10 cm3 and aqueous DMSO 1–16 cm3 and concentrations of carbazole 5.0, 10 and 25 mmol m23, CD 0–10 mol m23 and DMSO 10, 20, 30, 40, 50 and 60% v/v.The distribution ratio for carbazole determined from these extractions is plotted as a function of the initial concentration of CD (Fig. 1), together with the distribution coefficient, Kd = D when [CD]a0 = 0. Also, a few values accessible for the correlation between Kf and Kd were calculated from the extraction data. The parameter values given in Table 1 are grouped into six columns based on differences in the DMSO concentration.In each column, Kd was determined by averaging a series of values available from the eight different extractions with only the Fig. 1 Plot of the distribution ratio for carbazole versus the initial concentration of CD observed in a liquid–liquid extraction system consisting of heptane and aqueous DMSO (10–60% v/v). Analyst, September 1997, Vol. 122 905water–DMSO medium. The RSD was then calculated for the Kd, in order to assess the precision. Each D versus [CD]a0 plot is composed of the 64 data points measured at the same DMSO concentration.As can be seen from Fig. 1, deviation from linearity of the plot is observed at higher [CD]a0 and lower DMSO concentrations. In order to determine the Kf value according to eqn. (5), therefore, a linear regression analysis for the D versus [CD]a0 plot was applied to only 40 data points in the lower [CD]a0 range of 0–6 mol m23.The slope of the linear plot is given in Table 1, together with the correlation coefficient. Finally, the apparent Kf value for the carbazole–CD complex in the water–DMSO medium was calculated from dividing the slope by the observed Kd value as mentioned above. Table 1 indicates that the increase in Kd becomes pronounced with increasing DMSO concentration, having good RSD values at the higher concentrations. This is to be expected owing to the high solvent power of DMSO for carbazole.This power will result in a large variation in concentration of carbazole in the heptane phase before and after extraction. On the other hand, the slope of the D versus [CD]a0 plots does not increase as much as Kd, and the correlation coefficient goes through a maximum value at a DMSO concentration of 30% v/v. The smaller correlation at lower concentrations was attributable to the large deviation from the linear plots, and that at higher concentrations to the slight difference between the amounts of carbazole extracted with CD-free and -containing solutions. As can be seen from Table 1, the apparent Kf value given by the slope–Kd correlation decreases successively from 480 dm3 mol21 in 10% to 13.0 dm3 mol21 in 60% v/v DMSO medium.When the third component is dissolved in an aqueous CD medium as the organic modifier, other workers have suggested competitive complexation of the modifier compounds22,23 or the presence of a ternary complex.24,25 The decrease in Kf in this system may also be interpreted in terms of the assumption that the presence of DMSO in water breaks up the carbazole–CD complex competitively and assists carbazole in leaving the CD cavity.Dodecane–Aqueous DMSO Extraction System Carbazole in dodecane was extracted with water–DMSO media under the same conditions as described for the heptane–aqueous DMSO system, and the D versus [CD]a0 plot was constructed in the same manner (Fig. 2).Table 2 gives the parameter values for carbazole in this extraction system. Comparison of Table 2 with Table 1 reveals that there is little difference in the parameter values for carbazole between dodecane and heptane. This similarity should be predictable to some extent from the similar structures of these two straight-chain hydrocarbons. On closer inspection, however, it is found that there are some differences on comparing the dodecane with the heptane solvent. The values of Kd and the slope in the dodecane system are slightly smaller than the corresponding values in the heptane system, except for the slope at a 50% v/v DMSO concentration.The slight decrease in Kd can be ascribed to the difference in solubility of carbazole between the two hydrocarbons, but it is difficult to interpret the decrease in the slope of the D versus [CD]a0 plot. Further, the correlation for the D versus [CD]a0 plot in the dodecane system is poor compared with that in the heptane system, although it is better at a 10% v/v DMSO concentration.Probably this is due to the lower reproducibility of the synchronous spectrofluorimetric measurements of carbazole in dodecane. Cyclohexane–Aqueous DMSO Extraction System Extractions of carbazole from cyclohexane into CD solutions were tried under the same conditions as described in the above two systems. At DMSO concentrations < 30% v/v, however, it was impossible to estimate valid values of D and Kd for carbazole owing to precipitation from the aqueous DMSO solutions.Although the precipitation was observed even at 30% v/v DMSO at higher CD concentrations, the D values were measurable for [CD]a0 lower than approximately 6 mol m23. A possible explanation for this phenomenon has already been offered by Sanemasa et al.,26 who reported26 that CD is easily precipitated from its aqueous solution upon introducing cyclohexane vapour into the solution, coprecipitating PACs at the same time.Therefore, such precipitation observed in the less concentrated DMSO media may be regarded as due to a complicated complex of CD with cyclohexane (and in part with carbazole). Fig. 3 and Table 3 provide the D versus [CD]a0 plot and the parameter values for carbazole, respectively, for DMSO concentrations ranging only from 30 to 50% v/v. The data at 60% v/v DMSO concentration were not evaluated because of the insensitivity to the extraction of carbazole. In this cyclohexane system, an appreciably poorer correlation for the linear plots of D versus [CD]a0 is revealed by Fig. 3 and/or the correlation coefficients in Table 3. This poor correlation is due to the nature of the carbazole extraction, that is, a slight increase in the amount of carbazole with a large increase in the amount of CD. There may be some uncertainty in this system regarding the determination of the values of both the slope and the related Kf. However, the results for the cyclic hydrocarbon should be useful for comparison with those for the above straight-chain hydrocarbons.Table 3 indicates that the apparent Kf values at 30 and 40% v/v DMSO concentrations can be clearly distinguished from those in the heptane and dodecane systems, in spite of the similar decreases in Kd values. Anigbogu et al.3 pointed out that differences in the experimental conditions and probing techniques could result in significant differences in Kf. In this extraction technique, however, large differences in Kf values were observed under the same conditions and using the same method, although there is a difference in type of hydrocarbon.This implies that the hydrocarbons used as the organic phase compete with carbazole for CD complexation. It may therefore be concluded that the estimated Kf for the carbazole–CD Table 1 Parameter values obtained from data on solvent extraction of carbazole from heptane into various aqueous DMSO solutions of CD Concentration of DMSO (% v/v) Value 10 20 30 40 50 60 Mean Kd for 8 measurements 0.0216 0.0651 0.180 0.490 1.50 4.75 RSD of Kd measurements (%) 21.1 7.7 4.5 4.5 3.1 2.7 Slope of D versus [CD]a0 plot*/dm3 mol21 10.4 14.7 19.7 27.4 38.4 61.6 Correlation coefficient for the plot* 0.9600 0.9687 0.9808 0.9536 0.9311 0.7658 Kf (slope/Kd)/dm3 mol21 480 226 109 55.9 25.6 13.0 * Linear regression analysis was applied to 40 data points in the [CD]a0 range 0–6 mol m23 in the plot. 906 Analyst, September 1997, Vol. 122complex must be considered as a relative value to that for the hydrocarbon–CD complex.Linear Relationship Between log Kf and log Kd The apparent formation constant is a measure of the complexation of carbazole with CD in a given water–DMSO medium. On the other hand, the distribution coefficient is a measure of the distribution of carbazole between a hydrocarbon solvent and the medium in the absence of CD. Therefore, these two equilibrium constants, Kf and Kd, should be essentially independent of each other even if the hydrocarbon has a competitive effect on the carbazole–CD complexation.Application of linear regression analysis to each of the six data from the heptane and dodecane extraction systems led to the equations log Kf = 20.675 log Kd + 1.547 (r = 20.9996) and log Kf = 20.672 log Kd + 1.511 (r = 20.9992), respectively. Because of the small number of data points (n = 3), regression analysis was not applied to the results from the cyclohexane system.As can be seen from the above equations, Fig. 2 Plot of the distribution ratio for carbazole versus the initial concentration of CD observed in a liquid–liquid extraction system consisting of dodecane and aqueous DMSO (10–60% v/v). Table 2 Parameter values obtained from data on solvent extraction of carbazole from dodecane into various aqueous DMSO solutions of CD Concentration of DMSO (% v/v) Value 10 20 30 40 50 60 Mean Kd for 8 measurements 0.0201 0.0593 0.157 0.412 1.27 4.19 RSD of Kd measurements (%) 16.4 13.6 2.9 4.4 2.6 2.7 Slope of D versus [CD]a0 plot*/dm3 mol21 9.06 12.9 17.1 23.6 38.6 49.5 Correlation coefficient of the plot* 0.9744 0.9635 0.9581 0.9535 0.90006 0.6395 Kf (slope/Kd)/dm3 mol21 451 217 109 57.2 30.5 11.8 * Linear rergression analysis was applied to 40 data points in the [CD]a0 range 0–6 mol m23 in the plot.Analyst, September 1997, Vol. 122 907the plot of log Kf versus log Kd gives straight lines with good correlations (0.9996 and 0.9992) and similar slopes (20.675 and 20.672).These results suggest that the slope of the log Kf versus log Kd plot remains unaltered with changes to the extraction system. Fig. 4 illustrates log–log plots of the two equilibrium constants in all three hydrocarbon–aqueous DMSO extraction systems. The slopes of the three lines are identical, being the mean of the two values obtained from the heptane and dodecane systems.It is found that the above assumption is satisfactorily applicable to the case of the cyclohexane system. From these results, therefore, we conclude that there is a linear correlation between log Kf and log Kd for carbazole evaluated by the extraction method. In addition, the slope of the linear plot should be constant in different liquid–liquid extraction systems consisting of hydrocarbons and aqueous DMSO solutions. The reason for the unchanged slope is not clear, but it might arise from an unchanged fitting of the carbazole molecule into the CD cavity in different solvent environments.Interpretation of Empirical Equation On the above supposition, the correlation between Kf and Kd for carbazole in the hydrocarbon–aqueous DMSO systems can be expressed as log Kf = 20.673 log Kd + A (6) where A denotes the intercept of the linear plot and varies according to the hydrocarbon used as the organic phase in extraction. It is found from eqn. (6) that the value of Kf depends only on A when log Kd = 0 (Kd = 1).In addition, the apparent Kf value for carbazole can, in fact, be considered as a relative value resulting in competition with the hydrocarbon solvent. In order to generalize eqn. (6), we shall assume from the above viewpoint that Kf (observed) = Kf (carbazole)/Kf (hydrocarbon). On the basis of this additional assumption, therefore, eqn. (6) can be replaced by log KfOB = 20.673 log Kd + log (KfCR/KfHY) (7) where KfOB is the observed Kf value for a 1 : 1 carbazole–CD complex in a given water–DMSO medium and KfCR and KfHY are the absolute Kf values in neat water for the carbazole– and hydrocarbon–CD complexes, respectively.Eqn. (7) is a plausible empirical equation, but it is difficult to verify the validity at present because of the lack of both KfCR and KfHY values measured at the same time. Fortunately, however, some apparent KfHY values for heptane and cyclohexane have been published and could serve for the inspection of eqn.(7). When eqn. (6) is applied to the two different hydrocarbon solvents, heptane and cyclohexane, the intercepts of the corresponding linear plots are expressed according to eqn. (7) as AHP = log (KfCRKfHP) and ACY = log (KfCR/KfCY), where the subscripts HP and CY represent heptane and cyclohexane, respectively. Consequently, the difference between AHP and ACY can be written as follows: AHP2ACY = log (KfCR/KfHP)2log (KfCR/KfCY) = log (KfCR KfCY/KfHP KfCR) = log (KfCY/KfHP) (8) In this work, the statistically established values of A for heptane, dodecane and cyclohexane were 1.548, 1.510 and 1.238, respectively.Substitution of the two values of interest into eqn. (8) resulted in log (KfCY/KfHP) = 0.310, that is, KfCY/ KfHP = 2.04. Sanemasa and co-workers27,28 studied the KfHY values in an aqueous medium at 25 °C by making use of the volatilization rate of hydrocarbon molecules. The KfHP and KfCY values that they determined were 69 ± 4 and 156 ± 8 dm3 mol21, respectively.Therefore, a simple calculation from their data gives a KfCY/KfHP ratio of 2.26 (with variations in the range 2.03–2.52). On the other hand, Wishnia and Lappi29 presented the dissociation constant; which is the reciprocal of the formation constant (1/Kf) for heptane– and cyclohexane–CD 1 : 1 complexes in the range 0–50 °C. The values given in their table for the heptane and cyclohexane complexes were 0.37 and 0.182 mol m23, respectively, at 20 °C and 0.39 and 0.199 mol m23, respectively, at 30 °C.Further application of their results indicated that KfCY/KfHP = 2.03 at 20 °C and 1.96 at 30 °C. These values of the KfCY/KfHP ratio are in fair agreement with the value calculated in our work. Therefore, these results suggest that there is a reasonable correlation between KfOB and Kd values for carbazole, and also that the correlation can be expressed in the form of the empirical eqn. (7).Fig. 3 Plot of the distribution ratio for carbazole versus the initial concentration of CD observed in a liquid–liquid extraction system consisting of cyclohexane and aqueous DMSO (30–50% v/v). 908 Analyst, September 1997, Vol. 122Conclusion Liquid–liquid extraction of carbazole was performed from three types of hydrocarbon solvents into water–DMSO solutions of CD in order to determine the values of the apparent formation constant and the distribution coefficient at the same time.A linear relationship between the log Kf and log Kd for carbazole was substantiated on the basis of the extraction data. Since the slope of the log Kf versus log Kd plot obtained with the dodecane system was similar to that with the heptane system, it was assumed that the slope does not vary with changes in hydrocarbon phase. The reason for the unchanged slope is not clear at the present stage of our limited study. Further and more extensive studies may be necessary in order to confirm the assumption.From the results with the cyclohexane system, on the other hand, it was concluded that the apparent Kf value observed for a 1 : 1 carbazole–CD complex is a relative value resulting in competition with the hydrocarbon solvent. An empirical equation was finally proposed as a reasonable correlation between the observed Kf and Kd values for carbazole in these hydrocarbon–aqueous DMSO extraction systems. Emphasis was placed on the interpretation of the intercept of the linear log Kf versus log Kd plot.Although it is unknown what the slope of the plot represents, the concept of the relative Kf value as the intercept was pertinent to the empirical equation. Also, it is not clear whether such a correlation can be found or not for species other than carbazole. However, the good linearity of the log Kf versus log Kd plot may be attractive in further analytical investigations, such as for the prediction of selective separations from complicated mixtures and application of CD-containing solvent extraction to practical samples.References 1 Mu�noz de la Pe�na, Adou, T., Zung, J. B., and Warner, I. M., J. Phys. Chem., 1991, 95, 3330. 2 Chokchainarong, S., Fennema, O. R., and Connors, K. A., Carbohydr. Res., 1992, 232, 161. 3 Anigbogu, V. C., Mu�noz de la Pe�na, A., Ndou, T. T., and Warner, I. M., Anal. Chem., 1992, 64, 484. 4 Loukas, Y. L., Antoniadou-Vyza, E., Papadaki-Valiraki, A., and Machera, K. G., J. Agric. Food Chem., 1994, 42, 944. 5 Gareil, P., Pernin, D., Gramond, J.-P., and Guyon, F., J. High Resolut. Chromatogr., 1993, 16, 195. 6 Baumy, Ph., Morin, Ph., Dreux, M., Viaud, M. C., Boye, S., and Guillaumet, G., J. Chromatogr. A, 1995, 707, 311. 7 Choi, H.-S., Chang, C.-J., and Knevel, A. M., Pharm. Res., 1992, 9, 582. 8 Georgiou, M. E., Georgiou, C. A., and Koupparis, M. A., Anal. Chem., 1995, 67, 114. 9 Li, S., and Purdy, W. C., Anal. Chem., 1992, 64, 1405. 10 Valsami, G. N., Koupparis, M. A., and Macheras, P. E., Pharm. Res., 1992, 9, 94. 11 McCormack, S., Russell, N. R., and Cassidy, J. F., Electrochim. Acta, 1992, 37, 1939. 12 Blyshak, L. A., Rossi, T. M., Patonay, G., and Warner, I. M., Anal. Chem., 1988, 60, 2127. 13 Harangi, J., and N�an�asi, P., Anal. Chim. Acta, 1984, 156, 103. 14 Matsunaga, K., Imanaka, M., Ishida, T., and Oda, T., Anal. Chem., 1984, 56, 1980. 15 Elliott, N. B., Prenni, A. J., Ndou, T. T., and Warner, I. M., J. Colloid Interface Sci., 1993, 156, 359. 16 Tachibana, M., and Furusawa, M., Analyst, 1995, 120, 437. 17 Tachibana, M., Furusawa, M., and Kiba, N., J. Inclusion Phenom. Mol. Recognit. Chem., 1995, 22, 313. 18 Matsui, Y., and Mochida, K., Bull. Chem. Soc. Jpn., 1979, 52, 2808. 19 Uekama, K., Hirayama, F., Nasu, S., Matsuo, N., and Irie, T., Chem. Pharm. Bull., 1978, 26, 3477. 20 Furusawa, M., Takeuchi, T., Sekido, K., and Shimizu, H., Kogyo Kagaku Zasshi, 1963, 66, 1811. 21 Tachibana, M., and Furusawa, M., Bull. Chem. Soc. Jpn., 1988, 61, 2353. 22 Mohseni, R. M., and Hurtubise, R. J., J. Chromatogr., 1990, 499, 395. 23 Mu�noz de la Pe�na, A., Ndou, T. T., Anigbogu, V. C., and Warner, I. M., Anal. Chem., 1991, 63, 1018. 24 Kano, K., Hashimoto, S., Imai, A., and Ogawa, T., J. Inclusion Phenom., 1984, 2, 737. 25 Mu�noz de la Pe�na, A., Ndou, T. T., Zung, J. B., Greene, K. L., Live, D. H., and Warner, I. M., J. Am. Chem. Soc., 1991, 113, 1572. 26 Sanemasa, I., Koga, I., and Deguchi, T., Anal. Sci., 1991, 7, 641. 27 Sanemasa, I., Osajima, T., and Deguchi, T., Bull. Chem. Soc. Jpn., 1990, 63, 2814. 28 Osajima, T., Deguchi, T., and Sanemasa, I., Bull. Chem. Soc. Jpn., 1991, 64, 2705. 29 Wishnia, A., and Lappi, S. J., J. Mol. Biol., 1974, 82, 77. Paper 7/01843F Received March 17, 1997 Accepted May 29, 1997 Table 3 Parameter values obtained from data on solvent extraction of carbazole from cyclohexane into various aqueous DMSO solutions of CD Concentration of DMSO (% v/v) Value 30 40 50 Mean Kd for 8 measurements 0.126 0.365 1.07 RSD of Kd measurements (%) 2.9 4.4 2.6 Slope of D versus [CD]a0 plot*/ dm3 mol21 8.69 9.99 22.3 Correlation coefficient for the plot* 0.8794 0.7755 0.6656 Kf (slope/Kd)/dm3 mol21 68.9 27.4 20.9 * Linear regression analysis was applied to 40 data points in the [CD]a0 range 0–6 mol m23 in the plot. Fig. 4 Correlation between the apparent Kf value for a 1 : 1 carbazole–CD complex present in water–DMSO medium and the Kd value for carbazole on extraction from (2) heptane, (8) dodecane and (Ç) cyclohexane into the medium. Analyst, September 19
ISSN:0003-2654
DOI:10.1039/a701843f
出版商:RSC
年代:1997
数据来源: RSC
|
9. |
Sequential Injection Analysis Technique for the Concentration,Stoichiometry and Formation Constant Studies of Promethazine HydrochlorideComplexed With Palladium(II) in Hydrochloric Acid |
|
Analyst,
Volume 122,
Issue 9,
1997,
Page 911-914
Salah M. Sultan,
Preview
|
|
摘要:
Sequential Injection Analysis Technique for the Concentration, Stoichiometry and Formation Constant Studies of Promethazine Hydrochloride Complexed With Palladium(II) in Hydrochloric Acid Salah M. Sultan* and Nabeel I. Desai The Chemistry Department, King Fahd University of Petroleum and Minerals, KFUPM, P.O. Box 2026, Dhahran 31261, Saudi Arabia The sequential injection analysis (SIA) technique was successfully applied to the determination of promethazine hydrochloride in drug formulations.The chemical system is based on the complexation reaction of promethazine hydrochloride with PdII in 8.0 3 1024 mol dm23 HCl and measurement of the absorbance at 504 nm. Promethazine was determined in the range 50–400 ppm using 1.0 3 1023 mol dm23 PdII with an aspiration volume of 147.5 ml. The SIA technique was utilized for the determination of the concentration, stoichiometry and formation constant of the complexation reaction. The technique was found to be superior to flow injection analysis.The SIA method was statistically compared with the official British Pharmacopoeia method and showed comparable accuracy, but with the advantages of selectivity, simplicity, speed and amounts of reagents consumed. Keywords: Sequential injection analysis; promethazine–palladium(II) complexation; stoichiometry Promethazine hydrochloride, dimethyl[1-methyl-2-(phenothiazine- 1-yl)ethyl]amine hydrochloride, is widely used for its antihistamine action and for various post-operative conditions.The available analytical methods have been reviewed,1 and include two examples using flow injection analysis (FIA) that are relevant to this paper.1,2 Both methods depend on the spectrophotometric determination of a red oxidized derivative of the drug, believed to be a dication radical, generated, respectively, by metavanadate1 and cerium(iv).2 The British Pharmacopoeia (BP) method3 for promethazine uses PdII to generate a purple derivative, suitable for spectrophotometry.Similar reactions between PdII and other phenothiazines have been investigated4–6 and used as the basis of a sequential injection analysis (SIA) method.7 The present work describes an SIA method for promethazine based on its reaction with PdII. Experimental Reagents and Stock Solutions PdII. A stock solution of 0.025 mol dm23 PdII in 0.02 mol dm23 hydrochloric acid was prepared by dissolving 0.4440 g of anhydrous PdCl2 (Fluka, Buchs, Switzerland) in 50 cm3 of 0.04 mol dm23 hydrochloric acid.The mixture was heated at 80 °C until all the solid had dissolved, then cooled to room temperature and made up to 100 cm3 with water to form a clear brownish yellow solution. A stock solution containing a hydrochloric acid concentration lower than 0.02 mol dm23 could not be prepared as PdII does not dissolve in less acidic solutions even after prolonged heating. The molar absorptivity of the absorbing species is such that at the optimum pH of approximately 4 the optimum PdII concentration range over which Beer’s law applied is 1023–1024 mol dm23.Hydrochloric acid. A stock solution of 10 mol dm23 HCl was prepared by diluting AnalaR concentrated acid (Merck, Poole, Dorset, UK). Promethazine. The stock solution was 1500 ppm (4.674 3 1023 mol dm23) C17H20N2S.HCl (Rhone-Poulenc, Dagenham, Essex, UK, Batch No. W3021), prepared by dissolving 0.15 g of the drug in 100 cm3 of de-ionized water. Pure drugs.The drugs studied together with the names of the suppliers and other relevant information are given in Table 1.Table 1 Statistical comparison of the results for the determination of promethazine hydrochloride in commercial formulations by the proposed method with those obtained by the official BP method3 Recovery ± s (%)* Active Formulation ingredient SIA BP t-value† Phenergen tablets Promethazine 99.72 ± 0.6 99.31 ± 0.5 1.53 (Specia, France) hydrochloride, 25 mg Phenergen syrup Promethazine 99.9 ± 0.7 99.8 ± 0.3 0.32 (Specia, France) hydrochloride, 666.67 ppm Phenergen syrup Promethazine 99.4 ± 1.0 99.9 ± 0.6 1.12 Expectorant hydrochloride, (Specia, France) 666.67 ppm Cigan Elixir Promethazine 100.1 ± 0.8 99.8 ± 0.4 0.84 (Cimabrex, Denmark) hydrochloride, 1000 ppm * Standard deviation (s) for five determinations based on label claim.† Theoretical t-value = 2.78 (p = 0.05). Analyst, September 1997, Vol. 122 (911–914) 911Stock solutions of 1000 ppm of each of the drugs were prepared by dissolving the drug in water or dilute acid solution at room temperature.Syrups and elixir. Test solutions were prepared by diluting a pipetted amount of the syrup or elixir directly with water or dilute acid solution in a calibrated flask. Working solutions for calibration, Job’s plot, molar ratio and all other studies were prepared by appropriate dilutions of the stock solutions. Apparatus Sequential injection analyser The sequential injection analyser (Fig. 1) was constructed from the following components: A peristaltic pump (C4V, Alitea, Medina, WA, USA) featuring eight stainless-steel rollers on individual bearings was employed to propel the solutions. A Valco 10-port selector valve (Cheminert, Valco Instruments, Houston, TX, USA) was used to select the flows. Upchurch fittings (Upchurch, Oak Harbor, WA, USA) were used to lock unused ports. The holding and reaction coil tubing and also the tubing connecting the different units was made of PTFE (0.8 mm id).Teflon nuts and ferrules (Upchurch) were used to assemble the manifold. The pump tubing was Phar Med 1.02 mm id tubing (Upchurch), and was held on the pump rollers by FIA peristaltic pump tubing adapters (Upchurch). A reactor module consisting of 0.5 mm id PTFE tubing (Thermoplastic Scientific, NY, USA) of different lengths was used for mixing the solutions. Absorbance measurements were made with a Spectronic Mini-20 spectrophotometer (Milton Roy, Rochester, NY, USA), equipped with a grating monochromator detector and a Unovic ultra-micro-flow-through cell (Unovic Instruments, NY, USA) (20 ml) with a pathlength of 1.0 mm.A personal computer (Austin Computer Systems, Austin, TX, USA) equipped with a 120 Mbyte hard disk, 4 Mbyte RAM and VGA graphics was used to monitor the pump and valve. Communication between the computer and the external devices was effected with a general-purpose I/O board (Model ADA- 110, Real Time Devices, State College, PA, USA).The computer was also used to collect the data; alternatively, the data were recorded by a Model 0555 single-channel strip-chart recorder (Cole-Parmer, Chicago, IL, USA). A Perkin-Elmer (Norwalk, CT, USA) Lambda 5 UV/visible spectrophotometer equipped with 10.0 mm cells was used for preliminary investigations. Software Packages Microsoft windows 3.10 with DOS 6.20 was utilized to run the following software: FIA Lab 2.0 (beta release) (Alitea) was used to program the SIA system.Sigmaplot, version 1.02 (Jandel Scientific, Erkrath, Germany) was employed for data handling calculations and constructing graphs. Microsoft Word, version 6.0 (Microsoft Corporation, USA) was used for constructing tables and writing text. Procedure Fig. 1 shows the SIA manifold used. Fixed amounts of reagent, sample and carrier (wash) solutions were sequentially aspirated into the lines from the ports by means of the selector valve (SV). In each case, the excess solution introduced into the holding coil (HC) was expelled through port 7 to the auxiliary waste.The remaining steps of the procedure, which were controlled by the computer, were as follows: 1, The carrier solution was pumped through port 1, by setting the pump in the forward direction, for 25–45 in order to flush the system (holding coil, reactor and detector) with the carrier solution. 2, A pre-determined volume of the reagent solution was aspirated through port 2 into the holding coil (HC) in the reverse direction. The remaining reagents were aspirated into the holding coil via ports 3–8, as appropriate. 3, The drug solution was aspirated into the holding coil from port 9. The last two steps were achieved by setting the pump in the reverse mode. Finally, the composite zone was propelled by the carrier solution through port 1 to the reaction coil and then the detector. When the mixing chamber was used instead of the reactor coil, the reagent(s) and the sample were initially injected into the mixing chamber.At fixed intervals, aliquots of the solution in the mixing chamber were withdrawn into the holding coil and then pushed to the detector using the carrier by forward pumping. The data were acquired by the computer and transferred to plotting software for further calculations. In each of the above steps the volume of the solutions aspirated was determined from the time of aspiration and the volumetric flow rate of the pump (using the relationship, volume = flow rate 3 time), which was 29.5 ml s21 in this study.Determination of Flow Rate The pump speed can be altered by changing the rev min21 setting on the peristaltic pump. The rate of flow of the reagent inside the pump tubes was determined by withdrawing liquid (distilled water) at one end of the tube and collecting it at the other end in a measuring cylinder. The volume of liquid collected at a certain time gives the flow rate in ml min21 at a particular rev min21 value.This process was repeated for different rev min21 values, giving different values of flow rate. A graph of flow rate versus pump speed (rev min21) was then plotted. A straight line described by the following equation was obtained: Flow rate = 0.73644 + 0.057263(pump speed/rev min21) Therefore, for a pump speed of 500 rev min21, which was used in this work, the flow rate will be 1.77 ml min21 or 29.5 ml s21.Stoichiometry For stoichiometric studies, the SIA system (Fig. 1), including all the tubes attached to the selector valve, was flushed with de- Fig. 1 SIA manifold: C, carrier; P, peristaltic pump; HC, holding coil; SV; selector valve; RC, reaction coil; D, detector; CP, computer; and W, waste. 912 Analyst, September 1997, Vol. 122ionized water (carrier) flowing at a rate of 29.5 ml s21. The following operations were then conducted: 1. Equimolar solutions of PdII and promethazine were connected to the selector valve through ports 2 and 3, respectively. About 80 ml of each reagent were introduced sequentially into the holding coil in the reverse mode for 3.0 s; the excess, together with some carrier, was transferred to the auxiliary waste through valve 7 in the forward mode for 8.0 s. 2. In Job’s method of continuous variation, different aliquots of equimolar solutions of PdII and promethazine were taken and mixed in the holding coil so as to give solutions of identical total concentration (PdII + drug) but different mole fractions; the solution was then pushed for 40.0 s to the detector for signal monitoring.The volume of each reagent aspirated was varied between 14.8 ml (0.5 s) and 147.5 ml (5.0 s). The total volume aspirated was maintained constant at 162.0 ml by adjusting the aspiration times. This step was repeated while varying the aspiration time between 14.8 ml (0.5 s) and 280.3 ml (9.5 s). 3. In the molar ratio method, the total concentration of the ligand (promethazine) was maintained constant by aspirating 147.5 ml (5.0 s) of solution into the holding coil by flow reversal, while the PdII solution volume was varied between 14.8 ml (0.5 s) and 192.0 ml (6.5 s). This step was repeated except that the total concentration of the metal was held constant by fixing the flow reversal for 147.5 ml (5.0 s), while the volume of drug solution was varied between 14.8 ml (0.5 s) and 192.0 ml (6.5 s).Ideally, in the molar ratio method, two straight lines with different slopes are obtained when the absorbance is plotted against the PdII-to-drug ratio, and the point of intersection of these two lines corresponds to the stoichiometric ratio on interpolation to the molar ratio axis. Steps 2 and 3 were repeated but using 8.0 3 1024 mol dm23 hydrochloric acid as the carrier instead of water. There was no significant difference in the absorbance values obtained using either acid or water as carrier. Results and Discussion Chemical System Prior to using the SIA technique, preliminary investigations were made in which the maximum absorbance (lmax) over a wide range of pH values from 8.0 31025 to 0.01 mol dm23 was determined.The optimum absorbance conditions as a function of pH were established by measurements on solutions of PdII (10 mol dm23) and promethazine (150 ppm) at 28 increments of pH between 2.0 and 4.1. Both the wavelength of maximum absorbance and the absorbance itself increased from pH 2 (496.2 nm and 0.303, respectively) to pH 3.12 (503.9 nm and 0.369) before decreasing at pH 4.1 (502.2 nm and 0.279). The optimum pH for this work was selected as 3.1.The stability of the complexes formed favoured this method over previous methods involving oxidation with stronger oxidizing agents than PdII in which a highly unstable radical product is monitored for quantification.1,2,8,9 The nature of the interaction between PdII and promethazine in the pH range 4.1–3.0 is believed to involve protonation of the drug followed by hydrogen bonding between the protonated nitrogen and one of the chlorine atoms of PdCl2.4–6,10 Complexes of the type ML2X2 and MLX3 have also been reported,10 where M = Pd, L = promethazine ligand and X = Cl.An investigation of the type, stoichiometry and formation constant of such a complex was carried out by the Job’s plot11 and molar ratio11,12 methods as described below.Job’s Plot Method In Job’s method of continuous variation, different aliquots of equimolar solutions of PdII and promethazine were mixed to give solutions of identical total concentration (PdII + drug) but different mole fractions. A volume of carrier solution was then aspirated; the carrier volume was selected to allow optimum mixing on flow reversal towards the detector. Analysis of the Job’s plot was used to determine the stoichiometry of the complexation of PdII with promethazine in hydrochloric acid medium.A typical Job’s plot, illustrated in Fig. 2, was obtained by the SIA procedure using 8 3 1024 mol dm23 hydrochloric acid and an ionic strength of 0.20 mol dm23, which was maintained by using lithium perchlorate. Water (de-ionized) was used as a carrier for analysis, but the experiment was also repeated several times with hydrochloric acid as carrier and also with mixing of the reactants in the holding coil. It was found that there was no significant difference when water or acid was used as a carrier, even if the mixing steps were included in the SIA program.The total volume of the two reagents for each run was kept constant at 295.1 and 162.3 ml. It is worth noting that both reagents, viz., PdII and promethazine, were injected, thus consuming minimal amounts of reagent compared with the manual procedure. It is clear that the curves A = f (Xdrug) exhibit a maximum for mole fractions of 0.5, indicating that the ratio of PdII : drug in the complex is 1 : 1.Only 10 min are needed to generate the Job’s plot using SIA and less than 10 ml of the drug is sufficient to repeat the procedure many times. Attempts to apply the molar ratio method12,13 failed because the PdII concentration was a greater effect on the system, resulting in the formation of a mixed complex together with the oxidized form of the drug at higher concentrations of PdII, which makes the method unsatisfactory.This is in agreement with previous work, which indicated that various complex forms are possible.7 Overall, the Job’s plot method is the more reliable and it has clearly shown that the molar ratio of the PdII–promethazine complex is 1 : 1. Fig. 2 Typical SIA trace representing a Job’s plot for the promethazine system: [PdII] = [promethazine] = 1 3 1023 mol dm23 in 8 3 1024 mol dm23 hydrochloric acid; ionic strength = 0.20 mol dm23. The total aspiration volume is equivalent to 295 ml and aspiration volumes were varied between 14.8 ml and 280.3 ml.Mole fraction of the drug: (1) 0; (2) 0.05; (3) 0.10; (4) 0.20; (5) 0.30; (6) 0.40; (7) 0.50; (8) 0.60; (9) 0.70; (10) 0.80; (11) 0.90; (12) 0.95; (13) 0.975; and (14) 0.10 ml. Analyst, September 1997, Vol. 122 913Formation Constant The formation constant and the composition of the PdII– promethazine complex were also investigated by numerical methods.14,15 The JOBCON16,17 program was used to analyse the continuous variation data.The program was modified by the authors and re-written in C-language and applied on a PC/AT computer. The calculations are based on fitting a function f(x, b) to a set of experimental data, using a least-squares method. Unknown parameters are estimated by minimizing U, the sum of squares of residuals, defined by the following equation: U A A i n = - = å ( ) exp calc 2 1 where n represents the number of experimental points, Aexp the experimental absorbance, and Acalc = f (x, b), the absorbance calculated by the program from formation constants and stoichiometric ratios.Therefore, various experimental models could be fitted to the experimental data iteratively by varying the values of the formation constant and stoichiometric ratio. The JOBCON program was used to calculate the formation constant from the continuous variation data for the PdII– promethazine complex. A 1 : 1 metal-to-ligand ratio was found to be the most probable, with a relative error of 14.99%, resulting in a value for the logarithm of the formation constant (log Kf) of 4.349.Other metal-to-ligand ratios (m: n) appeared to be improbable, giving high relative errors, particularly those where m: n > 2. The value of log Kf reported10 previously for promethazine was 5.52, which is close to the value obtained in the present work for the 1 : 1 PdII–promethazine complex. It was also observed that the stoichiometry and the value of the formation constant are independent of the total concentration of the metal ion and the ligand. Calibration Graph A series of standard solutions of promethazine hydrochloride were run in triplicate; a graph of absorbance versus concentration was found to be linear in the range 50–400 ppm.The calibration equation was: A = 0.008654 + 0.000110 C where A is absorbance and C is concentration in ppm; the correlation coefficient (r2) was 0.997. The peak width at the baseline was measured to be 18 s, thus giving a throughput of 200 samples h21.The peak width was also determined at 60% peak height and found to be 2.4 s, indicating minimal dispersion. An average relative standard deviation (RSD) of 0.92% was obtained for six replicate determinations of 250 ppm of promethazine, indicating excellent reproducibility. Application The method was applied to the determination of promethazine hydrochloride in several commercial formulations, namely, Phenergen tablets, Phenergen promethazine, Phenergen Expectorant and Cigan Elixir. All drug formulations contained starch and glucose as excipients.The same formulations were also analysed by the BP method3 and the results were statistically compared by calculating the percentage recoveries, standard deviations and the Student’s t-test values. The results are shown in Table 1. The accuracy of the proposed method was comparable to that of the BP method. The results also show that common excipients do not interfere with the determination.The proposed method is simpler and more selective than the BP and previous methods. The utilization of PdII avoided oxidation of the drug during the analysis. Conclusion The proposed SIA method for the determination of promethazine was validated by complexation of the drug with PdII in dilute hydrochloric acid medium and was found to be superior to earlier FIA methods. The method developed was applied to the determination of promethazine in commercial formulations and was found to be superior to the official BP method, with a wide dynamic range (50–400 ppm), excellent reproducibility and an RSD of less than 0.96%.The capacity of the SIA technique was demonstrated by its application to the determination of concentrations, stoichiometries and formation constants of complexation reactions. A large number of experimental trials were performed utilizing relatively small amounts of reagents.N.I.D. thanks KFUPM for the award of a scholarship to study for an M.S. degree. References 1 Sultan, S. M., Analyst, 1991, 116, 177. 2 Sultan, S. M., and Suliman, F. O., Anal. Sci., 1992, 8, 841. 3 British Pharmacopoeia, HM Stationery Office, London, 5th edn., 1988, pp. 749 and 995. 4 Pellizzetti, E., J. Chem. Soc., Dalton Trans., 1980, 484. 5 Geary, W. J., Mason, N. J., Nowell, I. W., and Nixon, L. A., J. Chem. Soc., Chem. Commun., 1984, 1064. 6 Duddell, D. A., Goggin, P.L., Goodfellow, R. J., Norton, M. G., and Smith, J. G., J. Chem. Soc. A, 1970, 546. 7 Sultan, S. M., Suliman, F. O., and Bahruddin, B. S., Analyst, 1995, 120, 561. 8 Sultan, S. M., Microchem. J., 1991, 44, 304. 9 Sultan, S. M., and Abdennabi, A. M., Microchem. J., 1993, 48, 343. 10 Geary, W. J., Mason, N. J., Nowell, I. W., and Nixon, L. A., J. Chem. Soc., Dalton Trans., 1982, 1103. 11 Job, P., Ann. Chim., 1928, 9, 113. 12 Momoki, K., Sekino, J., Sato, H., and Yamaguchi, N., Anal.Chem., 1969, 41, 1286. 13 Yoe, J. H., and Jones, A. L., Ind. Eng. Chem. Anal. Ed., 1944, 64, 111. 14 Sabatini, A., Vacca, A., and Gans, P., Coord. Chem. Rev., 1992, 120, 289. 15 Sultan, S. M., Anal. Lett., 1991, 24, 1785. 16 Likussar, W., Anal. Chem., 1973, 45, 1926. 17 Meloun, M., and Javurek, M., Talanta, 1984, 31, 1083. Paper 7/01810J Received March 14, 1997 Accepted May 7, 1997 914 Analyst, September 1997, Vol. 122 Sequential Injection Analysis Technique for the Concentration, Stoichiometry and Formation Constant Studies of Promethazine Hydrochloride Complexed With Palladium(II) in Hydrochloric Acid Salah M.Sultan* and Nabeel I. Desai The Chemistry Department, King Fahd University of Petroleum and Minerals, KFUPM, P.O. Box 2026, Dhahran 31261, Saudi Arabia The sequential injection analysis (SIA) technique was successfully applied to the determination of promethazine hydrochloride in drug formulations. The chemical system is based on the complexation reaction of promethazine hydrochloride with PdII in 8.0 3 1024 mol dm23 HCl and measurement of the absorbance at 504 nm.Promethazine was determined in the range 50–400 ppm using 1.0 3 1023 mol dm23 PdII with an aspiration volume of 147.5 ml. The SIA technique was utilized for the determination of the concentration, stoichiometry and formation constant of the complexation reaction. The technique was found to be superior to flow injection analysis.The SIA method was statistically compared with the official British Pharmacopoeia method and showed comparable accuracy, but with the advantages of selectivity, simplicity, speed and amounts of reagents consumed. Keywords: Sequential injection analysis; promethazine–palladium(II) complexation; stoichiometry Promethazine hydrochloride, dimethyl[1-methyl-2-(phenothiazine- 1-yl)ethyl]amine hydrochloride, is widely used for its antihistamine action and for various post-operative conditions.The available analytical methods have been reviewed,1 and include two examples using flow injection analysis (FIA) that are relevant to this paper.1,2 Both methods depend on the spectrophotometric determination of a red oxidized derivative of the drug, believed to be a dication radical, generated, respectively, by metavanadate1 and cerium(iv).2 The British Pharmacopoeia (BP) method3 for promethazine uses PdII to generate a purple derivative, suitable for spectrophotometry.Similar reactions between PdII and other phenothiazines have been investigated4–6 and used as the basis of a sequential injection analysis (SIA) method.7 The present work describes an SIA method for promethazine based on its reaction with PdII. Experimental Reagents and Stock Solutions PdII. A stock solution of 0.025 mol dm23 PdII in 0.02 mol dm23 hydrochloric acid was prepared by dissolving 0.4440 g of anhydrous PdCl2 (Fluka, Buchs, Switzerland) in 50 cm3 of 0.04 mol dm23 hydrochloric acid. The mixture was heated at 80 °C until all the solid had dissolved, then cooled to room temperature and made up to 100 cm3 with water to form a clear brownish yellow solution.A stock solution containing a hydrochloric acid concentration lower than 0.02 mol dm23 could not be prepared as PdII does not dissolve in less acidic solutions even after prolonged heating. The molar absorptivity of the absorbing species is such that at the optimum pH of approximately 4 the optimum PdII concentration range over which Beer’s law applied is 1023–1024 mol dm23.Hydrochloric acid. A stock solution of 10 mol dm23 HCl was prepared by diluting AnalaR concentrated acid (Merck, Poole, Dorset, UK). Promethazine. The stock solution was 1500 ppm (4.674 3 1023 mol dm23) C17H20N2S.HCl (Rhone-Poulenc, Dagenham, Essex, UK, Batch No. W3021), prepared by dissolving 0.15 g of the drug in 100 cm3 of de-ionized water. Pure drugs.The drugs studied together with the names of the suppliers and other relevant information are given in Table 1.Table 1 Statistical comparison of the results for the determination of promethazine hydrochloride in commercial formulations by the proposed method with those obtained by the official BP method3 Recovery ± s (%)* Active Formulation ingredient SIA BP t-value† Phenergen tablets Promethazine 99.72 ± 0.6 99.31 ± 0.5 1.53 (Specia, France) hydrochloride, 25 mg Phenergen syrup Promethazine 99.9 ± 0.7 99.8 ± 0.3 0.32 (Specia, France) hydrochloride, 666.67 ppm Phenergen syrup Promethazine 99.4 ± 1.0 99.9 ± 0.6 1.12 Expectorant hydrochloride, (Specia, France) 666.67 ppm Cigan Elixir Promethazine 100.1 ± 0.8 99.8 ± 0.4 0.84 (Cimabrex, Denmark) hydrochloride, 1000 ppm * Standard deviation (s) for five determinations based on label claim.† Theoretical t-value = 2.78 (p = 0.05). Analyst, September 1997, Vol. 122 (911–914) 911Stock solutions of 1000 ppm of each of the drugs were prepared by dissolving the drug in water or dilute acid solution at room temperature.Syrups and elixir. Test solutions were prepared by diluting a pipetted amount of the syrup or elixir directly with water or dilute acid solution in a calibrated flask. Working solutions for calibration, Job’s plot, molar ratio and all other studies were prepared by appropriate dilutions of the stock solutions. Apparatus Sequential injection analyser The sequential injection analyser (Fig. 1) was constructed from the following components: A peristaltic pump (C4V, Alitea, Medina, WA, USA) featuring eight stainless-steel rollers on individual bearings was employed to propel the solutions. A Valco 10-port selector valve (Cheminert, Valco Instruments, Houston, TX, USA) was used to select the flows. Upchurch fittings (Upchurch, Oak Harbor, WA, USA) were used to lock unused ports. The holding and reaction coil tubing and also the tubing connecting the different units was made of PTFE (0.8 mm id).Teflon nuts and ferrules (Upchurch) were used to assemble the manifold. The pump tubing was Phar Med 1.02 mm id tubing (Upchurch), and was held on the pump rollers by FIA peristaltic pump tubing adapters (Upchurch). A reactor module consisting of 0.5 mm id PTFE tubing (Thermoplastic Scientific, NY, USA) of different lengths was used for mixing the solutions. Absorbance measurements were made with a Spectronic Mini-20 spectrophotometer (Milton Roy, Rochester, NY, USA), equipped with a grating monochromator detector and a Unovic ultra-micro-flow-through cell (Unovic Instruments, NY, USA) (20 ml) with a pathlength of 1.0 mm.A personal computer (Austin Computer Systems, Austin, TX, USA) equipped with a 120 Mbyte hard disk, 4 Mbyte RAM and VGA graphics was used to monitor the pump and valve. Communication between the computer and the external devices was effected with a general-purpose I/O board (Model ADA- 110, Real Time Devices, State College, PA, USA).The computer was also used to collect the data; alternatively, the data were recorded by a Model 0555 single-channel strip-chart recorder (Cole-Parmer, Chicago, IL, USA). A Perkin-Elmer (Norwalk, CT, USA) Lambda 5 UV/visible spectrophotometer equipped with 10.0 mm cells was used for preliminary investigations. Software Packages Microsoft windows 3.10 with DOS 6.20 was utilized to run the following software: FIA Lab 2.0 (beta release) (Alitea) was used to program the SIA system.Sigmaplot, version 1.02 (Jandel Scientific, Erkrath, Germany) was employed for data handling calculations and constructing graphs. Microsoft Word, version 6.0 (Microsoft Corporation, USA) was used for constructing tables and writing text. Procedure Fig. 1 shows the SIA manifold used. Fixed amounts of reagent, sample and carrier (wash) solutions were sequentially aspirated into the lines from the ports by means of the selector valve (SV).In each case, the excess solution introduced into the holding coil (HC) was expelled through port 7 to the auxiliary waste. The remaining steps of the procedure, which were controlled by the computer, were as follows: 1, The carrier solution was pumped through port 1, by setting the pump in the forward direction, for 25–45 in order to flush the system (holding coil, reactor and detector) with the carrier solution. 2, A pre-determined volume of the reagent solution was aspirated through port 2 into the holding coil (HC) in the reverse direction.The remaining reagents were aspirated into the holding coil via ports 3–8, as appropriate. 3, The drug solution was aspirated into the holding coil from port 9. The last two steps were achieved by setting the pump in the reverse mode. Finally, the composite zone was propelled by the carrier solution through port 1 to the reaction coil and then the detector. When the mixing chamber was used instead of the reactor coil, the reagent(s) and the sample were initially injected into the mixing chamber.At fixed intervals, aliquots of the solution in the mixing chamber were withdrawn into the holding coil and then pushed to the detector using the carrier by forward pumping. The data were acquired by the computer and transferred to plotting software for further calculations. In each of the above steps the volume of the solutions aspirated was determined from the time of aspiration and the volumetric flow rate of the pump (using the relationship, volume = flow rate 3 time), which was 29.5 ml s21 in this study.Determination of Flow Rate The pump speed can be altered by changing the rev min21 setting on the peristaltic pump. The rate of flow of the reagent inside the pump tubes was determined by withdrawing liquid (distilled water) at one end of the tube and collecting it at the other end in a measuring cylinder. The volume of liquid collected at a certain time gives the flow rate in ml min21 at a particular rev min21 value.This process was repeated for different rev min21 values, giving different values of flow rate. A graph of flow rate versus pump speed (rev min21) was then plotted. A straight line described by the following equation was obtained: Flow rate = 0.73644 + 0.057263(pump speed/rev min21) Therefore, for a pump speed of 500 rev min21, which was used in this work, the flow rate will be 1.77 ml min21 or 29.5 ml s21.Stoichiometry For stoichiometric studies, the SIA system (Fig. 1), including all the tubes attached to the selector valve, was flushed with de- Fig. 1 SIA manifold: C, carrier; P, peristaltic pump; HC, holding coil; SV; selector valve; RC, reaction coil; D, detector; CP, computer; and W, waste. 912 Analyst, September 1997, Vol. 122ionized water (carrier) flowing at a rate of 29.5 ml s21. The following operations were then conducted: 1. Equimolar solutions of PdII and promethazine were connected to the selector valve through ports 2 and 3, respectively.About 80 ml of each reagent were introduced sequentially into the holding coil in the reverse mode for 3.0 s; the excess, together with some carrier, was transferred to the auxiliary waste through valve 7 in the forward mode for 8.0 s. 2. In Job’s method of continuous variation, different aliquots of equimolar solutions of PdII and promethazine were taken and mixed in the holding coil so as to give solutions of identical total concentration (PdII + drug) but different mole fractions; the solution was then pushed for 40.0 s to the detector for signal monitoring.The volume of each reagent aspirated was varied between 14.8 ml (0.5 s) and 147.5 ml (5.0 s). The total volume aspirated was maintained constant at 162.0 ml by adjusting the aspiration times. This step was repeated while varying the aspiration time between 14.8 ml (0.5 s) and 280.3 ml (9.5 s). 3. In the molar ratio method, the total concentration of the ligand (promethazine) was maintained constant by aspirating 147.5 ml (5.0 s) of solution into the holding coil by flow reversal, while the PdII solution volume was varied between 14.8 ml (0.5 s) and 192.0 ml (6.5 s). This step was repeated except that the total concentration of the metal was held constant by fixing the flow reversal for 147.5 ml (5.0 s), while the volume of drug solution was varied between 14.8 ml (0.5 s) and 192.0 ml (6.5 s).Ideally, in the molar ratio method, two straight lines with different slopes are obtained when the absorbance is plotted against the PdII-to-drug ratio, and the point of intersection of these two lines corresponds to the stoichiometric ratio on interpolation to the molar ratio axis. Steps 2 and 3 were repeated but using 8.0 3 1024 mol dm23 hydrochloric acid as the carrier instead of water. There was no significant difference in the absorbance values obtained using either acid or water as carrier.Results and Discussion Chemical System Prior to using the SIA technique, preliminary investigations were made in which the maximum absorbance (lmax) over a wide range of pH values from 8.0 31025 to 0.01 mol dm23 was determined. The optimum absorbance conditions as a function of pH were established by measurements on solutions of PdII (10 mol dm23) and promethazine (150 ppm) at 28 increments of pH between 2.0 and 4.1. Both the wavelength of maximum absorbance and the absorbance itself increased from pH 2 (496.2 nm and 0.303, respectively) to pH 3.12 (503.9 nm and 0.369) before decreasing at pH 4.1 (502.2 nm and 0.279).The optimum pH for this work was selected as 3.1. The stability of the complexes formed favoured this method over previous methods involving oxidation with stronger oxidizing agents than PdII in which a highly unstable radical product is monitored for quantification.1,2,8,9 The nature of the interaction between PdII and promethazine in the pH range 4.1–3.0 is believed to involve protonation of the drug followed by hydrogen bonding between the protonated nitrogen and one of the chlorine atoms of PdCl2.4–6,10 Complexes of the type ML2X2 and MLX3 have also been reported,10 where M = Pd, L = promethazine ligand and X = Cl.An investigation of the type, stoichiometry and formation constant of such a complex was carried out by the Job’s plot11 and molar ratio11,12 methods as described below.Job’s Plot Method In Job’s method of continuous variation, different aliquots of equimolar solutions of PdII and promethazine were mixed to give solutions of identical total concentration (PdII + drug) but different mole fractions. A volume of carrier solution was then aspirated; the carrier volume was selected to allow optimum mixing on flow reversal towards the detector. Analysis of the Job’s plot was used to determine the stoichiometry of the complexation of PdII with promethazine in hydrochloric acid medium.A typical Job’s plot, illustrated in Fig. 2, was obtained by the SIA procedure using 8 3 1024 mol dm23 hydrochloric acid and an ionic strength of 0.20 mol dm23, which was maintained by using lithium perchlorate. Water (de-ionized) was used as a carrier for analysis, but the experiment was also repeated several times with hydrochloric acid as carrier and also with mixing of the reactants in the holding coil. It was found that there was no significant difference when water or acid was used as a carrier, even if the mixing steps were included in the SIA program.The total volume of the two reagents for each run was kept constant at 295.1 and 162.3 ml. It is worth noting that both reagents, viz., PdII and promethazine, were injected, thus consuming minimal amounts of reagent compared with the manual procedure. It is clear that the curves A = f (Xdrug) exhibit a maximum for mole fractions of 0.5, indicating that the ratio of PdII : drug in the complex is 1 : 1.Only 10 min are needed to generate the Job’s plot using SIA and less than 10 ml of the drug is sufficient to repeat the procedure many times. Attempts to apply the molar ratio method12,13 failed because the PdII concentration was a greater effect on the system, resulting in the formation of a mixed complex together with the oxidized form of the drug at higher concentrations of PdII, which makes the method unsatisfactory. This is in agreement with previous work, which indicated that various complex forms are possible.7 Overall, the Job’s plot method is the more reliable and it has clearly shown that the molar ratio of the PdII–promethazine complex is 1 : 1.Fig. 2 Typical SIA trace representing a Job’s plot for the promethazine system: [PdII] = [promethazine] = 1 3 1023 mol dm23 in 8 3 1024 mol dm23 hydrochloric acid; ionic strength = 0.20 mol dm23. The total aspiration volume is equivalent to 295 ml and aspiration volumes were varied between 14.8 ml and 280.3 ml.Mole fraction of the drug: (1) 0; (2) 0.05; (3) 0.10; (4) 0.20; (5) 0.30; (6) 0.40; (7) 0.50; (8) 0.60; (9) 0.70; (10) 0.80; (11) 0.90; (12) 0.95; (13) 0.975; and (14) 0.10 ml. Analyst, September 1997, Vol. 122 913Formation Constant The formation constant and the composition of the PdII– promethazine complex were also investigated by numerical methods.14,15 The JOBCON16,17 program was used to analyse the continuous variation data.The program was modified by the authors and re-written in C-language and applied on a PC/AT computer. The calculations are based on fitting a function f(x, b) to a set of experimental data, using a least-squares method. Unknown parameters are estimated by minimizing U, the sum of squares of residuals, defined by the following equation: U A A i n = - = å ( ) exp calc 2 1 where n represents the number of experimental points, Aexp the experimental absorbance, and Acalc = f (x, b), the absorbance calculated by the program from formation constants and stoichiometric ratios.Therefore, various experimental models could be fitted to the experimental data iteratively by varying the values of the formation constant and stoichiometric ratio. The JOBCON program was used to calculate the formation constant from the continuous variation data for the PdII– promethazine complex. A 1 : 1 metal-to-ligand ratio was found to be the most probable, with a relative error of 14.99%, resulting in a value for the logarithm of the formation constant (log Kf) of 4.349.Other metal-to-ligand ratios (m: n) appeared to be improbable, giving high relative errors, particularly those where m: n > 2. The value of log Kf reported10 previously for promethazine was 5.52, which is close to the value obtained in the present work for the 1 : 1 PdII–promethazine complex. It was also observed that the stoichiometry and the value of the formation constant are independent of the total concentration of the metal ion and the ligand.Calibration Graph A series of standard solutions of promethazine hydrochloride were run in triplicate; a graph of absorbance versus concentration was found to be linear in the range 50–400 ppm. The calibration equation was: A = 0.008654 + 0.000110 C where A is absorbance and C is concentration in ppm; the correlation coefficient (r2) was 0.997. The peak width at the baseline was measured to be 18 s, thus giving a throughput of 200 samples h21.The peak width was also determined at 60% peak height and found to be 2.4 s, indicating minimal dispersion. An average relative standard deviation (RSD) of 0.92% was obtained for six replicate determinations of 250 ppm of promethazine, indicating excellent reproducibility. Application The method was applied to the determination of promethazine hydrochloride in several commercial formulations, namely, Phenergen tablets, Phenergen promethazine, Phenergen Expectorant and Cigan Elixir.All drug formulations contained starch and glucose as excipients. The same formulations were also analysed by the BP method3 and the results were statistically compared by calculating the percentage recoveries, standard deviations and the Student’s t-test values. The results are shown in Table 1. The accuracy of the proposed method was comparable to that of the BP method. The results also show that common excipients do not interfere with the determination. The proposed method is simpler and more selective than the BP and previous methods. The utilization of PdII avoided oxidation of the drug during the analysis. Conclusion The proposed SIA method for the determination of promethazine was validated by complexation of the drug with PdII in dilute hydrochloric acid medium and was found to be superior to earlier FIA methods. The method developed was applied to the determination of promethazine in commercial formulations and was found to be superior to the official BP method, with a wide dynamic range (50–400 ppm), excellent reproducibility and an RSD of less than 0.96%. The capacity of the SIA technique was demonstrated by its application to the determination of concentrations, stoichiometries and formation constants of complexation reactions. A large number of experimental trials were performed utilizing relatively small amounts of reagents. N.I.D. thanks KFUPM for the award of a scholarship to study for an M.S. degree. References 1 Sultan, S. M., Analyst, 1991, 116, 177. 2 Sultan, S. M., and Suliman, F. O., Anal. Sci., 1992, 8, 841. 3 British Pharmacopoeia, HM Stationery Office, London, 5th edn., 1988, pp. 749 and 995. 4 Pellizzetti, E., J. Chem. Soc., Dalton Trans., 1980, 484. 5 Geary, W. J., Mason, N. J., Nowell, I. W., and Nixon, L. A., J. Chem. Soc., Chem. Commun., 1984, 1064. 6 Duddell, D. A., Goggin, P. L., Goodfellow, R. J., Norton, M. G., and Smith, J. G., J. Chem. Soc. A, 1970, 546. 7 Sultan, S. M., Suliman, F. O., and Bahruddin, B. S., Analyst, 1995, 120, 561. 8 Sultan, S. M., Microchem. J., 1991, 44, 304. 9 Sultan, S. M., and Abdennabi, A. M., Microchem. J., 1993, 48, 343. 10 Geary, W. J., Mason, N. J., Nowell, I. W., and Nixon, L. A., J. Chem. Soc., Dalton Trans., 1982, 1103. 11 Job, P., Ann. Chim., 1928, 9, 113. 12 Momoki, K., Sekino, J., Sato, H., and Yamaguchi, N., Anal. Chem., 1969, 41, 1286. 13 Yoe, J. H., and Jones, A. L., Ind. Eng. Chem. Anal. Ed., 1944, 64, 111. 14 Sabatini, A., Vacca, A., and Gans, P., Coord. Chem. Rev., 1992, 120, 289. 15 Sultan, S. M., Anal. Lett., 1991, 24, 1785. 16 Likussar, W., Anal. Chem., 1973, 45, 1926. 17 Meloun, M., and Javurek, M., Talanta, 1984, 31, 1083. Paper 7/01810J Received March 14, 1997 Accepted May 7, 1997 914 Analyst, September 1997, Vol. 122
ISSN:0003-2654
DOI:10.1039/a701810j
出版商:RSC
年代:1997
数据来源: RSC
|
10. |
Determination of Selenium by Atomic Absorption Spectrometry WithSimultaneous Retention of Selenium(IV) and Tetrahydroborate(III) on anAnion-exchange Resin Followed by Flow Injection Hydride Generation Fromthe Solid Phase |
|
Analyst,
Volume 122,
Issue 9,
1997,
Page 915-919
Pablo E. Carrero,
Preview
|
|
摘要:
Determination of Selenium by Atomic Absorption Spectrometry With Simultaneous Retention of Selenium(IV) and Tetrahydroborate(III) on an Anion-exchange Resin Followed by Flow Injection Hydride Generation From the Solid Phase Pablo E. Carrero† and Julian F. Tyson* Department of Chemistry, Box 34510, University of Massachusetts, Amherst, MA 01003-4510, USA Selenium(IV) and tetrahydroborate(III) (borohydride) were simultaneously retained on a strong anion-exchange resin in a packed column.Hydrogen selenide was generated by passage of an injected zone of hydrochloric acid with subsequent detection by AAS with quartz tube atomization. The limits of detection, defined as the concentration giving a signal equal to 3s of the blank, were 0.24, 0.15 and 0.12 mg l21 of Se for 1, 2 and 3 min preconcentration at a sample flow rate of 3 ml min21, respectively. The precision of the procedure, expressed as the RSD of 10 successive determinations of 5, 10, and 20 mg l21 of Se, varied from 0.41 to 1.32, 0.24 to 0.81 and 0.18 to 0.61% for 1, 2 and 3 min preconcentration, respectively.The system was used for the determination of selenium in river, lake and tap water matrices. No appreciable matrix effects were observed and the system was calibrated with aqueous solutions of a pure selenium salt (Na2SeO3). The recoveries of spikes (0.5, 2, and 10 mg l21 of Se) added to the water samples ranged from 96.0 to 102.0, 96.0 to 107.0 and 98.9 to 108% for river, lake and tap water, respectively.Keywords: Selenium(IV); ion-exchange preconcentration; solid-phase hydride generation; water analysis; flow injection There are significant advantages in the use of hydride generation (HG) with AAS or other atomic spectrometric detection methods for the determination of elements which form volatile hydrides. Hydride generation has become one of the most powerful and well established techniques for the determination of arsenic, antimony, bismuth, germanium, lead, selenium, tellurium, tin, indium, and thallium.A variety of reactions have been used to convert the analyte in solution into the hydride.1–8 There are also several reports of electrochemical hydride generation9–11 and of thermochemical gas-phase hydride generation.12 In the case of solution reactions, a metal– acid combination, or more commonly reaction with borohydride is employed for hydride generation. In early applications of the technique, the March reaction, based on a metal–acid system such as Zn–HCl producing nascent hydrogen which reacted with the analyte to form the hydride, was used.3 The more commonly used reaction for hydride formation is the BH42–acid reaction.13 Although some precursors are hydrated cations in the appropriate oxidation state, as for example with PbIV; for most analytes the precursor may well exist as an oxo-anion which is reduced prior to the final hydride transfer reaction, as would be the case with SeIV.Initial applications used NaBH4 pellets, but an aqueous solution stabilized by potassium or sodium hydroxide4 may be conveniently used and is currently the most popular reagent. Other forms of borohydride have been used. Borohydride bound to a stationary phase in a column has been used for hydride generation of arsenic5,6 and selenium.7,8 However, the only advantage reported over the use of an aqueous solution of BH42 was the suppression of some matrix interferences.In our current work on hydride generation with in-atomizer trapping for determination of low concentrations of selenium by ETAAS,14 we have found that detection limits are governed by impurities in the borohydride reagent. We considered that the immobilization of borohydride on an anion-exchange resin might result in a purer reagent and hence improved detection limits. In the course of this study, we found that selenium was retained on the anion-exchange resin and that hydrogen selenide could be generated from the resin by the passage of acid when both selenium and borohydride were co-immobilized. In this paper, we report on the analytical performance of a system for the determination of selenium based on this method of generation of hydrogen selenide.Experimental Reagents All reagents were obtained from Fisher (Pittsburgh, PA, USA) and were of analytical-reagent grade, unless stated otherwise. Doubly distilled 18 MW E-pure water was used throughout the experiment. A stock standard solution of 1000 mg l21 of selenite (SeO3 22) was used.Other concentrations were obtained by dilution. Sodium borohydride, 10% m/v in 2% m/v sodium hydroxide, was filtered through Whatman No. 2 filter-paper and stored at 4 °C for 4 weeks without deterioration. Other concentrations were obtained by dilution. The resins used were Amberlite IRA-410 and Amberlyst A26 (Aldrich, Milwaukee, WI, USA), which are strongly basic anion-exchange resins (styrene–divinylbenzene skeletal structure).Apparatus The detection unit used was a Perkin-Elmer (Norwalk, CT, USA) Model 3100 atomic absorption spectrometer. The atomizer was a flame-heated quartz T-shaped tube. The quartz tube was cleaned weekly with acid following the procedure described by Hatfield.15 A hollow-cathode lamp (Perkin-Elmer) was used as a light source. The wavelength for Se was 196.0 nm. The spectral bandpass was 0.7 nm. † On leave from IVAIQUIM (Venezuelan Andean Institute for Chemical Research), Faculty of Sciences, University of Los Andes, P.O.Box 542, M�erida 5101-A, Venezuela. Analyst, September 1997, Vol. 122 (915–919) 915Manifold The manifold, shown schematically in Fig. 1, was constructed from 0.8 mm id PTFE tubing, except the tubing between the argon confluence point, the gas–liquid separator and quartz tube, which was PTFE tubing of 1.5 mm id. The gas–liquid separator consisted of a 25 ml separating funnel with a twoholed rubber cap.The column consisted of a glass tube of 150 mm length and 4 mm id. Two PTFE reducing unions of 1/4 3 1/8 in (Cole-Parmer, Chicago, IL, USA), fitted at either end of the column, were used to connect it to the manifold. A slurry of anion-exchange resin was introduced into the column with the aid of a syringe. A small amount of glass-wool was placed at the ends of the column to prevent loss of the resin. The column was ready for use after washing several times alternately with borohydride and hydrochloric acid solutions.Two six-port PTFE Rheodyne rotary valves (Supelco, Bellefonte, PA, USA) were used. The ion-exchange column was located in the injection loop of one and the other was used to introduce a discrete volume (500 ml) of acid. Two Ismatec SA MS-Reglo Model 7331-10 peristaltic pumps (from Cole- Palmer) were used, one for the carrier line and the other for the acid, sample, and borohydride lines. The waste from the gas– liquid separator was pumped with a Lachat Instruments (Milwaukee, WI, USA) Model 1200-000 peristaltic pump.The flows of reagents were regulated by using Tygon pump tubing (Cole-Parmer) of different internal diameters and by control of the pump head rotation speed. Recommended Procedure All the experiments were carried out using the manifold shown in Fig. 1. In the load position, Fig. 1(a), the sample of selenite (SeO3 22) and the borohydride were mixed and pumped through the column for a period of 1, 2 or 3 min, resulting in the simultaneous retention of both anions (SeO3 22, or possibly Se22, and BH42).At the same time, the loop of valve 1 was filled with acid, while the de-ionized water carrier was pumped constantly through the system. In the injection position, Fig. 1(b), valve 2 was switched and the column was washed for a period of 20 s with carrier solution. Valve 1 was then switched and the acid was carried through the column generating hydrogen selenide.The column was cleaned between samples and standards by passing 4% m/v borohydride in 0.5% m/v NaOH for 30 s followed by an injection of acid. The cleaning step was not needed between replicate injections of the same sample or standard. Method Development Determination of arsenic Initial experiments involved repeating the procedures described by Tesfalidet and Irgum5 and Narasaki et al.6 for theetermination of arsenic. In these procedures, the borohydride was first loaded on to the anion-exchange column and, after washing, the acidified analyte solution was passed through the column generating arsine and hydrogen.These experiments are not described in detail here. The detailed experimental work concerns the development of the new procedure for the determination of selenium. Gas–liquid separation Two types of gas–liquid separator were used, namely the Perkin-Elmer device from the FIAS unit (this was the empty plastic vessel) and a device constructed in-house from a 25 ml separating funnel. In neither case was any additional device used to reduce the transfer of water vapor or droplets to the atomizer.The drains from both these devices were pumped. In the case of the separating funnel device the drain pump rate was such that there was always 1–2 ml of liquid in the funnel. Column dimensions Two different strongly basic resins, Amberlite IRA-410 and Amberlyst A26, were packed in columns of various dimensions.Four different lengths (50, 100, 150 and 200 cm) and two different internal diameters (2 and 4 mm) were used in all possible combinations. Both resins have styrene–divinylbenzene skeletal structures; however, Amberlite IRA-410 is a gel-type resin and Amberlyst A26 is a porous or macroreticular resin. The resulting columns were tested for selenium retention and subsequent hydride generation for six replicate measurements at two concentrations of selenium, 10 and 25 mg l21.Parameter optimization The multi-cycle alternating variable search optimization method16,17 was used for the optimization of the following parameters: borohydride concentration, HCl concentration, carrier flow rate and stripping gas flow rate. The figure of merit for the optimization process was maximum net peak height sensitivity (i.e., signal minus blank). Other parameters that were studied include the dimensions of the column, nature of the anion-exchange resin, oxidation state of the selenium and the flow rates of the sample and borohydride solutions.Parameter optimization was carried out with a 1 min preconcentration (at 3 ml min21 sample and borohydride flow rates), using a column size of 150 3 4 mm id packed with Amberlite IRA-410 anion-exchange resin. The optimization of the borohydride and HCl concentrations was carried out using sample solutions of 0, 10 and 25 mg l21 of Se. The effect of such reagents was studied by varying these concentrations within the ranges 0.01–2% m/v NaBH4 in 0.01% m/v NaOH, and 0.1–2 mol l21 HCl.Samples of 10 and 25 mg l21 were used in the studies of the effects of the argon stripping gas flow rate and the carrier flow rate, i.e., the speed at which the acid passed through the column. The argon flow rates were varied between 50 and Fig. 1 Schematic diagram of the manifold for selenium preconcentration and hydride generation. (a) load position and (b) injection position.V1,2, six-port valves; P1,2,3, peristaltic pumps; QAC, quartz atomization cell; GLS, gas–liquid separator; CL, column packing with strong anionexchange resin; L, 500 ml loop; CS, carrier solution, distilled, deionized 18 MW E-pure water (14 ml min21); AS, 4 mol l21 HCl (1.5 ml min21); RS, 0.05% m/v NaBH4 in 0.01% m/v NaOH (3.0 ml min21); S, sample (3.0 ml min21); and W, waste. 916 Analyst, September 1997, Vol. 1221000 ml min21. The carrier flow rate was varied between 4 and 15 ml min21.Analytical Performance Using the optimal experimental conditions (given in Table 1), calibration graphs with 0, 5, 10, 15, 20, and 25 mg l21 of Se for 1, 2 and 3 min of preconcentration were made. The precision of the system was evaluated as the RSD of 10 successive determinations of 5, 10 and 20 mg l21 of Se for 1, 2 and 3 min of preconcentration, respectively. The accuracy was evaluated by means of the recovery of 0.5, 2 and 10 mg l21 of Se spiked in river, lake and tap water.The effect of the sample matrix was evaluated by the method of standard additions. Equal volumes of river, lake and tap water (45 ml) were taken, all but one were separately spiked with different amounts of selenium and then all were diluted to equal volumes (50 ml) to obtain three series of solutions with final concentrations added of 0, 5, 10, 15, 20 and 25 mg l21 of Se. Signals were obtained under the optimal experimental conditions (Table 1) for a 3 min preconcentration time.All calibrations were obtained by an unweighted least-squares procedure. Results and Discussion Determination of Arsenic The results reported previously5,6 for the determination of arsenic were confirmed. Signals were obtained for both AsIII and AsV. For a sample volume of 835 ml, the linear ranges for AsIII and AsV were 0.6–40 and 0.9–50 mg l21, respectively. The precisions, expressed as RSD for five replicate determinations of 20 mg l21, were between 1.8% and 3.2% for AsIII and between 2.4% and 4.1% for AsV.Gas–Liquid Separator The device based on the 25 ml separating funnel was used, rather than the smaller internal volume Perkin-Elmer device, to avoid the carry-over of liquid which tends to occur with certain combinations of argon gas flow rate and borohydride concentration with devices having small internal volumes. As the optimization studies required that these parameters be varied, the use of a more robust gas–liquid separator decreased the down-time spent cleaning and regenerating the surface of the quartz cell.However, for an FI-based procedure the gas–liquid separator will affect the peak height sensitivity by virtue of a contribution to the overall dispersion and it is likely that the device used here is sub-optimal with respect to this characteristic. Column Dimensions and Resin Type The best results were observed with larger column sizes (lengths and id) with both resins, i.e., when the amount of resin was increased.However, lengths above 150 mm for the same internal diameter did not produce any improvement in the signal. The Amberlite IRA-410 gel-type resin produced better results than the Amberlyst A26 macroreticular resin. Therefore, a 150 3 4 mm id column packed with Amberlite IRA-410 geltype resin was chosen for further experiments. Parameter Optimization The optimum conditions are given in Table 1. No signals were obtained from solutions of SeVI.To achieve the best sensitivity, three cycles of the optimization process were necessary. The results of the third cycle for each parameter are discussed below. The effect of the concentration of borohydride is shown in Fig. 2. There was a steady increase in the blank signal as the concentration of borohydride increased from 0.01 to 1.0% m/v, with a slower increase from 1.0 to 2.0% m/v. For both samples (10 and 25 mg l21 of Se) a sharp increase in the signal was observed as the borohydride concentration increased from 0.01 to 0.05% m/v NaBH4, followed by a sharp decrease from 0.5 to 1.0% m/v.No signal was obtained in the absence of borohydride. The 0.05% m/v concentration produced the optimum net signal. This concentration of borohydride is considerably lower than that used in typical flow injection and typical batch procedures. The effect of the concentration of HCl is shown in Fig. 3. As can be seen, the signals from the samples and the blank increased as the HCl concentration increased to 0.5 mol l21, but changed little thereafter.Therefore, 4 mol l21 was chosen for further experiments. The effect of carrier flow rate is shown in Fig. 4. Carrier flow rates between 12 and 15 ml min21 resulted in the best net absorbance signal. A carrier flow rate of 14 ml min21 was used in further experiments. The effect of the stripping gas flow rate is shown in Fig. 5. The most favorable rate was found to be 180 ml min21, hence this value was chosen for further experiments.Table 1 Optimum operation conditions Atomic absorption spectrometer— Wavelength 196 nm Slit width 0.7 nm Lamp current 16 mA Quartz cell temperature 900 °C Background correction On On-line preconcentration and hydride generation— HCl concentration 4 mol l21 NaBH4 concentration 0.05% m/v Sample flow rate 3.0 ml min21 NaBH4 flow rate 3.0 ml min21 Carrier flow rate 14.0 ml min21 Argon flow rate 100 ml min21 Column size (length3id) 15034 mm Resin Amberlite IRA-410 Fig. 2 Effect of concentration of NaBH4 solution on the signal peak height (absorbance). A, no Se; B, 10 mg l21 of Se; and C, 25 mg l21 of Se. The error bars represent the standard deviation for five replicate measurements. Analyst, September 1997, Vol. 122 917Analytical Performance The calibration equations and the other performance figures of merit are summarized in Table 2. The system responded linearly from the detection limit up to 180, 120 and 80 mg l21 of Se for 1, 2 and 3 min preconcentration, respectively.The precision of the procedure as a function of concentration, expressed as RSD, varied from 0.41 to 1.32, 0.24 to 0.81 and 0.18 to 0.61% for 1, 2 and 3 min preconcentration, respectively. The precision of the system improved with preconcentration time, but degraded severely if the residual sample and borohydride were not washed from the column. Carry-over between samples and standards was prevented by flushing residual selenium from the column with the 4% borohydride wash.The limits of detection, defined as the concentration giving a signal equal to three times the standard deviation of the blank signal, were 0.24, 0.15 and 0.12 mg l21 of Se for 1, 2 and 3 min preconcentration, respectively. The sample throughput was 21 h21 for 1 min preconcentration. The results of the standard additions to the water sample matrices were compared with those for the same concentrations in distilled water.The characteristics of the regression lines are summarized in Table 3. The 95% confidence interval about the slope of the calibration in the distilled water matrix contains the slopes of the calibrations in the other matrices, i.e., the confidence intervals of the slopes overlap in all cases and there is no significant difference between the slopes for the different matrices. As no appreciable matrix effect was observed, the analyses of the test samples were performed by calibration with aqueous solutions of a pure selenium salt (Na2SeO3).The recoveries (and standard deviations) of spikes added to the water samples are given in Table 4 and ranged from 96 ± 4 to 102 ± 6, 96 ± 4 to 107 ± 6 and 99 ± 4 to 108 ± 6%, depending on the concentration, for river, lake and tap water, respectively. These values indicate that selenium can be quantitatively determined in such samples. Fig. 3 Effect of concentration of HCl solution on the signal peak height (absorbance). A, no Se; B, 10 mg l21 of Se; and C, 25 mg l21 of Se.The error bars represent the standard deviation for five replicate measurements. Fig. 4 Effect of the carrier solution flow on the net signal peak height (absorbance). A, 10 mg l21 of Se; and B, 25 mg l21 of Se. The error bars represent the standard deviation for five replicate measurements. Table 2 Analytical performance of the system Regression RSD (%)¶ Time*/ equation: LOD (3s)§/ (min) A = b + mC† r‡ mg l21 5 mg l21 10 mg l21 20 mg l21 1 A = 0.011 + 0.0123C 0.9996 0.24 1.32 0.83 0.41 2 A = 0.022 + 0.0197C 0.9997 0.15 0.81 0.46 0.24 3 A = 0.035 + 0.0256C 0.9997 0.12 0.61 0.36 0.18 * Time of passage of sample through the column at 3 ml min21. † A is absorbance, b is intercept, m is slope and C is concentration of Se in mg l21.‡ Regression coefficient. § LOD (3s) is the detection limit, calculated for 3s/m, where s is the within-run standard deviation of a blank determination (n = 10).¶ RSD for 5, 10 and 20 mg l21 of Se (n = 10). Fig. 5 Effect of stripping gas (argon) flow rate on the net signal peak height (absorbance). A, 10 mg l21 of Se; and B, 25 mg l21 of Se. The error bars represent the standard deviation for five replicate measurements. 918 Analyst, September 1997, Vol. 122Determination of SeIV in Water The results of the application of the proposed method to the determination of selenium in river, lake and tap water are given in Table 4 for a sample volume of 9 ml (3 min preconcentration time). The concentration of selenium in the river water sample was 0.14 ± 0.01 mg l21.The concentrations of selenium in the lake and tap water samples analyzed were below the detection limit of 0.12 mg l21. It is known that the concentration of SeO3 22 in natural water18–20 is usually very low and that selenium is present in water in various oxidation states and chemicals forms, including organic species.18,21,22 Conclusions The co-immobilization of selenium and borohydride on an anion-exchange resin followed by passage of acid forms the basis of a viable method for the determination of selenium by HGAAS and quantitative recovery of selenium from water samples can be achieved.The procedure could be fully automated via a system of computer-controlled valves and pumps and can be used with any atomic spectrometric detection system. Work is in progress to implement the procedure on a Perkin-Elmer FIAS unit.The preconcentration of the analyte on the anion-exchange column allows the potential for improved sensitivity over conventional flow injection hydride generation techniques and the possibility of the separation of the analyte from cation interferences. Current development of the procedure is focused on overcoming such matrix interferences as it may be possible to immobilize the sample and the borohydride successively. Financial support from the University of the Andes, M�erida, Venezuela, for P.C.is gratefully acknowledged and the authors thank Robert I. Ellis for helpful discussions. References 1 Nakahara, T., Spectrochim. Acta Rev., 1991, 14, 95. 2 Godden, R. G., and Thomerson, D. R., Analyst, 1980, 105, 1137. 3 Branch, C. H., and Hutchison, D., Analyst, 1985, 110, 163. 4 Agterdenbos, J., and Bax, D., Fresenius’ Z. Anal. Chem., 1986, 323, 783. 5 Tesfalidet, S., and Irgum, K., Anal. Chem., 1989, 61, 2079. 6 Narasaki, H., Kato, Y., and Kimura, H., Anal.Sci., 1992, 8, 893. 7 Cao, J.Y., and Narasaki, H., Bunseki Kagaku, 1994, 43,169. 8 Tesfalidet, S., and Irgum, K., Fresenius’ J. Anal. Chem., 1991, 341, 532. 9 Brockmann, A., Nonn, C., and Golloch, A., J. Anal. At. Spectrom., 1993, 8, 397. 10 Lin, Y., Wang, X., Yuan, D., Yang, P., Huang, B., and Zhuang, Z., J. Anal. At. Spectrom., 1992, 7, 287. 11 Ding, W.-W., and Sturgeon, R. E., J. Anal. At. Spectrom., 1996, 11, 225. 12 Blais, J. S., Momplaisir, G. M., and Marshall, W.D., Anal. Chem., 1990, 62, 1161. 13 Howard, A. G., J. Anal. At. Spectrom., 1997, 12, 267. 14 Tyson, J. F., Sundin, N. G., Hanna, C. P., and McIntosh, S. A., Spectrochim. Acta Part B, in the press. 15 Hatfield, D. B., Anal. Chem., 1987, 59, 1887. 16 Greenfield, S., Salman, M. S., Thomsen, M., and Tyson, J. F., J. Anal. At. Spectrom., 1989, 4, 55. 17 Miller, J. C., and Miller, J. N., Statistics for Analytical Chemistry, 3rd edn, Ellis Horwood, Chichester, 1993, pp. 185–187. 18 Cutter, G. A., Anal. Chim. Acta, 1978, 98, 59. 19 Measures, C. I., and Burton, J. D., Nature (London), 1978, 273, 293. 20 Yu, Q., Liu, G. Q., and Jin, Q., Talanta, 1983, 30, 265. 21 Cutter, G. A., Anal. Chim. Acta, 1983, 149, 391. 22 Robberecht, H., and Van Grieken, R., Talanta, 1982, 29, 823. Paper 7/01648D Received March 10, 1997 Accepted June 3, 1997 Table 3 Standard additions of selenium to water samples Regression line characteristics Confidence Matrix Slope sslope limits* Intercept sint r† Distilled water 0.0257 0.00048 0.0013 0.032 0.0073 0.9995 River water 0.0249 0.00045 0.0012 0.039 0.0067 0.9998 Lake water 0.0251 0.00045 0.0012 0.035 0.0069 0.9998 Tap water 0.0248 0.00043 0.0012 0.034 0.0061 0.9999 * Confidence limits for the slope, given by ts; the t-value was taken at the 95% confidence level and n 2 2 degrees of freedom.† Correlation coefficient. Table 4 Analytical results for various water samples Selenium Selenium added/ found/ Sample mg l21 mg l21 Recovery (%) River water (Fort river, Amherst) 0 0.14 ± 0.01 — 0.5 0.65 ± 0.03 102 ± 6 2.0 2.06 ± 0.07 96 ± 4 10.0 10.05 ± 0.39 99 ± 4 Lake water (Puffers Pond, Sunderland) 0 < DL*mdash; 0.5 0.48 ± 0.02 96 ± 4 2.0 2.05 ± 0.05 103 ± 3 10.0 10.70 ± 0.56 107 ± 6 Tap water (Amherst) 0 < DL* — 0.5 0.54 ± 0.03 108 ± 6 2.0 2.01 ± 0.05 101 ± 3 10.0 9.89 ± 0.43 99 ± 4 * Detection limit, 0.12 mg l21 of Se.Analyst, September 1997, Vol. 122 919 Determination of Selenium by Atomic Absorption Spectrometry With Simultaneous Retention of Selenium(IV) and Tetrahydroborate(III) on an Anion-exchange Resin Followed by Flow Injection Hydride Generation From the Solid Phase Pablo E.Carrero† and Julian F. Tyson* Department of Chemistry, Box 34510, University of Massachusetts, Amherst, MA 01003-4510, USA Selenium(IV) and tetrahydroborate(III) (borohydride) were simultaneously retained on a strong anion-exchange resin in a packed column. Hydrogen selenide was generated by passage of an injected zone of hydrochloric acid with subsequent detection by AAS with quartz tube atomization.The limits of detection, defined as the concentration giving a signal equal to 3s of the blank, were 0.24, 0.15 and 0.12 mg l21 of Se for 1, 2 and 3 min preconcentration at a sample flow rate of 3 ml min21, respectively. The precision of the procedure, expressed as the RSD of 10 successive determinations of 5, 10, and 20 mg l21 of Se, varied from 0.41 to 1.32, 0.24 to 0.81 and 0.18 to 0.61% for 1, 2 and 3 min preconcentration, respectively.The system was used for the determination of selenium in river, lake and tap water matrices. No appreciable matrix effects were observed and the system was calibrated with aqueous solutions of a pure selenium salt (Na2SeO3). The recoveries of spikes (0.5, 2, and 10 mg l21 of Se) added to the water samples ranged from 96.0 to 102.0, 96.0 to 107.0 and 98.9 to 108% for river, lake and tap water, respectively.Keywords: Selenium(IV); ion-exchange preconcentration; solid-phase hydride generation; water analysis; flow injection There are significant advantages in the use of hydride generation (HG) with AAS or other atomic spectrometric detection methods for the determination of elements which form volatile hydrides. Hydride generation has become one of the most powerful and well established techniques for the determination of arsenic, antimony, bismuth, germanium, lead, selenium, tellurium, tin, indium, and thallium.A variety of reactions have been used to convert the analyte in solution into the hydride.1–8 There are also several reports of electrochemical hydride generation9–11 and of thermochemical gas-phase hydride generation.12 In the case of solution reactions, a metal– acid combination, or more commonly reaction with borohydride is employed for hydride generation. In early applications of the technique, the March reaction, based on a metal–acid system such as Zn–HCl producing nascent hydrogen which reacted with the analyte to form the hydride, was used.3 The more commonly used reaction for hydride formation is the BH42–acid reaction.13 Although some precursors are hydrated cations in the appropriate oxidation state, as for example with PbIV; for most analytes the precursor may well exist as an oxo-anion which is reduced prior to the final hydride transfer reaction, as would be the case with SeIV.Initial applications used NaBH4 pellets, but an aqueous solution stabilized by potassium or sodium hydroxide4 may be conveniently used and is currently the most popular reagent. Other forms of borohydride have been used. Borohydride bound to a stationary phase in a column has been used for hydride generation of arsenic5,6 and selenium.7,8 However, the only advantage reported over the use of an aqueous solution of BH42 was the suppression of some matrix interferences. In our current work on hydride generation with in-atomizer trapping for determination of low concentrations of selenium by ETAAS,14 we have found that detection limits are governed by impurities in the borohydride reagent. We considered that the immobilization of borohydride on an anion-exchange resin might result in a purer reagent and hence improved detection limits.In the course of this study, we found that selenium was retained on the anion-exchange resin and that hydrogen selenide could be generated from the resin by the passage of acid when both selenium and borohydride were co-immobilized.In this paper, we report on the analytical performance of a system for the determination of selenium based on this method of generation of hydrogen selenide. Experimental Reagents All reagents were obtained from Fisher (Pittsburgh, PA, USA) and were of analytical-reagent grade, unless stated otherwise. Doubly distilled 18 MW E-pure water was used throughout the experiment.A stock standard solution of 1000 mg l21 of selenite (SeO3 22) was used. Other concentrations were obtained by dilution. Sodium borohydride, 10% m/v in 2% m/v sodium hydroxide, was filtered through Whatman No. 2 filter-paper and stored at 4 °C for 4 weeks without deterioration. Other concentrations were obtained by dilution. The resins used were Amberlite IRA-410 and Amberlyst A26 (Aldrich, Milwaukee, WI, USA), which are strongly basic anion-exchange resins (styrene–divinylbenzene skeletal structure).Apparatus The detection unit used was a Perkin-Elmer (Norwalk, CT, USA) Model 3100 atomic absorption spectrometer. The atomizer was a flame-heated quartz T-shaped tube. The quartz tube was cleaned weekly with acid following the procedure described by Hatfield.15 A hollow-cathode lamp (Perkin-Elmer) was used as a light source. The wavelength for Se was 196.0 nm. The spectral bandpass was 0.7 nm.† On leave from IVAIQUIM (Venezuelan Andean Institute for Chemical Research), Faculty of Sciences, University of Los Andes, P.O. Box 542, M�erida 5101-A, Venezuela. Analyst, September 1997, Vol. 122 (915–919) 915Manifold The manifold, shown schematically in Fig. 1, was constructed from 0.8 mm id PTFE tubing, except the tubing between the argon confluence point, the gas–liquid separator and quartz tube, which was PTFE tubing of 1.5 mm id. The gas–liquid separator consisted of a 25 ml separating funnel with a twoholed rubber cap.The column consisted of a glass tube of 150 mm length and 4 mm id. Two PTFE reducing unions of 1/4 3 1/8 in (Cole-Parmer, Chicago, IL, USA), fitted at either end of the column, were used to connect it to the manifold. A slurry of anion-exchange resin was introduced into the column with the aid of a syringe. A small amount of glass-wool was placed at the ends of the column to prevent loss of the resin. The column was ready for use after washing several times alternately with borohydride and hydrochloric acid solutions.Two six-port PTFE Rheodyne rotary valves (Supelco, Bellefonte, PA, USA) were used. The ion-exchange column was located in the injection loop of one and the other was used to introduce a discrete volume (500 ml) of acid. Two Ismatec SA MS-Reglo Model 7331-10 peristaltic pumps (from Cole- Palmer) were used, one for the carrier line and the other for the acid, sample, and borohydride lines. The waste from the gas– liquid separator was pumped with a Lachat Instruments (Milwaukee, WI, USA) Model 1200-000 peristaltic pump.The flows of reagents were regulated by using Tygon pump tubing (Cole-Parmer) of different internal diameters and by control of the pump head rotation speed. Recommended Procedure All the experiments were carried out using the manifold shown in Fig. 1. In the load position, Fig. 1(a), the sample of selenite (SeO3 22) and the borohydride were mixed and pumped through the column for a period of 1, 2 or 3 min, resulting in the simultaneous retention of both anions (SeO3 22, or possibly Se22, and BH42).At the same time, the loop of valve 1 was filled with acid, while the de-ionized water carrier was pumped constantly through the system. In the injection position, Fig. 1(b), valve 2 was switched and the column was washed for a period of 20 s with carrier solution. Valve 1 was then switched and the acid was carried through the column generating hydrogen selenide.The column was cleaned between samples and stanrds by passing 4% m/v borohydride in 0.5% m/v NaOH for 30 s followed by an injection of acid. The cleaning step was not needed between replicate injections of the same sample or standard. Method Development Determination of arsenic Initial experiments involved repeating the procedures described by Tesfalidet and Irgum5 and Narasaki et al.6 for the determination of arsenic.In these procedures, the borohydride was first loaded on to the anion-exchange column and, after washing, the acidified analyte solution was passed through the column generating arsine and hydrogen. These experiments are not described in detail here. The detailed experimental work concerns the development of the new procedure for the determination of selenium. Gas–liquid separation Two types of gas–liquid separator were used, namely the Perkin-Elmer device from the FIAS unit (this was the empty plastic vessel) and a device constructed in-house from a 25 ml separating funnel.In neither case was any additional device used to reduce the transfer of water vapor or droplets to the atomizer. The drains from both these devices were pumped. In the case of the separating funnel device the drain pump rate was such that there was always 1–2 ml of liquid in the funnel. Column dimensions Two different strongly basic resins, Amberlite IRA-410 and Amberlyst A26, were packed in columns of various dimensions.Four different lengths (50, 100, 150 and 200 cm) and two different internal diameters (2 and 4 mm) were used in all possible combinations. Both resins have styrene–divinylbenzene skeletal structures; however, Amberlite IRA-410 is a gel-type resin and Amberlyst A26 is a porous or macroreticular resin. The resulting columns were tested for selenium retention and subsequent hydride generation for six replicate measurements at two concentrations of selenium, 10 and 25 mg l21.Parameter optimization The multi-cycle alternating variable search optimization method16,17 was used for the optimization of the following parameters: borohydride concentration, HCl concentration, carrier flow rate and stripping gas flow rate. The figure of merit for the optimization process was maximum net peak height sensitivity (i.e., signal minus blank). Other parameters that were studied include the dimensions of the column, nature of the anion-exchange resin, oxidation state of the selenium and the flow rates of the sample and borohydride solutions.Parameter optimization was carried out with a 1 min preconcentration (at 3 ml min21 sample and borohydride flow rates), using a column size of 150 3 4 mm id packed with Amberlite IRA-410 anion-exchange resin. The optimization of the borohydride and HCl concentrations was carried out using sample solutions of 0, 10 and 25 mg l21 of Se. The effect of such reagents was studied by varying these concentrations within the ranges 0.01–2% m/v NaBH4 in 0.01% m/v NaOH, and 0.1–2 mol l21 HCl.Samples of 10 and 25 mg l21 were used in the studies of the effects of the argon stripping gas flow rate and the carrier flow rate, i.e., the speed at which the acid passed through the column. The argon flow rates were varied between 50 and Fig. 1 Schematic diagram of the manifold for selenium preconcentration and hydride generation. (a) load position and (b) injection position.V1,2, six-port valves; P1,2,3, peristaltic pumps; QAC, quartz atomization cell; GLS, gas–liquid separator; CL, column packing with strong anionexchange resin; L, 500 ml loop; CS, carrier solution, distilled, deionized 18 MW E-pure water (14 ml min21); AS, 4 mol l21 HCl (1.5 ml min21); RS, 0.05% m/v NaBH4 in 0.01% m/v NaOH (3.0 ml min21); S, sample (3.0 ml min21); and W, waste. 916 Analyst, September 1997, Vol. 1221000 ml min21. The carrier flow rate was varied between 4 and 15 ml min21.Analytical Performance Using the optimal experimental conditions (given in Table 1), calibration graphs with 0, 5, 10, 15, 20, and 25 mg l21 of Se for 1, 2 and 3 min of preconcentration were made. The precision of the system was evaluated as the RSD of 10 successive determinations of 5, 10 and 20 mg l21 of Se for 1, 2 and 3 min of preconcentration, respectively. The accuracy was evaluated by means of the recovery of 0.5, 2 and 10 mg l21 of Se spiked in river, lake and tap water.The effect of the sample matrix was evaluated by the method of standard additions. Equal volumes of river, lake and tap water (45 ml) were taken, all but one were separately spiked with different amounts of selenium and then all were diluted to equal volumes (50 ml) to obtain three series of solutions with final concentrations added of 0, 5, 10, 15, 20 and 25 mg l21 of Se. Signals were obtained under the optimal experimental conditions (Table 1) for a 3 min preconcentration time.All calibrations were obtained by an unweighted least-squares procedure. Results and Discussion Determination of Arsenic The results reported previously5,6 for the determination of arsenic were confirmed. Signals were obtained for both AsIII and AsV. For a sample volume of 835 ml, the linear ranges for AsIII and AsV were 0.6–40 and 0.9–50 mg l21, respectively. The precisions, expressed as RSD for five replicate determinations of 20 mg l21, were between 1.8% and 3.2% for AsIII and between 2.4% and 4.1% for AsV.Gas–Liquid Separator The device based on the 25 ml separating funnel was used, rather than the smaller internal volume Perkin-Elmer device, to avoid the carry-over of liquid which tends to occur with certain combinations of argon gas flow rate and borohydride concentration with devices having small internal volumes. As the optimization studies required that these parameters be varied, the use of a more robust gas–liquid separator decreased the down-time spent cleaning and regenerating the surface of the quartz cell.However, for an FI-based procedure the gas–liquid separator will affect the peak height sensitivity by virtue of a contribution to the overall dispersion and it is likely that the device used here is sub-optimal with respect to this characteristic. Column Dimensions and Resin Type The best results were observed with larger column sizes (lengths and id) with both resins, i.e., when the amount of resin was increased.However, lengths above 150 mm for the same internal diameter did not produce any improvement in the signal. The Amberlite IRA-410 gel-type resin produced better results than the Amberlyst A26 macroreticular resin. Therefore, a 150 3 4 mm id column packed with Amberlite IRA-410 geltype resin was chosen for further experiments. Parameter Optimization The optimum conditions are given in Table 1. No signals were obtained from solutions of SeVI.To achieve the best sensitivity, three cycles of the optimization process were necessary. The results of the third cycle for each parameter are discussed below. The effect of the concentration of borohydride is shown in Fig. 2. There was a steady increase in the blank signal as the concentration of borohydride increased from 0.01 to 1.0% m/v, with a slower increase from 1.0 to 2.0% m/v. For both samples (10 and 25 mg l21 of Se) a sharp increase in the signal was observed as the borohydride concentration increased from 0.01 to 0.05% m/v NaBH4, followed by a sharp decrease from 0.5 to 1.0% m/v.No signal was obtained in the absence of borohydride. The 0.05% m/v concentration produced the optimum net signal. This concentration of borohydride is considerably lower than that used in typical flow injection and typical batch procedures. The effect of the concentration of HCl is shown in Fig. 3. As can be seen, the signals from the samples and the blank increased as the HCl concentration increased to 0.5 mol l21, but changed little thereafter.Therefore, 4 mol l21 was chosen for further experiments. The effect of carrier flow rate is shown in Fig. 4. Carrier flow rates between 12 and 15 ml min21 resulted in the best net absorbance signal. A carrier flow rate of 14 ml min21 was used in further experiments. The effect of the stripping gas flow rate is shown in Fig. 5. The most favorable rate was found to be 180 ml min21, hence this value was chosen for further experiments.Table 1 Optimum operation conditions Atomic absorption spectrometer— Wavelength 196 nm Slit width 0.7 nm Lamp current 16 mA Quartz cell temperature 900 °C Background correction On On-line preconcentration and hydride generation— HCl concentration 4 mol l21 NaBH4 concentration 0.05% m/v Sample flow rate 3.0 ml min21 NaBH4 flow rate 3.0 ml min21 Carrier flow rate 14.0 ml min21 Argon flow rate 100 ml min21 Column size (length3id) 15034 mm Resin Amberlite IRA-410 Fig. 2 Effect of concentration of NaBH4 solution on the signal peak height (absorbance). A, no Se; B, 10 mg l21 of Se; and C, 25 mg l21 of Se. The error bars represent the standard deviation for five replicate measurements. Analyst, September 1997, Vol. 122 917Analytical Performance The calibration equations and the other performance figures of merit are summarized in Table 2. The system responded linearly from the detection limit up to 180, 120 and 80 mg l21 of Se for 1, 2 and 3 min preconcentration, respectively.The precision of the procedure as a function of concentration, expressed as RSD, varied from 0.41 to 1.32, 0.24 to 0.81 and 0.18 to 0.61% for 1, 2 and 3 min preconcentration, respectively. The precision of the system improved with preconcentration time, but degraded severely if the residual sample and borohydride were not washed from the column. Carry-over between samples and standards was prevented by flushing residual selenium from the column with the 4% borohydride wash.The limits of detection, defined as the concentration giving a signal equal to three times the standard deviation of the blank signal, were 0.24, 0.15 and 0.12 mg l21 of Se for 1, 2 and 3 min preconcentration, respectively. The sample throughput was 21 h21 for 1 min preconcentration. The results of the standard additions to the water sample matrices were compared with those for the same concentrations in distilled water.The characteristics of the regression lines are summarized in Table 3. The 95% confidence interval about the slope of the calibration in the distilled water matrix contains the slopes of the calibrations in the other matrices, i.e., the confidence intervals of the slopes overlap in all cases and there is no significant difference between the slopes for the different matrices. As no appreciable matrix effect was observed, the analyses of the test samples were performed by calibration with aqueous solutions of a pure selenium salt (Na2SeO3).The recoveries (and standard deviations) of spikes added to the water samples are given in Table 4 and ranged from 96 ± 4 to 102 ± 6, 96 ± 4 to 107 ± 6 and 99 ± 4 to 108 ± 6%, depending on the concentration, for river, lake and tap water, respectively. These values indicate that selenium can be quantitatively determined in such samples. Fig. 3 Effect of concentration of HCl solution on the signal peak height (absorbance).A, no Se; B, 10 mg l21 of Se; and C, 25 mg l21 of Se. The error bars represent the standard deviation for five replicate measurements. Fig. 4 Effect of the carrier solution flow on the net signal peak height (absorbance). A, 10 mg l21 of Se; and B, 25 mg l21 of Se. The error bars represent the standard deviation for five replicate measurements. Table 2 Analytical performance of the system Regression RSD (%)¶ Time*/ equation: LOD (3s)§/ (min) A = b + mC† r‡ mg l21 5 mg l21 10 mg l21 20 mg l21 1 A = 0.011 + 0.0123C 0.9996 0.24 1.32 0.83 0.41 2 A = 0.022 + 0.0197C 0.9997 0.15 0.81 0.46 0.24 3 A = 0.035 + 0.0256C 0.9997 0.12 0.61 0.36 0.18 * Time of passage of sample through the column at 3 ml min21.† A is absorbance, b is intercept, m is slope and C is concentration of Se in mg l21. ‡ Regression coefficient. § LOD (3s) is the detection limit, calculated for 3s/m, where s is the within-run standard deviation of a blank determination (n = 10).¶ RSD for 5, 10 and 20 mg l21 of Se (n = 10). Fig. 5 Effect of stripping gas (argon) flow rate on the net signal peak height (absorbance). A, 10 mg l21 of Se; and B, 25 mg l21 of Se. The error bars represent the standard deviation for five replicate measurements. 918 Analyst, September 1997, Vol. 122Determination of SeIV in Water The results of the application of the proposed method to the determination of selenium in river, lake and tap water are given in Table 4 for a sample volume of 9 ml (3 min preconcentration time).The concentration of selenium in the river water sample was 0.14 ± 0.01 mg l21. The concentrations of selenium in the lake and tap water samples analyzed were below the detection limit of 0.12 mg l21. It is known that the concentration of SeO3 22 in natural water18–20 is usually very low and that selenium is present in water in various oxidation states and chemicals forms, including organic species.18,21,22 Conclusions The co-immobilization of selenium and borohydride on an anion-exchange resin followed by passage of acid forms the basis of a viable method for the determination of selenium by HGAAS and quantitative recovery of selenium from water samples can be achieved.The procedure could be fully automated via a system of computer-controlled valves and pumps and can be used with any atomic spectrometric detection system. Work is in progress to implement the procedure on a Perkin-Elmer FIAS unit.The preconcentration of the analyte on the anion-exchange column allows the potential for improved sensitivity over conventional flow injection hydride generation techniques and the possibility of the separation of the analyte from cation interferences. Current development of the procedure is focused on overcoming such matrix interferences as it may be possible to immobilize the sample and the borohydride successively. Financial support from the University of the Andes, M�erida, Venezuela, for P.C.is gratefully acknowledged and the authors thank Robert I. Ellis for helpful discussions. References 1 Nakahara, T., Spectrochim. Acta Rev., 1991, 14, 95. 2 Godden, R. G., and Thomerson, D. R., Analyst, 1980, 105, 1137. 3 Branch, C. H., and Hutchison, D., Analyst, 1985, 110, 163. 4 Agterdenbos, J., and Bax, D., Fresenius’ Z. Anal. Chem., 1986, 323, 783. 5 Tesfalidet, S., and Irgum, K., Anal. Chem., 1989, 61, 2079. 6 Narasaki, H., Kato, Y., and Kimura, H., Anal. Sci., 1992, 8, 893. 7 Cao, J.Y., and Narasaki, H., Bunseki Kagaku, 1994, 43,169. 8 Tesfalidet, S., and Irgum, K., Fresenius’ J. Anal. Chem., 1991, 341, 532. 9 Brockmann, A., Nonn, C., and Golloch, A., J. Anal. At. Spectrom., 1993, 8, 397. 10 Lin, Y., Wang, X., Yuan, D., Yang, P., Huang, B., and Zhuang, Z., J. Anal. At. Spectrom., 1992, 7, 287. 11 Ding, W.-W., and Sturgeon, R. E., J. Anal. At. Spectrom., 1996, 11, 225. 12 Blais, J. S., Momplaisir, G. M., and Marshall, W. D., Anal. Chem., 1990, 62, 1161. 13 Howard, A. G., J. Anal. At. Spectrom., 1997, 12, 267. 14 Tyson, J. F., Sundin, N. G., Hanna, C. P., and McIntosh, S. A., Spectrochim. Acta Part B, in the press. 15 Hatfield, D. B., Anal. Chem., 1987, 59, 1887. 16 Greenfield, S., Salman, M. S., Thomsen, M., and Tyson, J. F., J. Anal. At. Spectrom., 1989, 4, 55. 17 Miller, J. C., and Miller, J. N., Statistics for Analytical Chemistry, 3rd edn, Ellis Horwood, Chichester, 1993, pp. 185–187. 18 Cutter, G. A., Anal. Chim. Acta, 1978, 98, 59. 19 Measures, C. I., and Burton, J. D., Nature (London), 1978, 273, 293. 20 Yu, Q., Liu, G. Q., and Jin, Q., Talanta, 1983, 30, 265. 21 Cutter, G. A., Anal. Chim. Acta, 1983, 149, 391. 22 Robberecht, H., and Van Grieken, R., Talanta, 1982, 29, 823. Paper 7/01648D Received March 10, 1997 Accepted June 3, 1997 Table 3 Standard additions of selenium to water samples Regression line characteristics Confidence Matrix Slope sslope limits* Intercept sint r† Distilled water 0.0257 0.00048 0.0013 0.032 0.0073 0.9995 River water 0.0249 0.00045 0.0012 0.039 0.0067 0.9998 Lake water 0.0251 0.00045 0.0012 0.035 0.0069 0.9998 Tap water 0.0248 0.00043 0.0012 0.034 0.0061 0.9999 * Confidence limits for the slope, given by ts; the t-value was taken at the 95% confidence level and n 2 2 degrees of freedom. † Correlation coeicient. Table 4 Analytical results for various water samples Selenium Selenium added/ found/ Sample mg l21 mg l21 Recovery (%) River water (Fort river, Amherst) 0 0.14 ± 0.01 — 0.5 0.65 ± 0.03 102 ± 6 2.0 2.06 ± 0.07 96 ± 4 10.0 10.05 ± 0.39 99 ± 4 Lake water (Puffers Pond, Sunderland) 0 < DL* — 0.5 0.48 ± 0.02 96 ± 4 2.0 2.05 ± 0.05 103 ± 3 10.0 10.70 ± 0.56 107 ± 6 Tap water (Amherst) 0 < DL* — 0.5 0.54 ± 0.03 108 ± 6 2.0 2.01 ± 0.05 101 ± 3 10.0 9.89 ± 0.43 99 ± 4 * Detection limit, 0.12 mg l21 of Se. Analyst, September 1997, Vol. 122 919
ISSN:0003-2654
DOI:10.1039/a701648d
出版商:RSC
年代:1997
数据来源: RSC
|
|